From aa65bd8c0e64e4c3b294777b0df94a509d179d6c Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Tue, 3 Nov 2026 22:50:10 -0500
Subject: [PATCH 01/29] CHANGELOG: the route record's failure-cost entry names
 no product

The entry for version 2 of the route record named the product whose proposal it
adopted, against CONTRIBUTING's public-text rule. The name is redacted in place
with a dated marker, as this changelog's history redacts names, and the sentence
keeps its disclosure: the proposal came from a product of WAI's maintainers
[name removed 2026-11-03].

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index 6a64d746d..dea0c8f33 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -672,7 +672,8 @@ and no `depends_on :macos`. It now has both, so the tap carries it as the releas
   unknown: the expected figures are `null` and rank after every measured one, by the existing rule for `null`, never
   as no failures. The record restates every basis and expected figure. The job never pays for failures: the
   unit-price and spend limits read the list price and cost, and a line bills the billed attempt. This is the failure
-  cost waiTransport proposed, with its physics and its basis, except that an unknown failure is not counted as zero.
+  cost a product of WAI's maintainers [name removed 2026-11-03] proposed, with its physics and its basis, except that
+  an unknown failure is not counted as zero.
 - **A hop receipt is bound to its output** (§5.2, §6): in version 2 the attempt that served states `outputs`, the
   content hashes (jwp-receipts §2.1) of what it delivered in order, and its OER/2 receipt's `subject.output` must be
   their commitment, or the record is refused **`receipt_output_mismatch`**: a receipt cannot be moved to vouch for an
-- 
2.54.0 (Apple Git-157)


From 779b203c35ab6663110497edf37f14988b91d342 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Tue, 3 Nov 2026 22:50:15 -0500
Subject: [PATCH 02/29] check-public-names: the products WAI's maintainers sell
 are on the list

The name check read only the private list, which never held the maintainers' own
product names, so a changelog line naming one passed it. Those names are public
already, and the private list also serves checks of text about the products, which
names them where it should; so they are listed in the script, MAINTAINER_PRODUCTS,
in any case, joined or with one hyphen or underscore, the shorter one standing alone.
They are applied beside the private list, with the same opt-out, and an unset private
list still fails closed. The self-test gains a case for each spelling, a generic
phrase that is not a hit, an honoured opt-out, and the unset list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 scripts/check-public-names.sh      | 22 +++++++++++++++++++---
 scripts/test-public-text-checks.sh | 20 ++++++++++++++++++--
 wai/CONTRIBUTING.md                |  7 ++++---
 3 files changed, 41 insertions(+), 8 deletions(-)

diff --git a/scripts/check-public-names.sh b/scripts/check-public-names.sh
index dcb52b8a6..c291788a2 100755
--- a/scripts/check-public-names.sh
+++ b/scripts/check-public-names.sh
@@ -6,8 +6,8 @@
 # customer or commercial product. Standards, standards bodies, protocols,
 # papers and open-source projects may be named, factually.
 #
-# The patterns are NOT in this repository. They are read from the
-# environment variable WAI_NAME_DENYLIST: newline-separated POSIX extended
+# The patterns for other parties' names are NOT in this repository. They are
+# read from the environment variable WAI_NAME_DENYLIST: newline-separated POSIX extended
 # regular expressions, matched case-sensitively (write `[Ff]oo` where both
 # cases matter). In CI the variable comes from a forge Actions secret of the
 # same name; locally, export it from a private file before committing:
@@ -16,6 +16,13 @@
 #
 # A public list would itself publish the names it exists to keep out.
 #
+# The names of the products WAI's maintainers sell are in this script, as
+# MAINTAINER_PRODUCTS below. They are public already, so listing them here
+# publishes nothing; and the private list cannot hold them, because it also
+# serves the checks of text that is about those products. The script requires
+# the private list as before, and applies both lists together, with the same
+# opt-out.
+#
 # Scope: the tracked files under wai/ with the extensions .md .rs .toml .py
 # .sh .mjs .js .json .html .css .ttl .h .c .txt, and wai/'s .gitignore files;
 # or, when files are given as arguments, those files (the self-test,
@@ -50,6 +57,14 @@ if [ -z "${WAI_NAME_DENYLIST:-}" ]; then
   exit 2
 fi
 
+# The products of WAI's maintainers, in any case, joined or with one hyphen or
+# underscore (a crate or a namespace spells them so). The shorter name must
+# stand alone, so that a run of letters inside an encoded string is not a hit.
+MAINTAINER_PRODUCTS=(
+  '[Ww][Aa][Ii][-_]?[Tt][Rr][Aa][Nn][Ss][Pp][Oo][Rr][Tt]'
+  '(^|[^[:alnum:]])[Ww][Aa][Ii][-_]?[Oo][Ss]([^[:alnum:]]|$)'
+)
+
 patterns="$(mktemp)"
 out="$(mktemp)"
 trap 'rm -f "$patterns" "$out"' EXIT
@@ -58,6 +73,7 @@ if [ ! -s "$patterns" ]; then
   echo "WAI_NAME_DENYLIST holds no pattern" >&2
   exit 2
 fi
+printf '%s\n' "${MAINTAINER_PRODUCTS[@]}" >> "$patterns"
 # Compile every pattern once before scanning. grep exits 1 when nothing matches
 # and 2 on an error; only 0 and 1 mean the patterns are usable. grep's message
 # is suppressed: it could quote a private pattern.
@@ -122,7 +138,7 @@ while IFS= read -r line; do
   fi
   # Print the location only: the matched text may be a name the list keeps private.
   if [ $hits -eq 0 ]; then
-    echo "::error::public text names a company, vendor, customer or product (patterns: WAI_NAME_DENYLIST)"
+    echo "::error::public text names a company, vendor, customer or product (patterns: WAI_NAME_DENYLIST and MAINTAINER_PRODUCTS)"
   fi
   echo "${line%%:*}:$(printf '%s' "${line#*:}" | cut -d: -f1)"
   hits=$((hits + 1))
diff --git a/scripts/test-public-text-checks.sh b/scripts/test-public-text-checks.sh
index a2c8b0bea..a148fd10f 100755
--- a/scripts/test-public-text-checks.sh
+++ b/scripts/test-public-text-checks.sh
@@ -9,8 +9,10 @@
 #   - an opt-out with a reason is honoured, and one with an empty reason (the
 #     comment form `<!-- allow-name: -->`, whose `-->` must not count as the
 #     reason) is not;
-#   - a pattern grep cannot compile exits 2, never "clean".
-# The fixture names a made-up company, never a real one.
+#   - a pattern grep cannot compile exits 2, never "clean";
+#   - the products of WAI's maintainers are hits beside any private list.
+# The fixtures name a made-up company, never a real one, and the maintainers'
+# products, which check-public-names.sh lists itself.
 #
 # Exit status: 0 when every case gives its stated status, 1 otherwise.
 
@@ -76,6 +78,20 @@ expect 1 "names: a hit among many relative paths is a hit" in_dir "$tmp" env WAI
 expect 0 "names: a missing file among many fails closed and is named" names_file "gone.md" in_dir "$tmp" env WAI_NAME_DENYLIST="$list" bash "$names" "${many[@]}" sub/gone.md
 expect 0 "names: a directory given as a file fails closed and is named" names_file "/sub" in_dir "$tmp" env WAI_NAME_DENYLIST="$list" bash "$names" "${many[@]}" sub
 
+printf 'The failure cost waiTransport proposed.\n' > "$tmp/product-joined.md"
+printf 'A sidecar under waitransport.provisional.route.\n' > "$tmp/product-lower.md"
+printf 'A crate named wai_transport.\n' > "$tmp/product-underscore.md"
+printf 'It runs on waiOS.\n' > "$tmp/product-short.md"
+printf 'The WAI transport layer, WAI OS integration and xWAIOSy.\n' > "$tmp/product-generic.md"
+printf 'waiTransport <!-- allow-name: a repository URL -->\n' > "$tmp/product-reasoned.md"
+expect 1 "names: a maintainer's product is a hit beside an unrelated private list" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-joined.md"
+expect 1 "names: a maintainer's product in lower case is a hit" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-lower.md"
+expect 1 "names: a maintainer's product joined by an underscore is a hit" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-underscore.md"
+expect 1 "names: the shorter product name standing alone is a hit" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-short.md"
+expect 0 "names: the generic words, and the shorter name inside a word, are not hits" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-generic.md"
+expect 0 "names: an opt-out with a reason is honoured for a product" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-reasoned.md"
+expect 2 "names: with the products listed, an unset private list still fails closed" env -u WAI_NAME_DENYLIST bash "$names" "$tmp/product-joined.md"
+
 printf 'Its reproducibility is the superiority claim.\n' > "$tmp/superiority.md"
 printf 'About 8 kbps, well below Xcodec at the same quality.\n' > "$tmp/well-below.md"
 printf 'Its token rate is among the lowest of any codec.\n' > "$tmp/among.md"
diff --git a/wai/CONTRIBUTING.md b/wai/CONTRIBUTING.md
index f1c2fd26f..12e22619d 100644
--- a/wai/CONTRIBUTING.md
+++ b/wai/CONTRIBUTING.md
@@ -107,9 +107,10 @@ Three checks enforce this. `scripts/check-no-vanity-metrics.sh` (opt-out marker
 `allow-phrase: <reason>`, for an identifier or a quotation) run on every change.
 `scripts/check-public-names.sh` reads its patterns from a private list (the
 `WAI_NAME_DENYLIST` secret in CI, given only to the run on a push to main, so a
-pull request is checked when it lands); a functional identifier — a capability string,
-a model-hub id a tool defaults to, a repository URL, an OS API name, a security
-advisory id — is kept and its line marked `allow-name: <reason>`. Put the marker
+pull request is checked when it lands), and holds the names of the products
+WAI's maintainers sell in the script itself, since they are public already. A
+functional identifier — a capability string, a model-hub id a tool defaults to,
+a repository URL, an OS API name, a security advisory id — is kept and its line marked `allow-name: <reason>`. Put the marker
 in a comment of the file's own syntax (`<!-- ... -->`, `// ...`, `# ...`); the
 reason must contain a letter or a digit. JSON has no comments, so a JSON file
 carries no such identifier: the codec glossary titles a format by its own name,
-- 
2.54.0 (Apple Git-157)


From 98ee3e74b96f0a5e219d37db76f10fed4003e5d0 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Tue, 3 Nov 2026 23:01:44 -0500
Subject: [PATCH 03/29] =?UTF-8?q?sink-energy=20=C2=A74.6:=20outside=20meas?=
 =?UTF-8?q?urements=20of=20decode=20energy,=20each=20cited=20by=20arXiv=20?=
 =?UTF-8?q?identifier=20and=20table?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

A hardware decoder against software decoders of the same format (arXiv 2402.09001,
Table IV), fewer frames chosen by the encoder (arXiv 2603.18305, Table V) and a lower
resolution or coarser quantiser chosen by the encoder (arXiv 2511.00707, Table 4), each
with what was measured and against what. They are informative: no record states them,
no rule reads them, and none is a figure of a WAI decoder. The section says how they bear
on the text: the two encoder choices cost frames or quality, SPEC §4's energy order
already ranks the platform's power-efficient decode first, and a hardware row of this
revision reads no media engine, so it and the first point are not like for like.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/extensions/sink-energy.md | 61 +++++++++++++++++++++++++++++++----
 1 file changed, 55 insertions(+), 6 deletions(-)

diff --git a/wai/extensions/sink-energy.md b/wai/extensions/sink-energy.md
index c3e2e31f6..2ad8ffa28 100644
--- a/wai/extensions/sink-energy.md
+++ b/wai/extensions/sink-energy.md
@@ -24,16 +24,17 @@
 > Texts read for this revision: RFC 8032 (Ed25519); RFC 8785 (JSON Canonicalization
 > Scheme); RFC 8259; FIPS 180-4 (SHA-256); JCGM 100:2008 (GUM) §4.3.7 and F.2.2.1;
 > Commission Delegated Regulation (EU) 2019/2013, for the on-mode power a display's
-> energy label states. A section number in this text is this extension's unless it names
-> another document.
+> energy label states; arXiv 2402.09001, 2603.18305 and 2511.00707, for §4.6's
+> outside measurements of decode energy. A section number in this text is this
+> extension's unless it names another document.
 
 ## 1. Scope
 
 In scope: the three records, their signatures and identities (§3); the decode table, the
-procedure that produces it, and how each row's status and figure follow from its rounds
-(§4); the segment receipt and its four methods (§5); the session receipt and its per-hop
-split (§6); verification (§7); the reasons (§8); and what a conforming implementation
-publishes (§9).
+procedure that produces it, how each row's status and figure follow from its rounds, and
+outside measurements of decode energy (§4); the segment receipt and its four methods
+(§5); the session receipt and its per-hop split (§6); verification (§7); the reasons
+(§8); and what a conforming implementation publishes (§9).
 
 Out of scope: a transport; how a counter works, which energy-measurement §9 states per
 platform; the energy of networks, which no sink can read and this extension never
@@ -272,6 +273,54 @@ every segment that is not unmetered, is `Partial` in this revision, and a `Whole
 refused (`table_row_figure`, `segment_method_class`); `Whole` is kept for a profile a
 later revision registers.
 
+### 4.6 Outside measurements of decode energy (informative)
+
+The three points below are outside measurements of decode energy. Their figures are not
+figures of a WAI decoder or of a reference sink: no record of this extension states them,
+and no rule of it reads them. Each is cited by its arXiv identifier and the table it is
+read from, with what was measured and against what.
+
+1. **A platform's hardware decoder, against software decoders of the same format.**
+   arXiv 2402.09001, Table IV. Over AVC, VP9, HEVC and AV1, each in a random-access and a
+   low-delay configuration, a hardware decoder spent from 0.89% to 8.86% of the energy of
+   a software decoder of the same format, and from 4.44% to 8.86% of the energy of the
+   optimised software decoders with vector instructions enabled. The figures are dynamic
+   energy: the energy measured at idle is subtracted. The software decoders ran
+   single-threaded on a desktop processor, read by the processor's own energy counter;
+   the hardware decoders ran on a single-board computer's video unit, read at its power
+   input by an external meter. The point sets two decode paths side by side as two
+   machines built them.
+2. **Fewer frames, chosen by the encoder.** arXiv 2603.18305, Table V, its row for all
+   sequences. Decoding energy was 17.60% lower, encoding energy 17.46% lower and the
+   bitrate 3.38% lower, each a Bjøntegaard delta at equal matched PSNR (mPSNR, computed
+   against the source), against encoding each sequence at its source's 120 frames per
+   second. A trained classifier picks, per sequence and per quantisation setting, the
+   frame rate the encoder codes at, the source's or a lower one; where it is lower, the
+   decoder decodes fewer frames than the source has. One software encoder and one
+   software decoder of HEVC were measured, by the processor's own energy counter, idle
+   subtracted.
+3. **Lower resolution or coarser quantisation, chosen by the encoder.** arXiv 2511.00707,
+   Table 4, its row ρ = 0.05. Decoding energy was 53.54% lower and encoding energy 51.22%
+   lower, for an average loss of 1.68 VMAF points and 3.67 dB PSNR, against the
+   configuration chosen with no loss allowed (ρ = 0). Per sequence, the encoder takes, of
+   the resolutions and quantisers whose predicted VMAF is within 5% of the best predicted,
+   the one of least predicted energy. One software encoder and one software decoder of VVC
+   were measured on UHD sources at 60 frames per second, by the processor's own energy
+   counter read through a software tool.
+
+How they bear on this extension:
+
+- Points 2 and 3 are choices of what to encode, and each costs frames or picture quality.
+  Point 1 is a choice of decoder for the same stream. None of the three changes what a
+  decoder computes from a given stream.
+- SPEC §4's *Energy order* already ranks first a classical decode that the sink's platform
+  answers supported, smooth and power-efficient. Point 1 measures what a hardware decode
+  spent beside a software one; the platform's answer stays a hint that orders, never an
+  energy figure or evidence of hardware (SPEC §4).
+- A `hardware` row of this revision reads no media engine (§4.5, *Coverage*): its figure
+  is the processors' part of a fixed-function decode. It and point 1 measure different
+  things, and are not set against each other as like for like.
+
 ## 5. The segment receipt
 
 A segment receipt is a sink's statement of what one segment of one stream cost it to
-- 
2.54.0 (Apple Git-157)


From b8bbb4036117b6906ccccca8367dcde43655c895 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:46:14 -0500
Subject: [PATCH 04/29] check-public-names: the maintainers' product names are
 checked on every pull request

The private list is a secret, given only to the run on a push to main, so a pull
request that named one of the maintainers' products was caught only after it landed.
Those names need no secret: --maintainer-products-only applies MAINTAINER_PRODUCTS
alone and reads no private list, and the repository checks run it on every change.
The default still requires the private list and fails closed without it.

The script's header and CONTRIBUTING now say which products the script lists: those
the maintainers' public site names. A product not yet announced is named on no site,
so the private list holds it until the day it is announced, when it moves to the
script. The self-test gains cases for the new mode: a product is a hit, a clean file
is clean, the private list is not read, and a missing file fails closed. Two
paragraphs are rewrapped to their file's width.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/repo-checks.yml  |  5 +++
 scripts/check-public-names.sh      | 61 +++++++++++++++++++-----------
 scripts/test-public-text-checks.sh |  7 +++-
 wai/CHANGELOG.md                   |  6 +++
 wai/CONTRIBUTING.md                | 26 +++++++------
 5 files changed, 70 insertions(+), 35 deletions(-)

diff --git a/.github/workflows/repo-checks.yml b/.github/workflows/repo-checks.yml
index c1a71cd09..3fd6a9476 100644
--- a/.github/workflows/repo-checks.yml
+++ b/.github/workflows/repo-checks.yml
@@ -108,6 +108,11 @@ jobs:
       - name: "public text: WAI's public text names no comparison"
         if: ${{ !cancelled() }}
         run: bash scripts/check-comparative-prose.sh
+      # The maintainers' product names are public, held in the script itself: this run
+      # reads no secret, so it checks every change, a pull request included.
+      - name: "public text: WAI's public text names no product of its maintainers"
+        if: ${{ !cancelled() }}
+        run: bash scripts/check-public-names.sh --maintainer-products-only
       # The pattern list is a secret, never given to pull-request code: the check runs
       # on the push to main.
       - name: "public text: WAI's public text names no company or product (push to main)"
diff --git a/scripts/check-public-names.sh b/scripts/check-public-names.sh
index c291788a2..7415e6936 100755
--- a/scripts/check-public-names.sh
+++ b/scripts/check-public-names.sh
@@ -7,21 +7,28 @@
 # papers and open-source projects may be named, factually.
 #
 # The patterns for other parties' names are NOT in this repository. They are
-# read from the environment variable WAI_NAME_DENYLIST: newline-separated POSIX extended
-# regular expressions, matched case-sensitively (write `[Ff]oo` where both
-# cases matter). In CI the variable comes from a forge Actions secret of the
-# same name; locally, export it from a private file before committing:
+# read from the environment variable WAI_NAME_DENYLIST: newline-separated
+# POSIX extended regular expressions, matched case-sensitively (write `[Ff]oo`
+# where both cases matter). In CI the variable comes from a forge Actions
+# secret of the same name; locally, export it from a private file before
+# committing:
 #
 #   WAI_NAME_DENYLIST="$(cat /path/to/private/patterns)" scripts/check-public-names.sh
 #
 # A public list would itself publish the names it exists to keep out.
 #
-# The names of the products WAI's maintainers sell are in this script, as
-# MAINTAINER_PRODUCTS below. They are public already, so listing them here
-# publishes nothing; and the private list cannot hold them, because it also
-# serves the checks of text that is about those products. The script requires
-# the private list as before, and applies both lists together, with the same
-# opt-out.
+# The products of WAI's maintainers that their public site names
+# (waiTransport, waiOS) are in this script, as MAINTAINER_PRODUCTS below. They
+# are public already, so listing them here publishes nothing. The private list
+# cannot hold them: it also checks the text of the product sites, which name
+# them. A product the maintainers have not announced is named on no site, so
+# the private list holds it until the day it is announced, when it moves here.
+# The script requires the private list as before, and applies both lists
+# together, with the same opt-out.
+#
+# With --maintainer-products-only as its first argument, the script applies
+# MAINTAINER_PRODUCTS alone and reads no private list, so it needs no secret;
+# CI runs it so on every change, a pull request included.
 #
 # Scope: the tracked files under wai/ with the extensions .md .rs .toml .py
 # .sh .mjs .js .json .html .css .ttl .h .c .txt, and wai/'s .gitignore files;
@@ -39,10 +46,10 @@
 # a digit. Prose around such an identifier describes the standard or the
 # function, not the owner.
 #
-# Exit status: 0 clean; 1 a hit; 2 WAI_NAME_DENYLIST unset or empty, a pattern
-# grep cannot compile, a file given that is not a readable regular file (each
-# such file is named), or any other grep error. The check never reads an error
-# as "clean".
+# Exit status: 0 clean; 1 a hit; 2 WAI_NAME_DENYLIST unset or empty (unless
+# --maintainer-products-only is given), a pattern grep cannot compile, a file
+# given that is not a readable regular file (each such file is named), or any
+# other grep error. The check never reads an error as "clean".
 
 set -euo pipefail
 
@@ -52,7 +59,15 @@ REPO_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
 CALLER_DIR="$(pwd)"
 cd "$REPO_ROOT"
 
-if [ -z "${WAI_NAME_DENYLIST:-}" ]; then
+products_only=0
+lists="WAI_NAME_DENYLIST and MAINTAINER_PRODUCTS"
+if [ "${1:-}" = "--maintainer-products-only" ]; then
+  products_only=1
+  lists="MAINTAINER_PRODUCTS"
+  shift
+fi
+
+if [ $products_only -eq 0 ] && [ -z "${WAI_NAME_DENYLIST:-}" ]; then
   echo "WAI_NAME_DENYLIST is not set: nothing to check against (see the header of $0)" >&2
   exit 2
 fi
@@ -68,10 +83,12 @@ MAINTAINER_PRODUCTS=(
 patterns="$(mktemp)"
 out="$(mktemp)"
 trap 'rm -f "$patterns" "$out"' EXIT
-printf '%s\n' "$WAI_NAME_DENYLIST" | grep -v '^[[:space:]]*$' > "$patterns" || true
-if [ ! -s "$patterns" ]; then
-  echo "WAI_NAME_DENYLIST holds no pattern" >&2
-  exit 2
+if [ $products_only -eq 0 ]; then
+  printf '%s\n' "$WAI_NAME_DENYLIST" | grep -v '^[[:space:]]*$' > "$patterns" || true
+  if [ ! -s "$patterns" ]; then
+    echo "WAI_NAME_DENYLIST holds no pattern" >&2
+    exit 2
+  fi
 fi
 printf '%s\n' "${MAINTAINER_PRODUCTS[@]}" >> "$patterns"
 # Compile every pattern once before scanning. grep exits 1 when nothing matches
@@ -82,7 +99,7 @@ grep -E -f "$patterns" /dev/null >/dev/null 2>&1
 rc=$?
 set -e
 if [ $rc -gt 1 ]; then
-  echo "::error::WAI_NAME_DENYLIST holds a pattern grep cannot compile (grep exit $rc); nothing was checked" >&2
+  echo "::error::a pattern grep cannot compile is among $lists (grep exit $rc); nothing was checked" >&2
   exit 2
 fi
 
@@ -138,7 +155,7 @@ while IFS= read -r line; do
   fi
   # Print the location only: the matched text may be a name the list keeps private.
   if [ $hits -eq 0 ]; then
-    echo "::error::public text names a company, vendor, customer or product (patterns: WAI_NAME_DENYLIST and MAINTAINER_PRODUCTS)"
+    echo "::error::public text names a company, vendor, customer or product (patterns: $lists)"
   fi
   echo "${line%%:*}:$(printf '%s' "${line#*:}" | cut -d: -f1)"
   hits=$((hits + 1))
@@ -149,4 +166,4 @@ if [ $hits -gt 0 ]; then
   echo "for a functional identifier, mark the line with allow-name: <reason>."
   exit 1
 fi
-echo "public-name check: clean (${#files[@]} files)"
+echo "public-name check ($lists): clean (${#files[@]} files)"
diff --git a/scripts/test-public-text-checks.sh b/scripts/test-public-text-checks.sh
index a148fd10f..0d856bcf2 100755
--- a/scripts/test-public-text-checks.sh
+++ b/scripts/test-public-text-checks.sh
@@ -10,7 +10,8 @@
 #     comment form `<!-- allow-name: -->`, whose `-->` must not count as the
 #     reason) is not;
 #   - a pattern grep cannot compile exits 2, never "clean";
-#   - the products of WAI's maintainers are hits beside any private list.
+#   - the products of WAI's maintainers are hits beside any private list, and
+#     alone with --maintainer-products-only, which reads no private list.
 # The fixtures name a made-up company, never a real one, and the maintainers'
 # products, which check-public-names.sh lists itself.
 #
@@ -91,6 +92,10 @@ expect 1 "names: the shorter product name standing alone is a hit" env WAI_NAME_
 expect 0 "names: the generic words, and the shorter name inside a word, are not hits" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-generic.md"
 expect 0 "names: an opt-out with a reason is honoured for a product" env WAI_NAME_DENYLIST="$list" bash "$names" "$tmp/product-reasoned.md"
 expect 2 "names: with the products listed, an unset private list still fails closed" env -u WAI_NAME_DENYLIST bash "$names" "$tmp/product-joined.md"
+expect 1 "names: products only, a product is a hit with no private list" env -u WAI_NAME_DENYLIST bash "$names" --maintainer-products-only "$tmp/product-joined.md"
+expect 0 "names: products only, a clean file is clean with no private list" env -u WAI_NAME_DENYLIST bash "$names" --maintainer-products-only "$tmp/clean.md"
+expect 0 "names: products only, the private list is not read" env WAI_NAME_DENYLIST="$list" bash "$names" --maintainer-products-only "$tmp/bare.md"
+expect 2 "names: products only, a missing file fails closed" env -u WAI_NAME_DENYLIST bash "$names" --maintainer-products-only "$tmp/no-such-file.md"
 
 printf 'Its reproducibility is the superiority claim.\n' > "$tmp/superiority.md"
 printf 'About 8 kbps, well below Xcodec at the same quality.\n' > "$tmp/well-below.md"
diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index dea0c8f33..a57e82118 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -6,6 +6,12 @@ wire format may change and such changes are recorded here.
 
 ## Unreleased
 
+### CI: the maintainers' product names are checked on every pull request
+
+- **CI:** the public-name check applies the maintainers' product names, held in the script, on every change, a pull
+  request included (`--maintainer-products-only`); the private list still runs on the push to main. CONTRIBUTING
+  states both.
+
 ### Contribution links: a packet taken after its sequence number's drop point is late, whatever the timer did
 
 - **[live-contribution](extensions/live-contribution.md) §6.1:** a packet *arrives* when the receiver takes it. An
diff --git a/wai/CONTRIBUTING.md b/wai/CONTRIBUTING.md
index 12e22619d..6e84b2b25 100644
--- a/wai/CONTRIBUTING.md
+++ b/wai/CONTRIBUTING.md
@@ -105,18 +105,20 @@ the CHANGELOG state what WAI does, plainly:
 Three checks enforce this. `scripts/check-no-vanity-metrics.sh` (opt-out marker
 `allow-vanity: <reason>`) and `scripts/check-comparative-prose.sh` (opt-out
 `allow-phrase: <reason>`, for an identifier or a quotation) run on every change.
-`scripts/check-public-names.sh` reads its patterns from a private list (the
-`WAI_NAME_DENYLIST` secret in CI, given only to the run on a push to main, so a
-pull request is checked when it lands), and holds the names of the products
-WAI's maintainers sell in the script itself, since they are public already. A
-functional identifier — a capability string, a model-hub id a tool defaults to,
-a repository URL, an OS API name, a security advisory id — is kept and its line marked `allow-name: <reason>`. Put the marker
-in a comment of the file's own syntax (`<!-- ... -->`, `// ...`, `# ...`); the
-reason must contain a letter or a digit. JSON has no comments, so a JSON file
-carries no such identifier: the codec glossary titles a format by its own name,
-FourCC, file extension or designation, and never by a company, a product or a
-person ([`glossary/SCHEMA.md`](glossary/SCHEMA.md)). The checks fail closed: a
-pattern that does not compile, or any grep error, fails them, and
+`scripts/check-public-names.sh` holds the names of the maintainers' products
+that their public site names in the script itself, since they are public
+already, and checks them on every change (`--maintainer-products-only`). Its
+other patterns are a private list (the `WAI_NAME_DENYLIST` secret in CI, given
+only to the run on a push to main, so a pull request is checked against that
+list when it lands). A functional identifier — a capability string, a model-hub
+id a tool defaults to, a repository URL, an OS API name, a security advisory
+id — is kept and its line marked `allow-name: <reason>`. Put the marker in a
+comment of the file's own syntax (`<!-- ... -->`, `// ...`, `# ...`); the reason
+must contain a letter or a digit. JSON has no comments, so a JSON file carries
+no such identifier: the codec glossary titles a format by its own name, FourCC,
+file extension or designation, and never by a company, a product or a person
+([`glossary/SCHEMA.md`](glossary/SCHEMA.md)). The checks fail closed: a pattern
+that does not compile, or any grep error, fails them, and
 `scripts/test-public-text-checks.sh` proves that on every run.
 
 Dated research records are not rewritten: a correction is a dated annotation
-- 
2.54.0 (Apple Git-157)


From 98f9792c3476bcae219df55439746df51949f154 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:46:14 -0500
Subject: [PATCH 05/29] =?UTF-8?q?sink-energy=20=C2=A74.6:=20what=20the=20t?=
 =?UTF-8?q?hird=20point's=20tool=20read,=20and=20what=20its=20paper=20leav?=
 =?UTF-8?q?es=20unstated?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Point 3 said its energy was read by the processor's own counter through a software
tool. The paper's tool reads the processor's counter and the graphics processor's,
and the paper states neither whether the idle energy is subtracted nor whether Table
4's savings are the measured or the predicted energy of the configurations chosen;
the point now says so. Point 2's classifier chooses per rate-factor setting, of which
Table V covers four, not per quantisation setting.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/extensions/sink-energy.md | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)

diff --git a/wai/extensions/sink-energy.md b/wai/extensions/sink-energy.md
index 2ad8ffa28..e8e488029 100644
--- a/wai/extensions/sink-energy.md
+++ b/wai/extensions/sink-energy.md
@@ -294,19 +294,21 @@ read from, with what was measured and against what.
    sequences. Decoding energy was 17.60% lower, encoding energy 17.46% lower and the
    bitrate 3.38% lower, each a Bjøntegaard delta at equal matched PSNR (mPSNR, computed
    against the source), against encoding each sequence at its source's 120 frames per
-   second. A trained classifier picks, per sequence and per quantisation setting, the
-   frame rate the encoder codes at, the source's or a lower one; where it is lower, the
-   decoder decodes fewer frames than the source has. One software encoder and one
-   software decoder of HEVC were measured, by the processor's own energy counter, idle
-   subtracted.
+   second. A trained classifier picks, per sequence and per rate-factor setting (the
+   table covers four), the frame rate the encoder codes at, the source's or a lower one;
+   where it is lower, the decoder decodes fewer frames than the source has. One software
+   encoder and one software decoder of HEVC were measured, by the processor's own energy
+   counter, idle subtracted.
 3. **Lower resolution or coarser quantisation, chosen by the encoder.** arXiv 2511.00707,
    Table 4, its row ρ = 0.05. Decoding energy was 53.54% lower and encoding energy 51.22%
    lower, for an average loss of 1.68 VMAF points and 3.67 dB PSNR, against the
    configuration chosen with no loss allowed (ρ = 0). Per sequence, the encoder takes, of
    the resolutions and quantisers whose predicted VMAF is within 5% of the best predicted,
    the one of least predicted energy. One software encoder and one software decoder of VVC
-   were measured on UHD sources at 60 frames per second, by the processor's own energy
-   counter read through a software tool.
+   were measured on UHD sources at 60 frames per second, by a software tool reading the
+   processor's energy counter and the graphics processor's; the paper does not state
+   whether the idle energy is subtracted, nor whether Table 4's savings are the measured
+   or the predicted energy of the configurations chosen.
 
 How they bear on this extension:
 
-- 
2.54.0 (Apple Git-157)


From 6eb0747681a03f8d437d045aa4f304f676898111 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Tue, 3 Nov 2026 22:44:27 -0500
Subject: [PATCH 06/29] CI: one pinned wasmtime, run with canonical NaNs and
 deterministic relaxed SIMD
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

WAI's WebAssembly legs ran two stock wasmtime versions: 49.0.2 in the
quantum and face install steps, and 38.0.3 in wai-ops.yml and wai-vector.yml
(wai-ops runs the wasm32-wasip2 decode determinism-tiers §5 cites). A stock
wasmtime leaves NaN canonicalisation off and lets each relaxed SIMD
instruction take the host's own behaviour, so the bits of a NaN a module reads
back, and the result of a relaxed instruction, were the engine's choice.

Every install step now pins wasmtime 50.0.1 by version and SHA-256 (the
x86_64 Linux release archive, 685466455dd4...98a5, the digest the release
lists), and every run passes

    -W nan-canonicalization=y -W relaxed-simd-deterministic=y

wasmtime's two determinism settings: every NaN a float instruction makes is
the canonical one, and each relaxed SIMD instruction has one behaviour on
every host. The flag names are 50.0.1's own, read from that release's
`wasmtime run -W help` on the build host. The two 38.0.3 steps gain the
architecture check and the digest the others already had.

Both settings took effect on the build host (arm64, the 50.0.1 release for
that platform). f32.add of a NaN with payload 0x200001 and 1.0 gives
0x7fe00001 without the flags and the canonical 0x7fc00000 with them. The
fixed relaxed behaviour is the engine's, not always the one WebAssembly's
deterministic profile prescribes: f32x4.relaxed_madd(1+2^-12, 1+2^-12,
-(1+2^-11)) gives the fused 2^-24 with and without the flag, where the
profile's choice (parameter 0) is the unfused fadd(fmul), which gives 0.
The workflows' comment says so.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-conformance.yml      | 32 +++++++++++++++-------
 .github/workflows/wai-ops.yml              | 29 +++++++++++++++-----
 .github/workflows/wai-quantum-cli.yml      | 22 +++++++++------
 .github/workflows/wai-quantum-formats.yml  | 16 +++++++----
 .github/workflows/wai-quantum-lang.yml     | 14 +++++++---
 .github/workflows/wai-quantum-pulse.yml    | 14 +++++++---
 .github/workflows/wai-quantum-registry.yml | 14 +++++++---
 .github/workflows/wai-quantum-verdict.yml  | 16 +++++++----
 .github/workflows/wai-vector.yml           | 29 +++++++++++++++-----
 9 files changed, 132 insertions(+), 54 deletions(-)

diff --git a/.github/workflows/wai-conformance.yml b/.github/workflows/wai-conformance.yml
index 7ec3144d5..cd3b57a60 100644
--- a/.github/workflows/wai-conformance.yml
+++ b/.github/workflows/wai-conformance.yml
@@ -969,7 +969,13 @@ jobs:
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
 
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -979,18 +985,18 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
 
       - name: tests/bit_identity.rs under wasmtime
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: cargo test --locked --target wasm32-wasip2 --features full --test bit_identity
 
       # The filters are module paths, so a module that is renamed or loses its tests would
@@ -1000,7 +1006,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: |
           set -euo pipefail
           mods="quantum ijson receipt_pins quantum_fabric quantum_source quantum_channel quantum_qudit quantum_receipt engine::synth engine::compile engine::estimate"
@@ -1083,7 +1089,7 @@ jobs:
           set -euo pipefail
           cargo build --locked --target wasm32-wasip2 --no-default-features --features quantum --bin wai_quantum_vectors \
             --config 'profile.dev.package.wai.codegen-units=1'
-          wasmtime run --dir ../quantum-conformance::/corpus \
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir ../quantum-conformance::/corpus \
             target/wasm32-wasip2/debug/wai_quantum_vectors.wasm verify /corpus
 
   # The face generator under wasm32-wasip2, run by wasmtime (extensions/avatars.md §7.8):
@@ -1113,7 +1119,13 @@ jobs:
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
 
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -1123,11 +1135,11 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
 
@@ -1141,7 +1153,7 @@ jobs:
           cargo build --locked --release --target wasm32-wasip2 --no-default-features --features avatar_face --bin wai_avatar_face_vectors \
             --config 'profile.release.package.wai.codegen-units=1'
           t="${CARGO_TARGET_DIR:-target}"
-          wasmtime run --dir ../live-conformance/avatar-face::/corpus \
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir ../live-conformance/avatar-face::/corpus \
             "$t/wasm32-wasip2/release/wai_avatar_face_vectors.wasm" verify /corpus | tee "$RUNNER_TEMP/face-wasm.log"
           [ "$(grep -c '^PASS ' "$RUNNER_TEMP/face-wasm.log")" = 3 ] || { echo "::error::not every check passed under wasm32"; exit 1; }
 
diff --git a/.github/workflows/wai-ops.yml b/.github/workflows/wai-ops.yml
index dad94e41b..996ad1b1a 100644
--- a/.github/workflows/wai-ops.yml
+++ b/.github/workflows/wai-ops.yml
@@ -207,14 +207,29 @@ jobs:
           rustup target add wasm32-wasip2
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
 
-      - name: Install wasmtime 38.0.3
+      # The release binary, pinned by version and SHA-256, rather than a build from source.
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
-          v=38.0.3
-          curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/wasmtime.tar.xz" \
-            "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/wasmtime-v$v-x86_64-linux.tar.xz"
-          tar -C "$RUNNER_TEMP" -xf "$RUNNER_TEMP/wasmtime.tar.xz"
+          [ "$(uname -m)" = x86_64 ] || { echo "::error::the pinned wasmtime archive is x86_64; this host is $(uname -m)"; exit 1; }
+          if ! command -v xz >/dev/null 2>&1; then
+            SUDO=""; [ "$(id -u)" -ne 0 ] && SUDO="sudo"
+            $SUDO apt-get update -qq
+            $SUDO apt-get install -y -qq xz-utils
+          fi
+          v=50.0.1
+          f="wasmtime-v$v-x86_64-linux.tar.xz"
+          curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
+            "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
+          tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
 
       # The tests read committed files through CARGO_MANIFEST_DIR/.., so the wai
@@ -225,7 +240,7 @@ jobs:
         run: |
           set -euo pipefail
           W=$(cd .. && pwd)
-          CARGO_TARGET_WASM32_WASIP2_RUNNER="wasmtime run --dir $W::$W" \
+          CARGO_TARGET_WASM32_WASIP2_RUNNER="wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir $W::$W" \
             cargo test --locked --release --target wasm32-wasip2
 
       - name: Build wai-verify for wasm32-wasip2
@@ -239,7 +254,7 @@ jobs:
         run: |
           set -euo pipefail
           wasmtime --version
-          wasmtime run --dir .::/wai wai-verify/target/wasm32-wasip2/release/wai-verify.wasm vectors verify /wai
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir .::/wai wai-verify/target/wasm32-wasip2/release/wai-verify.wasm vectors verify /wai
 
   node:
     needs: relevant
diff --git a/.github/workflows/wai-quantum-cli.yml b/.github/workflows/wai-quantum-cli.yml
index d990712c0..93916b1a1 100644
--- a/.github/workflows/wai-quantum-cli.yml
+++ b/.github/workflows/wai-quantum-cli.yml
@@ -111,7 +111,13 @@ jobs:
           cargo build --locked --release
           bash smoke.sh ./target/release/wai-quantum
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -121,11 +127,11 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
       # The same source as a WASI component: a circuit's state hash, a run's
@@ -142,23 +148,23 @@ jobs:
           printf 'OPENQASM 3.0; qubit[3] q; h q[0]; cx q[0], q[1]; t q[2]; cx q[1], q[2];' > "$RUNNER_TEMP/p.qasm"
           for args in "lab run ghz -n 5" "lab receipt ghz -n 3 --demo-key"; do
             $n $args > "$RUNNER_TEMP/native.txt" 2> /dev/null
-            wasmtime run "$w" $args > "$RUNNER_TEMP/wasi.txt" 2> /dev/null
+            wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y "$w" $args > "$RUNNER_TEMP/wasi.txt" 2> /dev/null
             cmp "$RUNNER_TEMP/native.txt" "$RUNNER_TEMP/wasi.txt" || { echo "::error::$args differs under wasmtime"; diff "$RUNNER_TEMP/native.txt" "$RUNNER_TEMP/wasi.txt" || true; exit 1; }
           done
           $n run --demo-key --json --shots 100 < "$RUNNER_TEMP/p.qasm" 2> /dev/null | python3 -c 'import json,sys; j=json.load(sys.stdin); j["result"].pop("elapsed_ms"); j["result"]["receipt"].pop("digest"); j["result"]["receipt"].pop("signature"); print(json.dumps(j, sort_keys=True))' > "$RUNNER_TEMP/native.json"
-          wasmtime run "$w" run --demo-key --json --shots 100 < "$RUNNER_TEMP/p.qasm" 2> /dev/null | python3 -c 'import json,sys; j=json.load(sys.stdin); j["result"].pop("elapsed_ms"); j["result"]["receipt"].pop("digest"); j["result"]["receipt"].pop("signature"); print(json.dumps(j, sort_keys=True))' > "$RUNNER_TEMP/wasi.json"
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y "$w" run --demo-key --json --shots 100 < "$RUNNER_TEMP/p.qasm" 2> /dev/null | python3 -c 'import json,sys; j=json.load(sys.stdin); j["result"].pop("elapsed_ms"); j["result"]["receipt"].pop("digest"); j["result"]["receipt"].pop("signature"); print(json.dumps(j, sort_keys=True))' > "$RUNNER_TEMP/wasi.json"
           cmp "$RUNNER_TEMP/native.json" "$RUNNER_TEMP/wasi.json" || { echo "::error::a run's result differs under wasmtime"; exit 1; }
           # A compile to Clifford+T (rotation synthesis in exact and fixed-point
           # arithmetic, the passes, the certificate and its seal): byte for byte.
           printf 'OPENQASM 3.0; qubit[2] q; h q[0]; rz(0.3) q[0]; cx q[0], q[1]; ry(1.1) q[1]; cp(0.7) q[0], q[1];' > "$RUNNER_TEMP/angles.qasm"
           $n compile --demo-key --json --level 2 --epsilon 1e-8 < "$RUNNER_TEMP/angles.qasm" > "$RUNNER_TEMP/native-compile.json" 2> /dev/null
-          wasmtime run "$w" compile --demo-key --json --level 2 --epsilon 1e-8 < "$RUNNER_TEMP/angles.qasm" > "$RUNNER_TEMP/wasi-compile.json" 2> /dev/null
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y "$w" compile --demo-key --json --level 2 --epsilon 1e-8 < "$RUNNER_TEMP/angles.qasm" > "$RUNNER_TEMP/wasi-compile.json" 2> /dev/null
           cmp "$RUNNER_TEMP/native-compile.json" "$RUNNER_TEMP/wasi-compile.json" || { echo "::error::a compile differs under wasmtime"; exit 1; }
           # A run on a device target: the layout search, the noise derived and
           # the density matrix, the same bits.
           for t in synthetic-line-5 synthetic-all-5; do
             $n run --demo-key --json --shots 100 --target $t < "$RUNNER_TEMP/p.qasm" 2> /dev/null | python3 -c 'import json,sys; j=json.load(sys.stdin); j["result"]["receipt"].pop("digest"); j["result"]["receipt"].pop("signature"); print(json.dumps(j, sort_keys=True))' > "$RUNNER_TEMP/native-$t.json"
-            wasmtime run "$w" run --demo-key --json --shots 100 --target $t < "$RUNNER_TEMP/p.qasm" 2> /dev/null | python3 -c 'import json,sys; j=json.load(sys.stdin); j["result"]["receipt"].pop("digest"); j["result"]["receipt"].pop("signature"); print(json.dumps(j, sort_keys=True))' > "$RUNNER_TEMP/wasi-$t.json"
+            wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y "$w" run --demo-key --json --shots 100 --target $t < "$RUNNER_TEMP/p.qasm" 2> /dev/null | python3 -c 'import json,sys; j=json.load(sys.stdin); j["result"]["receipt"].pop("digest"); j["result"]["receipt"].pop("signature"); print(json.dumps(j, sort_keys=True))' > "$RUNNER_TEMP/wasi-$t.json"
             cmp "$RUNNER_TEMP/native-$t.json" "$RUNNER_TEMP/wasi-$t.json" || { echo "::error::a run on $t differs under wasmtime"; exit 1; }
           done
           # A resource estimate (the counts, each rotation's synthesis, the
@@ -166,7 +172,7 @@ jobs:
           # the seal): byte for byte.
           printf 'OPENQASM 3.0; qubit[4] q; h q; ctrl(3) @ x q[0], q[1], q[2], q[3]; rz(0.3) q[0]; ry(1.1) q[2]; t q[1];' > "$RUNNER_TEMP/est.qasm"
           $n estimate --demo-key --json --qubit us-e3 < "$RUNNER_TEMP/est.qasm" > "$RUNNER_TEMP/native-estimate.json" 2> /dev/null
-          wasmtime run "$w" estimate --demo-key --json --qubit us-e3 < "$RUNNER_TEMP/est.qasm" > "$RUNNER_TEMP/wasi-estimate.json" 2> /dev/null
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y "$w" estimate --demo-key --json --qubit us-e3 < "$RUNNER_TEMP/est.qasm" > "$RUNNER_TEMP/wasi-estimate.json" 2> /dev/null
           cmp "$RUNNER_TEMP/native-estimate.json" "$RUNNER_TEMP/wasi-estimate.json" || { echo "::error::an estimate differs under wasmtime"; exit 1; }
           echo "the WASI component's output equals the native build's"
 
diff --git a/.github/workflows/wai-quantum-formats.yml b/.github/workflows/wai-quantum-formats.yml
index 093378137..0e2f88e3b 100644
--- a/.github/workflows/wai-quantum-formats.yml
+++ b/.github/workflows/wai-quantum-formats.yml
@@ -129,7 +129,13 @@ jobs:
           rustup target add --toolchain 1.98.0 wasm32-wasip2
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -139,11 +145,11 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
       # A debug build, so integer overflow panics instead of wrapping. The test
@@ -153,7 +159,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: |
           set -euo pipefail
           cargo test --locked --target wasm32-wasip2 --features full --test formats_conformance 2>&1 | tee "$RUNNER_TEMP/wasip2.log"
@@ -165,7 +171,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
           CARGO_PROFILE_RELEASE_OVERFLOW_CHECKS: "true"
         run: |
           set -euo pipefail
diff --git a/.github/workflows/wai-quantum-lang.yml b/.github/workflows/wai-quantum-lang.yml
index a0842fe9c..2cf0b349b 100644
--- a/.github/workflows/wai-quantum-lang.yml
+++ b/.github/workflows/wai-quantum-lang.yml
@@ -116,7 +116,13 @@ jobs:
           rustup target add --toolchain 1.98.0 wasm32-wasip2
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -126,11 +132,11 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
       # A debug build, so integer overflow panics instead of wrapping. The test
@@ -140,7 +146,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: |
           set -euo pipefail
           cargo test --locked --target wasm32-wasip2 --features full --test qasm3_conformance 2>&1 | tee "$RUNNER_TEMP/wasip2.log"
diff --git a/.github/workflows/wai-quantum-pulse.yml b/.github/workflows/wai-quantum-pulse.yml
index 9ec9024a0..4ef882a75 100644
--- a/.github/workflows/wai-quantum-pulse.yml
+++ b/.github/workflows/wai-quantum-pulse.yml
@@ -172,7 +172,13 @@ jobs:
           rustup target add --toolchain 1.98.0 wasm32-wasip2
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -182,11 +188,11 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
       # A debug build, so integer overflow panics instead of wrapping.
@@ -194,7 +200,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: |
           set -euo pipefail
           cargo test --locked --target wasm32-wasip2 --features full --lib -- qasm::pulse:: quantum_pulse:: engine::pulse:: 2>&1 | tee "$RUNNER_TEMP/wasip2.log"
diff --git a/.github/workflows/wai-quantum-registry.yml b/.github/workflows/wai-quantum-registry.yml
index 52a7a0471..502f2257e 100644
--- a/.github/workflows/wai-quantum-registry.yml
+++ b/.github/workflows/wai-quantum-registry.yml
@@ -140,7 +140,13 @@ jobs:
           rustup target add --toolchain 1.98.0 wasm32-wasip2
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -150,11 +156,11 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
       # A debug build, so integer overflow panics instead of wrapping.
@@ -162,7 +168,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: |
           set -euo pipefail
           cargo test --locked --target wasm32-wasip2 --features full --lib engine::registry:: 2>&1 | tee "$RUNNER_TEMP/wasip2.log"
diff --git a/.github/workflows/wai-quantum-verdict.yml b/.github/workflows/wai-quantum-verdict.yml
index 597d631c9..6fc3e62f9 100644
--- a/.github/workflows/wai-quantum-verdict.yml
+++ b/.github/workflows/wai-quantum-verdict.yml
@@ -135,7 +135,13 @@ jobs:
           rustup target add --toolchain 1.98.0 wasm32-wasip2
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
       # The release binary, pinned by version and SHA-256, rather than a build from source.
-      - name: Install wasmtime 49.0.2
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
@@ -145,11 +151,11 @@ jobs:
             $SUDO apt-get update -qq
             $SUDO apt-get install -y -qq xz-utils
           fi
-          v=49.0.2
+          v=50.0.1
           f="wasmtime-v$v-x86_64-linux.tar.xz"
           curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
             "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
-          echo "a4d6e9e3a5a60f527cf7793d674c48930c80c2e8977995b8a275cad3254b9322  $RUNNER_TEMP/$f" | sha256sum -c -
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
           tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
       # A debug build, so integer overflow panics instead of wrapping. The test
@@ -159,7 +165,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: |
           set -euo pipefail
           cargo test --locked --target wasm32-wasip2 --features full --test intent_conformance 2>&1 | tee "$RUNNER_TEMP/wasip2.log"
@@ -170,7 +176,7 @@ jobs:
         shell: bash
         working-directory: wai/wai-quantum
         env:
-          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
         run: |
           set -euo pipefail
           cargo test --locked --target wasm32-wasip2 --features full --lib -- engine::verdict:: engine::repair:: engine::refusal:: 2>&1 | tee "$RUNNER_TEMP/wasip2-lib.log"
diff --git a/.github/workflows/wai-vector.yml b/.github/workflows/wai-vector.yml
index 10a2e156d..01757a92a 100644
--- a/.github/workflows/wai-vector.yml
+++ b/.github/workflows/wai-vector.yml
@@ -178,14 +178,29 @@ jobs:
           rustup target add wasm32-wasip2
           echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
 
-      - name: Install wasmtime 38.0.3
+      # The release binary, pinned by version and SHA-256, rather than a build from source.
+      # Every WAI workflow that runs wasmtime pins this one version, and every run passes
+      # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
+      # instruction makes is the canonical one, and each relaxed SIMD instruction has one
+      # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
+      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
+      # where WebAssembly's deterministic profile prescribes the unfused result.
+      - name: Install wasmtime 50.0.1
         shell: bash
         run: |
           set -euo pipefail
-          v=38.0.3
-          curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/wasmtime.tar.xz" \
-            "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/wasmtime-v$v-x86_64-linux.tar.xz"
-          tar -C "$RUNNER_TEMP" -xf "$RUNNER_TEMP/wasmtime.tar.xz"
+          [ "$(uname -m)" = x86_64 ] || { echo "::error::the pinned wasmtime archive is x86_64; this host is $(uname -m)"; exit 1; }
+          if ! command -v xz >/dev/null 2>&1; then
+            SUDO=""; [ "$(id -u)" -ne 0 ] && SUDO="sudo"
+            $SUDO apt-get update -qq
+            $SUDO apt-get install -y -qq xz-utils
+          fi
+          v=50.0.1
+          f="wasmtime-v$v-x86_64-linux.tar.xz"
+          curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
+            "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
+          tar -xJf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
           echo "$RUNNER_TEMP/wasmtime-v$v-x86_64-linux" >> "$GITHUB_PATH"
 
       # The tests read committed files through CARGO_MANIFEST_DIR/.., so the wai
@@ -196,7 +211,7 @@ jobs:
         run: |
           set -euo pipefail
           W=$(cd .. && pwd)
-          CARGO_TARGET_WASM32_WASIP2_RUNNER="wasmtime run --dir $W::$W" \
+          CARGO_TARGET_WASM32_WASIP2_RUNNER="wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir $W::$W" \
             cargo test --locked --release --target wasm32-wasip2
 
       - name: Build the tool for wasm32-wasip2
@@ -210,7 +225,7 @@ jobs:
         run: |
           set -euo pipefail
           wasmtime --version
-          wasmtime run --dir .::/wai wai-vector/target/wasm32-wasip2/release/wai-vector.wasm vectors verify /wai
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir .::/wai wai-vector/target/wasm32-wasip2/release/wai-vector.wasm vectors verify /wai
 
   node:
     needs: relevant
-- 
2.54.0 (Apple Git-157)


From 290410864df7cdc1d6bac567e21d822176281349 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Tue, 3 Nov 2026 22:44:27 -0500
Subject: [PATCH 07/29] wai-web-wasm: the packages validate with relaxed SIMD
 and threads off
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

determinism-tiers §5 says a WebAssembly decoder at decode-equivalence MUST
NOT use relaxed SIMD, and that the browser crate was validated without it by
hand, "which CI does not check". CI now checks it. wasm-tools 1.250.0, pinned
by version and SHA-256, validates the node bundle and the platform_decode
bundle the job builds, and the package of record the site serves
(reach/record/wai_web_bg.wasm), with --features=-relaxed-simd,-threads. A
relaxed SIMD instruction, a shared memory or an atomic in any of them fails
the job.

Before the packages, the step shows that it can fail. A planted module with
i8x16.relaxed_swizzle and one with a shared memory each validate with the
default features, and each is refused with them off for its own feature's
reason (the error text is matched).

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-web-wasm.yml | 48 +++++++++++++++++++++++++++++-
 1 file changed, 47 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/wai-web-wasm.yml b/.github/workflows/wai-web-wasm.yml
index 0d018d5fb..454539028 100644
--- a/.github/workflows/wai-web-wasm.yml
+++ b/.github/workflows/wai-web-wasm.yml
@@ -31,7 +31,11 @@ name: wai-web wasm32
 #     Python verifier with the same verdicts;
 #   - a byte comparison of the demo's staged int_synth copies with the corpus;
 #   - the crate's wasm_tests under wasm-bindgen-test-runner, and its native unit
-#     tests on the host.
+#     tests on the host;
+#   - wasm-tools validate on both bundles built here and on the package of record
+#     (reach/record/wai_web_bg.wasm) with relaxed SIMD and threads turned off: a
+#     WebAssembly decoder at decode-equivalence uses no relaxed SIMD instruction
+#     (determinism-tiers §5), and the bundles use no shared memory or atomic.
 #
 # The full-scale int_hyper wasm decode needs vectors regenerated from a
 # pretrained model; wai-sota-vectors.yml runs it after regenerating them.
@@ -148,6 +152,48 @@ jobs:
             target/wasm32-unknown-unknown/release/wai_web.wasm
           test -s demo/pkg-platform/wai_platform_decode.js
           grep -q "platformChoose" demo/pkg-platform/wai_platform_decode.js
+      # The release binary, pinned by version and SHA-256, rather than a build from source.
+      - name: Install wasm-tools 1.250.0
+        shell: bash
+        run: |
+          set -euo pipefail
+          [ "$(uname -m)" = x86_64 ] || { echo "::error::the pinned wasm-tools archive is x86_64; this host is $(uname -m)"; exit 1; }
+          v=1.250.0
+          f="wasm-tools-$v-x86_64-linux.tar.gz"
+          curl --proto '=https' --tlsv1.2 -sSfL -o "$RUNNER_TEMP/$f" \
+            "https://github.com/bytecodealliance/wasm-tools/releases/download/v$v/$f"
+          echo "b746c34e7c4162b8812eb29397ebe076834e496a8c46fe68d793379a2741eb50  $RUNNER_TEMP/$f" | sha256sum -c -
+          tar -xzf "$RUNNER_TEMP/$f" -C "$RUNNER_TEMP"
+          echo "$RUNNER_TEMP/wasm-tools-$v-x86_64-linux" >> "$GITHUB_PATH"
+      # A WebAssembly decoder at decode-equivalence MUST NOT use a relaxed SIMD
+      # instruction, whose result the specification lets each engine choose
+      # (determinism-tiers §5). Both bundles built above and the package of record the
+      # site serves must validate with relaxed SIMD and threads (shared memory and
+      # atomics) turned off. First the check is shown able to fail: a planted relaxed
+      # SIMD module and a planted shared-memory module are each valid with the features
+      # on, and refused with them off for the feature's own reason.
+      - name: wai-web's packages validate with relaxed SIMD and threads off
+        shell: bash
+        working-directory: wai/wai-web
+        run: |
+          set -euo pipefail
+          F=--features=-relaxed-simd,-threads
+          wasm-tools --version
+          printf '(module (func (export "f") (param v128 v128) (result v128) local.get 0 local.get 1 i8x16.relaxed_swizzle))' > "$RUNNER_TEMP/relaxed.wat"
+          printf '(module (memory 1 1 shared) (func (export "f") (result i32) i32.const 0 i32.atomic.load))' > "$RUNNER_TEMP/shared.wat"
+          for p in "relaxed:relaxed SIMD support is not enabled" "shared:threads must be enabled"; do
+            m="$RUNNER_TEMP/${p%%:*}.wat"
+            wasm-tools validate "$m"
+            if wasm-tools validate "$F" "$m" 2> "$RUNNER_TEMP/planted.err"; then
+              echo "::error::the planted module ${p%%:*}.wat validated with $F"; exit 1
+            fi
+            grep -q "${p#*:}" "$RUNNER_TEMP/planted.err" || { echo "::error::${p%%:*}.wat was refused for another reason"; cat "$RUNNER_TEMP/planted.err"; exit 1; }
+            echo "planted ${p%%:*}.wat: refused ($(grep -o "${p#*:}" "$RUNNER_TEMP/planted.err"))"
+          done
+          for m in demo/pkg-node/wai_web_bg.wasm demo/pkg-platform/wai_platform_decode_bg.wasm reach/record/wai_web_bg.wasm; do
+            wasm-tools validate "$F" "$m" || { echo "::error::$m does not validate with relaxed SIMD and threads off"; exit 1; }
+            echo "$m (SHA-256 $(sha256sum "$m" | cut -d' ' -f1)): valid with relaxed SIMD and threads off"
+          done
       # The JS surface itself (pick, companionsOf, verifyPin, rendition() and
       # companion() refusing what is never presented or does not match its digest,
       # the held-sets conversion, and the WAI2 parse at 32-bit usize) runs only on
-- 
2.54.0 (Apple Git-157)


From b57022ecfea66248c97c8e72fb3644b7d67520de Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Tue, 3 Nov 2026 22:44:27 -0500
Subject: [PATCH 08/29] wai-byte-equality: a full runner disk fails as
 infrastructure, before the checkout

Main's run 17676 at 2bef11f71 failed on both arm64 legs with "No space left
on device": linux-aarch64 in the checkout's `git fetch`, and macOS in the
toolchain's temporary files. Both runners live on one host, whose disk was
full. No fixture was decoded, yet the cross-architecture matrix read as red.

Each leg now begins, before the checkout, with a step named
"infrastructure: this runner's disk has room for the leg". It measures the
free space under the workspace, the runner's temporary directory, TMPDIR and
HOME. Below a floor of 4 GiB it fails with an error that begins
"infrastructure:" and says that no fixture was decoded. The leg itself writes
under 1 GiB (measured on the build host: about 270 MiB of working tree and a
67 MiB release target), and a cold toolchain install about 1 GiB more.

A disk can also fill while the leg runs, from the jobs that share it. A
second step runs only when a step after the check failed. It measures again,
and names the failure infrastructure when the disk is below the floor, or
says that the failure is not the disk.

Both steps are bash and skip on Windows, as the matrix comment says.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-byte-equality.yml | 55 +++++++++++++++++++++++++
 1 file changed, 55 insertions(+)

diff --git a/.github/workflows/wai-byte-equality.yml b/.github/workflows/wai-byte-equality.yml
index 16f3c1580..deab15f3f 100644
--- a/.github/workflows/wai-byte-equality.yml
+++ b/.github/workflows/wai-byte-equality.yml
@@ -164,11 +164,44 @@ jobs:
         # Windows x86_64 leg removed 2026-09-30: runner offline. Restore by
         # re-adding `- { os: windows, label: windows-x86_64 }` here once the
         # runner is back online.
+        # The two disk steps below are bash, and skip on Windows.
         include:
           - { os: linux,       label: linux-x86_64 }
           - { os: linux-arm64, label: linux-aarch64 }
           - { os: macos,       label: macos-arm64 (reference) }
+    env:
+      # The floor of the disk check below, in GiB. The leg writes under 1 GiB (a
+      # shallow checkout of about 0.3 GiB plus its pack, and a release target under
+      # 0.1 GiB, measured 2026-11-03), and a cold toolchain install about 1 GiB more
+      # under HOME; the rest is room for the jobs that share the disk.
+      DISK_FLOOR_GIB: '4'
     steps:
+      # Infrastructure first, before the checkout writes anything. Main's run 17676
+      # failed on both arm64 legs with "No space left on device", linux-aarch64 in
+      # `git fetch` and macOS in the toolchain's temporary files: the hosts' disks were
+      # full and no fixture was decoded, yet the matrix read as a byte-equality failure.
+      # This step measures the free space under every directory the leg writes to (the
+      # workspace, the runner's temporary directory, TMPDIR, and HOME, where the
+      # toolchain lives). Below the floor it fails, and its name and its error say
+      # infrastructure, so a full disk is not read as a divergence.
+      - name: "infrastructure: this runner's disk has room for the leg"
+        id: disk
+        if: matrix.os != 'windows'
+        shell: bash
+        run: |
+          set -euo pipefail
+          short=0
+          for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
+            p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
+            set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
+            echo "$d: $(($1 / 1048576)) GiB free on $2"
+            if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
+              echo "::error::infrastructure: ${{ matrix.label }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The runner host's disk is full: no fixture was decoded, so this is not a byte-equality failure."
+              short=1
+            fi
+          done
+          exit $short
+
       - name: Checkout
         uses: actions/checkout@v4
 
@@ -195,3 +228,25 @@ jobs:
       - name: Decode fixtures, assert golden bytes
         working-directory: wai/byte-exact-conformance
         run: cargo test --release -- --nocapture
+
+      # A disk can also fill while the leg runs, from the jobs that share it (three
+      # conformance jobs at once filled the arm64 VM's data disk mid-link). When a step
+      # after the check failed, the same measurement is taken again: below the floor,
+      # the failure is named infrastructure; with room, it stands as it is.
+      - name: "infrastructure: was the disk full when the leg failed?"
+        if: failure() && steps.disk.outcome == 'success'
+        shell: bash
+        run: |
+          set -euo pipefail
+          short=0
+          for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
+            p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
+            set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
+            echo "$d: $(($1 / 1048576)) GiB free on $2"
+            if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
+              echo "::error::infrastructure: ${{ matrix.label }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
+              short=1
+            fi
+          done
+          [ "$short" = 1 ] || echo "Every disk this leg writes to has room: the failure above is not the disk."
+          exit $short
-- 
2.54.0 (Apple Git-157)


From 888cc3ba20de8a94a7a9d442922d89e330443495 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:51:25 -0500
Subject: [PATCH 09/29] CI: a check holds every workflow to the one pinned
 wasmtime and both its settings

The install steps' comment said every WAI workflow that runs wasmtime pins one
version and passes both determinism settings, but nothing enforced it: a step
that ran a stock or differently pinned wasmtime would have merged green.
scripts/check-wasmtime-pin.py now fails, in the repository checks, any workflow
that

1. downloads a wasmtime release other than 50.0.1, or checks it against any
   SHA-256 but that archive's;
2. installs wasmtime another way (cargo install, the install script, an action);
3. runs wasmtime (run, serve, compile, or a module given to it) or names it as
   a CARGO_TARGET_*_RUNNER without -W nan-canonicalization=y and
   -W relaxed-simd-deterministic=y.

It also fails when it finds no download or no run, so a check that looked in
the wrong place cannot pass. On this tree it finds the downloads and runs it
should. Its self-test (scripts/check-wasmtime-pin-selftest.sh) passes a scratch
workflow shaped like these and fails each way of breaking it that the
self-test lists, matching the reason in the output.

The comment's relaxed_madd claim was measured on arm64 only. It now rests on
both architectures, measured on an arm64 macOS host and an x86_64 Linux host
with a small probe (wasmtime 50.0.1's release for the host, checked against
the digest the release lists):

    host Darwin arm64; wasmtime 50.0.1 (c307ec81d 2026-10-29)
    f32.add(bits 0x7fa00001, 1.0)                     [no flags]: 0x7fe00001
    f32x4.relaxed_madd(1+2^-12, 1+2^-12, -(1+2^-11)) [no flags]: 0x33800000
    f32.add(bits 0x7fa00001, 1.0)                     [both flags]: 0x7fc00000
    f32x4.relaxed_madd(1+2^-12, 1+2^-12, -(1+2^-11)) [both flags]: 0x33800000
    an unknown -W option is refused: error: failed to parse -W / --wasm option:
      unknown option: no-such-option

    host Linux x86_64; wasmtime 50.0.1 (c307ec81d 2026-10-29)
    (the same five lines, bit for bit)

0x33800000 is 2^-24, the fused result; the unfused fadd(fmul) gives 0. So the
NaN setting takes effect on both, and relaxed_madd is fused on both with and
without the relaxed setting. That matches 50.0.1's source (the
wasmtime-internal-cranelift crate, read from the registry): the translator
lowers F32x4/F64x2RelaxedMadd through fma_f32x4/fma_f64x2
(src/translate/code_translator.rs L2530-2539), which emit a native fma on a
host that has one, a fused library call under relaxed-simd-deterministic on a
host that has none, and the unfused fmul and fadd only on a host with no FMA
and the setting off (src/func_environ.rs L5907-5925). The setting is what
makes the result fused on every host. And since an unknown -W option stops
wasmtime, a run that exits 0 with the two flags has parsed them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/repo-checks.yml          |   8 ++
 .github/workflows/wai-conformance.yml      |  12 +-
 .github/workflows/wai-ops.yml              |   6 +-
 .github/workflows/wai-quantum-cli.yml      |   6 +-
 .github/workflows/wai-quantum-formats.yml  |   6 +-
 .github/workflows/wai-quantum-lang.yml     |   6 +-
 .github/workflows/wai-quantum-pulse.yml    |   6 +-
 .github/workflows/wai-quantum-registry.yml |   6 +-
 .github/workflows/wai-quantum-verdict.yml  |   6 +-
 .github/workflows/wai-vector.yml           |   6 +-
 scripts/check-wasmtime-pin-selftest.sh     | 107 +++++++++++++++
 scripts/check-wasmtime-pin.py              | 144 +++++++++++++++++++++
 12 files changed, 299 insertions(+), 20 deletions(-)
 create mode 100755 scripts/check-wasmtime-pin-selftest.sh
 create mode 100755 scripts/check-wasmtime-pin.py

diff --git a/.github/workflows/repo-checks.yml b/.github/workflows/repo-checks.yml
index 3fd6a9476..81791ead5 100644
--- a/.github/workflows/repo-checks.yml
+++ b/.github/workflows/repo-checks.yml
@@ -87,6 +87,14 @@ jobs:
         if: ${{ !cancelled() }}
         run: ./scripts/pr-touches-paths-selftest.sh
 
+      # ── One pinned wasmtime, run with both determinism settings ─────────────
+      - name: "workflow hygiene: the wasmtime pin check fails closed (self-test)"
+        if: ${{ !cancelled() }}
+        run: bash scripts/check-wasmtime-pin-selftest.sh
+      - name: "workflow hygiene: every wasmtime is the pinned release, run with both determinism settings"
+        if: ${{ !cancelled() }}
+        run: python3 -I scripts/check-wasmtime-pin.py
+
       # ── Code cites no section its SPEC lacks; WAI's cited sections exist ────
       - name: "spec references: the check can fail (self-test)"
         if: ${{ !cancelled() }}
diff --git a/.github/workflows/wai-conformance.yml b/.github/workflows/wai-conformance.yml
index cd3b57a60..db7fe7c5e 100644
--- a/.github/workflows/wai-conformance.yml
+++ b/.github/workflows/wai-conformance.yml
@@ -973,8 +973,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
@@ -1123,8 +1125,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-ops.yml b/.github/workflows/wai-ops.yml
index 996ad1b1a..00273b74c 100644
--- a/.github/workflows/wai-ops.yml
+++ b/.github/workflows/wai-ops.yml
@@ -212,8 +212,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-quantum-cli.yml b/.github/workflows/wai-quantum-cli.yml
index 93916b1a1..cf3d37363 100644
--- a/.github/workflows/wai-quantum-cli.yml
+++ b/.github/workflows/wai-quantum-cli.yml
@@ -115,8 +115,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-quantum-formats.yml b/.github/workflows/wai-quantum-formats.yml
index 0e2f88e3b..459cfe70e 100644
--- a/.github/workflows/wai-quantum-formats.yml
+++ b/.github/workflows/wai-quantum-formats.yml
@@ -133,8 +133,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-quantum-lang.yml b/.github/workflows/wai-quantum-lang.yml
index 2cf0b349b..752323b87 100644
--- a/.github/workflows/wai-quantum-lang.yml
+++ b/.github/workflows/wai-quantum-lang.yml
@@ -120,8 +120,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-quantum-pulse.yml b/.github/workflows/wai-quantum-pulse.yml
index 4ef882a75..d8d878a81 100644
--- a/.github/workflows/wai-quantum-pulse.yml
+++ b/.github/workflows/wai-quantum-pulse.yml
@@ -176,8 +176,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-quantum-registry.yml b/.github/workflows/wai-quantum-registry.yml
index 502f2257e..e6be4c3fc 100644
--- a/.github/workflows/wai-quantum-registry.yml
+++ b/.github/workflows/wai-quantum-registry.yml
@@ -144,8 +144,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-quantum-verdict.yml b/.github/workflows/wai-quantum-verdict.yml
index 6fc3e62f9..033e5e093 100644
--- a/.github/workflows/wai-quantum-verdict.yml
+++ b/.github/workflows/wai-quantum-verdict.yml
@@ -139,8 +139,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/.github/workflows/wai-vector.yml b/.github/workflows/wai-vector.yml
index 01757a92a..21da478c6 100644
--- a/.github/workflows/wai-vector.yml
+++ b/.github/workflows/wai-vector.yml
@@ -183,8 +183,10 @@ jobs:
       # -W nan-canonicalization=y -W relaxed-simd-deterministic=y: every NaN a float
       # instruction makes is the canonical one, and each relaxed SIMD instruction has one
       # behaviour on every host. A stock wasmtime leaves both off. The fixed relaxed
-      # behaviour is the engine's own: on 50.0.1 (measured on arm64) relaxed_madd is fused,
-      # where WebAssembly's deterministic profile prescribes the unfused result.
+      # behaviour is the engine's own: on 50.0.1 relaxed_madd is fused (measured on arm64 and
+      # x86_64), where WebAssembly's deterministic profile prescribes the unfused result.
+      # The repository checks (scripts/check-wasmtime-pin.py) fail a workflow that pins another
+      # release or runs wasmtime without both settings.
       - name: Install wasmtime 50.0.1
         shell: bash
         run: |
diff --git a/scripts/check-wasmtime-pin-selftest.sh b/scripts/check-wasmtime-pin-selftest.sh
new file mode 100755
index 000000000..4c68863d7
--- /dev/null
+++ b/scripts/check-wasmtime-pin-selftest.sh
@@ -0,0 +1,107 @@
+#!/usr/bin/env bash
+#
+# check-wasmtime-pin-selftest.sh — check-wasmtime-pin.py can fail.
+#
+# A check that cannot fail is decoration. This builds a scratch tree with one workflow
+# that installs the pinned wasmtime and runs it the ways the repository's workflows do,
+# asserts that the check passes it, and then breaks it each way the property can break
+# and asserts that the check fails, saying why.
+set -uo pipefail
+
+here=$(cd "$(dirname "$0")" && pwd)
+check="$here/check-wasmtime-pin.py"
+tmp=$(mktemp -d)
+trap 'rm -rf "$tmp"' EXIT
+W="$tmp/.github/workflows/w.yml"
+
+mk() {
+  rm -rf "$tmp/.github"
+  mkdir -p "$tmp/.github/workflows"
+  cat > "$W" <<'YML'
+jobs:
+  t:
+    steps:
+      # Every run passes -W nan-canonicalization=y -W relaxed-simd-deterministic=y.
+      - name: Install wasmtime 50.0.1
+        run: |
+          v=50.0.1
+          f="wasmtime-v$v-x86_64-linux.tar.xz"
+          curl -sSfL -o "$RUNNER_TEMP/$f" \
+            "https://github.com/bytecodealliance/wasmtime/releases/download/v$v/$f"
+          echo "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5  $RUNNER_TEMP/$f" | sha256sum -c -
+      - name: tests under wasmtime
+        env:
+          CARGO_TARGET_WASM32_WASIP2_RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y
+        run: cargo test --target wasm32-wasip2
+      - name: the corpus under wasmtime
+        run: |
+          W=$(pwd)
+          CARGO_TARGET_WASM32_WASIP2_RUNNER="wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir $W::$W" \
+            cargo test --release --target wasm32-wasip2
+          wasmtime --version
+          wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir .::/w \
+            x.wasm verify /w
+          wasmtime run \
+            -W nan-canonicalization=y -W relaxed-simd-deterministic=y x.wasm
+          cmp a b || { echo "::error::the output differs under wasmtime"; exit 1; }
+YML
+}
+
+rep() { # rep <old> <new>: replace the first occurrence in the scratch workflow
+  python3 -I -c 'import sys; p, a, b = sys.argv[1:]; s = open(p).read(); assert a in s, a; open(p, "w").write(s.replace(a, b, 1))' "$W" "$1" "$2"
+}
+
+fails=0
+expect() { # expect <pass|fail> <what> [text the output must hold]
+  local want=$1 what=$2 needle=${3:-} out rc
+  out=$(python3 -I "$check" --root "$tmp" 2>&1); rc=$?
+  if [ "$want" = pass ] && [ $rc -ne 0 ]; then echo "FAIL  $what: the check failed"; printf '%s\n' "$out" | sed 's/^/      /'; fails=$((fails+1)); return; fi
+  if [ "$want" = fail ] && [ $rc -eq 0 ]; then echo "FAIL  $what: the check passed"; fails=$((fails+1)); return; fi
+  if [ -n "$needle" ] && ! grep -qF -- "$needle" <<<"$out"; then echo "FAIL  $what: the output does not say '$needle'"; printf '%s\n' "$out" | sed 's/^/      /'; fails=$((fails+1)); return; fi
+  echo "ok    $what"
+}
+
+mk; expect pass "the pinned release, run every way with both settings, passes"
+
+mk; rep "v=50.0.1" "v=49.0.2"
+expect fail "another version" "downloads wasmtime 49.0.2"
+
+mk; rep '/download/v$v/$f"' '/download/v38.0.3/wasmtime-v38.0.3-x86_64-linux.tar.xz"'; rep "v=50.0.1" ":"
+expect fail "another version, named in the URL" "downloads wasmtime 38.0.3"
+
+mk; rep "685466455dd4" "685466455dd5"
+expect fail "another digest" "not its pinned digest"
+
+mk; rep '| sha256sum -c -' '> /dev/null'
+expect fail "no digest checked" "checks no SHA-256"
+
+mk; rep "x86_64-linux.tar.xz" "aarch64-linux.tar.xz"
+expect fail "an archive with no pinned digest" "has no pinned digest here"
+
+mk; rep "    steps:" "    steps:
+      - run: cargo install wasmtime-cli --locked"
+expect fail "wasmtime built from source" "installs wasmtime some other way"
+
+mk; rep "RUNNER: wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y" "RUNNER: wasmtime"
+expect fail "a cargo runner with neither setting" "runs wasmtime without -W nan-canonicalization=y and -W relaxed-simd-deterministic=y"
+
+mk; rep '--dir $W::$W"' '--dir $W::$W -W nan-canonicalization=y"'; rep 'RUNNER="wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' 'RUNNER="wasmtime run'
+expect fail "a runner with one setting" "runs wasmtime without -W relaxed-simd-deterministic=y"
+
+mk; rep "-W relaxed-simd-deterministic=y --dir .::/w" "--dir .::/w"
+expect fail "a run without deterministic relaxed SIMD" "runs wasmtime without -W relaxed-simd-deterministic=y"
+
+mk; rep "            -W nan-canonicalization=y -W relaxed-simd-deterministic=y x.wasm" "            x.wasm"
+expect fail "a run continued over two lines, with neither setting" "runs wasmtime without"
+
+mk; rep "wasmtime --version" "wasmtime x.wasm"
+expect fail "a module given to wasmtime with no subcommand" "runs wasmtime without"
+
+mk; rep "wasmtime --version" "wasmtime serve -W nan-canonicalization=y app.wasm"
+expect fail "a serve without deterministic relaxed SIMD" "runs wasmtime without -W relaxed-simd-deterministic=y"
+
+mk; printf 'jobs:\n  t:\n    steps:\n      - run: true\n' > "$W"
+expect fail "no workflow that runs wasmtime" "looked in the wrong place"
+
+if [ $fails -ne 0 ]; then echo "check-wasmtime-pin-selftest: $fails failure(s)"; exit 1; fi
+echo "check-wasmtime-pin-selftest: the check passes the pinned, flagged runs and fails each other."
diff --git a/scripts/check-wasmtime-pin.py b/scripts/check-wasmtime-pin.py
new file mode 100755
index 000000000..1550e0da2
--- /dev/null
+++ b/scripts/check-wasmtime-pin.py
@@ -0,0 +1,144 @@
+#!/usr/bin/env python3
+"""check-wasmtime-pin.py: every workflow that runs wasmtime runs the one pinned release,
+with both of its determinism settings.
+
+WAI's WebAssembly legs hold a module's output to pinned bits. A stock wasmtime leaves NaN
+canonicalisation off and lets each relaxed SIMD instruction take the host's behaviour, and
+two versions of it can lower one instruction two ways, so a leg's bits would be the
+engine's choice. The install steps' comment says every workflow pins one version and
+passes both settings; this makes that true of every workflow, not only of the ones that
+comment sits in.
+
+In every `.github/workflows/*.yml`, with continued lines (`\\` at the end) joined, this
+requires that
+
+  1. every download of a wasmtime release names version PIN (the `v=` the step sets, or
+     the version in the URL), and the SHA-256 it checks is the digest of that archive in
+     DIGESTS;
+  2. wasmtime is installed no other way (`cargo install wasmtime-cli`, the install
+     script, or an action);
+  3. every `wasmtime run`, `wasmtime serve` or `wasmtime compile`, every `wasmtime` given
+     a module path, and every `CARGO_TARGET_<triple>_RUNNER` that is wasmtime, passes
+     `-W nan-canonicalization=y` and `-W relaxed-simd-deterministic=y`;
+  4. at least one download and one run were found, so a check that looked in the wrong
+     place fails rather than passing on nothing.
+
+    check-wasmtime-pin.py [--root DIR]
+
+Exit 0 when all four hold, 1 with one line a finding otherwise. Standard library only.
+"""
+import os
+import re
+import sys
+
+PIN = "50.0.1"
+# The release archives a workflow may download, by name, with the SHA-256 the release lists.
+DIGESTS = {
+    "wasmtime-v50.0.1-x86_64-linux.tar.xz": "685466455dd42a14774a86c36c8f2e45cbb9bd6d92a850139315df60a75498a5",
+}
+FLAGS = ("nan-canonicalization=y", "relaxed-simd-deterministic=y")
+
+DOWNLOAD = re.compile(r"bytecodealliance/wasmtime/releases/download/")
+URL_VERSION = re.compile(r"releases/download/v(\d[^/\"'\s]*)/")
+SET_VERSION = re.compile(r"^\s*v=[\"']?([^\"'\s;]+)")
+ARCHIVE = re.compile(r"wasmtime-v(\$v|\$\{v\}|\d[^-\"'\s]*)-([a-z0-9_]+-[a-z]+)\.tar\.xz")
+DIGEST = re.compile(r"\b([0-9a-f]{64})\b.*sha256sum -c")
+OTHER_INSTALL = re.compile(
+    r"cargo\s+(?:\S+\s+)*install\s+(?:\S+\s+)*wasmtime-cli|wasmtime\.dev/install|uses:\s*\S*wasmtime"
+)
+# wasmtime as a command: run, serve or compile, or given a module path (`wasmtime x.wasm`).
+INVOKE = re.compile(
+    r"(?:^|[\s;&|(`/\"'=])wasmtime\s+(?:(?:run|serve|compile)\b|[^\s-]\S*\.(?:c?wasm|wat)\b)"
+)
+RUNNER = re.compile(r"CARGO_TARGET_\w+_RUNNER\s*[:=]\s*[\"']?wasmtime\b")
+
+
+def logical_lines(text):
+    """(first line number, text) with backslash-continued lines joined."""
+    out, buf, start = [], "", 0
+    for n, line in enumerate(text.splitlines(), 1):
+        if not buf:
+            start = n
+        if line.rstrip().endswith("\\"):
+            buf += line.rstrip()[:-1] + " "
+            continue
+        out.append((start, buf + line))
+        buf = ""
+    if buf:
+        out.append((start, buf))
+    return out
+
+
+def check_file(path, rel, findings):
+    lines = logical_lines(open(path, encoding="utf-8").read())
+    downloads = runs = 0
+    for i, (n, line) in enumerate(lines):
+        code = line.lstrip()
+        if code.startswith("#"):
+            continue
+        if OTHER_INSTALL.search(line):
+            findings.append(f"{rel}:{n}: installs wasmtime some other way; download the pinned release ({PIN}) by version and SHA-256")
+        if DOWNLOAD.search(line):
+            downloads += 1
+            m = URL_VERSION.search(line)
+            ver = m.group(1) if m else None
+            if ver is None:  # the URL takes the version from the step's `v=`
+                for _, prev in reversed(lines[max(0, i - 12) : i]):
+                    s = SET_VERSION.match(prev)
+                    if s:
+                        ver = s.group(1)
+                        break
+            if ver != PIN:
+                findings.append(f"{rel}:{n}: downloads wasmtime {ver or '(version not found)'}; every workflow pins {PIN}")
+            near = lines[max(0, i - 12) : i + 5]
+            arch = None
+            for _, other in near:
+                a = ARCHIVE.search(other)
+                if a:
+                    arch = a.group(2)
+                    break
+            name = f"wasmtime-v{ver}-{arch}.tar.xz" if arch else None
+            digests = [d.group(1) for _, other in lines[i : i + 5] for d in [DIGEST.search(other)] if d]
+            if name not in DIGESTS:
+                findings.append(f"{rel}:{n}: downloads {name or 'an archive whose name was not found'}, which has no pinned digest here")
+            elif not digests:
+                findings.append(f"{rel}:{n}: downloads {name} and checks no SHA-256 within four lines")
+            elif digests[0] != DIGESTS[name]:
+                findings.append(f"{rel}:{n}: checks {name} against {digests[0][:12]}..., not its pinned digest {DIGESTS[name][:12]}...")
+        for rx in (INVOKE, RUNNER):
+            m = rx.search(line)
+            if m:
+                runs += 1
+                rest = line[m.start() :]
+                missing = [f for f in FLAGS if f not in rest]
+                if missing:
+                    findings.append(f"{rel}:{n}: runs wasmtime without {' and '.join('-W ' + f for f in missing)}")
+                break
+    return downloads, runs
+
+
+def main(argv):
+    root = "."
+    if len(argv) == 3 and argv[1] == "--root":
+        root = argv[2]
+    elif len(argv) != 1:
+        sys.exit("usage: check-wasmtime-pin.py [--root DIR]")
+    wf = os.path.join(root, ".github", "workflows")
+    findings, downloads, runs = [], 0, 0
+    for name in sorted(os.listdir(wf)) if os.path.isdir(wf) else []:
+        if name.endswith((".yml", ".yaml")):
+            d, r = check_file(os.path.join(wf, name), f".github/workflows/{name}", findings)
+            downloads += d
+            runs += r
+    if downloads == 0 or runs == 0:
+        findings.append(f"found {downloads} wasmtime download(s) and {runs} run(s) under {wf}: this check looked in the wrong place")
+    for f in findings:
+        print(f"::error::{f}")
+    if findings:
+        return 1
+    print(f"check-wasmtime-pin: {downloads} downloads of wasmtime {PIN}, each checked against its digest; {runs} runs, each with both settings")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv))
-- 
2.54.0 (Apple Git-157)


From 8a3141366d01c76c25951869257d65313b053b61 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:54:27 -0500
Subject: [PATCH 10/29] =?UTF-8?q?determinism-tiers=20=C2=A75:=20CI=20valid?=
 =?UTF-8?q?ates=20the=20browser=20packages;=20the=20READMEs=20run=20CI's?=
 =?UTF-8?q?=20wasmtime?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

§5 said the browser crate's SIMD validation was done by hand, "which CI does
not check". Since "wai-web-wasm: the packages validate with relaxed SIMD and
threads off", wai-web-wasm.yml validates both bundles and
reach/record/wai_web_bg.wasm with relaxed SIMD and threads disabled. §5 now
says so, and says what CI still does not check: that they hold no other SIMD
instruction. The changelog gains the CI entry for this branch.

The commands that reproduce CI's wasmtime legs ran a stock wasmtime: the WAI
Quantum READMEs (English and Chinese), the bit_identity test's header, the WAI
README's edge example, and the conformance READMEs (quantum, qasm3, formats,
answers, intents). Each now passes -W nan-canonicalization=y
-W relaxed-simd-deterministic=y, as CI does. The determinism-cost run scripts
under wai/research keep their commands: they record runs already made.

The quantum docs check could not see that drift. For a `norun=ci:` block it
stripped every leading VAR=value before matching the command against the
workflow, and its pattern, \S*, did not read a quoted value. It now reads a
quoted or bare value, and when the workflow sets VAR (as an env: key or in a
script) the documented value must be one the workflow gives it, so
CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime is refused against a workflow that
sets it to the flagged runner.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md                          | 14 +++++++++++
 wai/README.md                             |  4 ++--
 wai/extensions/determinism-tiers.md       |  5 +++-
 wai/quantum-conformance/README.md         |  2 +-
 wai/quantum-conformance/answers/README.md |  2 +-
 wai/quantum-conformance/formats/README.md |  4 ++--
 wai/quantum-conformance/intents/README.md |  2 +-
 wai/quantum-conformance/qasm3/README.md   |  2 +-
 wai/tools/quantum-docs/check_docs.py      | 29 +++++++++++++++++++----
 wai/wai-quantum/README.md                 |  2 +-
 wai/wai-quantum/README.zh-CN.md           |  2 +-
 wai/wai-quantum/tests/bit_identity.rs     |  2 +-
 12 files changed, 54 insertions(+), 16 deletions(-)

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index a57e82118..aaf219337 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -6,6 +6,20 @@ wire format may change and such changes are recorded here.
 
 ## Unreleased
 
+### CI: WebAssembly runs under one pinned wasmtime with both its determinism settings, and the browser packages hold no relaxed SIMD
+
+- **CI:** every workflow that runs a module under wasmtime runs wasmtime 50.0.1, pinned by version and SHA-256, with
+  `-W nan-canonicalization=y -W relaxed-simd-deterministic=y`: every NaN a float instruction makes is the canonical
+  one, and each relaxed SIMD instruction has one behaviour on every host. `wai-ops.yml` and `wai-vector.yml` ran
+  38.0.3, and the quantum and face jobs 49.0.2, each with neither setting. The repository checks
+  (`scripts/check-wasmtime-pin.py`) fail a workflow that pins another release or runs wasmtime without both.
+- **CI:** `wai-web-wasm.yml` validates both of `wai-web`'s bundles and `reach/record/wai_web_bg.wasm` with relaxed SIMD
+  and threads disabled (`wasm-tools` 1.250.0, pinned the same way), so a relaxed SIMD instruction, a shared memory or
+  an atomic in any of them fails the job ([determinism-tiers](extensions/determinism-tiers.md) §5).
+- The READMEs' commands that reproduce these jobs pass the same two settings. The quantum docs check holds a
+  `VAR=value` before a command a `ci:` block names to a value the workflow gives `VAR`, so a documented runner cannot
+  drift from CI's.
+
 ### CI: the maintainers' product names are checked on every pull request
 
 - **CI:** the public-name check applies the maintainers' product names, held in the script, on every change, a pull
diff --git a/wai/README.md b/wai/README.md
index 51364b22f..15f655a79 100644
--- a/wai/README.md
+++ b/wai/README.md
@@ -329,8 +329,8 @@ earlier releases gave.
 ```bash
 cd wai-quantum-cli && cargo build --release --target wasm32-wasip2 --no-default-features
 
-wasmtime run target/wasm32-wasip2/release/wai-quantum.wasm lab run ghz -n 5
-wasmtime run target/wasm32-wasip2/release/wai-quantum.wasm lab receipt ghz -n 3 --demo-key
+wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y target/wasm32-wasip2/release/wai-quantum.wasm lab run ghz -n 5
+wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y target/wasm32-wasip2/release/wai-quantum.wasm lab receipt ghz -n 3 --demo-key
 ```
 
 Both print the identical `state-hash`. A receipt sealed in cloud WASM verifies
diff --git a/wai/extensions/determinism-tiers.md b/wai/extensions/determinism-tiers.md
index bad01d753..778f75701 100644
--- a/wai/extensions/determinism-tiers.md
+++ b/wai/extensions/determinism-tiers.md
@@ -533,7 +533,10 @@ shows, by fabric:
   `wasm32-wasip2` under wasmtime (§6). The browser crate's release build contains no
   SIMD instruction: on the source of the change that added this subsection it
   validated with WebAssembly's SIMD and relaxed-SIMD features disabled
-  (`wasm-tools validate`), which CI does not check. A WebAssembly decoder at this
+  (`wasm-tools validate`). CI (`wai-web-wasm.yml`) validates both of the crate's
+  bundles and the package of record, `reach/record/wai_web_bg.wasm`, with relaxed SIMD
+  and threads disabled, and fails on a module that uses either. CI does not check that
+  they contain no other SIMD instruction. A WebAssembly decoder at this
   tier MUST NOT use relaxed-SIMD instructions, whose results the WebAssembly
   specification lets engines choose among, and MUST NOT depend on the bits of a NaN.
 - **GPU.** A decoder's `wai-ops/1-i32` graphs ([integer-operators](integer-operators.md)
diff --git a/wai/quantum-conformance/README.md b/wai/quantum-conformance/README.md
index f87283d9e..a05beafd9 100644
--- a/wai/quantum-conformance/README.md
+++ b/wai/quantum-conformance/README.md
@@ -43,7 +43,7 @@ Same binary, same source, different architecture:
 ```sh
 cargo build --release --target wasm32-wasip2 --no-default-features --features quantum \
   --bin wai_quantum_vectors
-wasmtime run --dir wai/quantum-conformance::/corpus \
+wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y --dir wai/quantum-conformance::/corpus \
   target/wasm32-wasip2/release/wai_quantum_vectors.wasm verify /corpus
 ```
 
diff --git a/wai/quantum-conformance/answers/README.md b/wai/quantum-conformance/answers/README.md
index 442a55bda..b99f0f7df 100644
--- a/wai/quantum-conformance/answers/README.md
+++ b/wai/quantum-conformance/answers/README.md
@@ -37,7 +37,7 @@ WAI_ANSWER_BUNDLE_WRITE=1 cargo test --release --features full --test answer_con
 ## WASI (the edge)
 
 ```sh
-CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime \
+CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' \
   cargo test --target wasm32-wasip2 --features full --test answer_conformance
 ```
 
diff --git a/wai/quantum-conformance/formats/README.md b/wai/quantum-conformance/formats/README.md
index 1329bfb35..2f215ae79 100644
--- a/wai/quantum-conformance/formats/README.md
+++ b/wai/quantum-conformance/formats/README.md
@@ -47,7 +47,7 @@ cargo test --release --features full --test formats_mutation -- --nocapture
 ## WASI (the edge)
 
 ```sh
-CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime \
+CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' \
   cargo test --target wasm32-wasip2 --features full --test formats_conformance
 ```
 
@@ -55,7 +55,7 @@ The cases come from the bundle compiled into the test. The hostile-input
 test runs there too, as a release build with overflow checks on:
 
 ```sh
-CARGO_PROFILE_RELEASE_OVERFLOW_CHECKS=true CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime \
+CARGO_PROFILE_RELEASE_OVERFLOW_CHECKS=true CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' \
   cargo test --release --target wasm32-wasip2 --features full --test formats_mutation
 ```
 
diff --git a/wai/quantum-conformance/intents/README.md b/wai/quantum-conformance/intents/README.md
index da6291705..a45a0b0fe 100644
--- a/wai/quantum-conformance/intents/README.md
+++ b/wai/quantum-conformance/intents/README.md
@@ -39,7 +39,7 @@ WAI_INTENT_BUNDLE_WRITE=1 cargo test --release --features full --test intent_con
 ## WASI (the edge)
 
 ```sh
-CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime \
+CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' \
   cargo test --target wasm32-wasip2 --features full --test intent_conformance
 ```
 
diff --git a/wai/quantum-conformance/qasm3/README.md b/wai/quantum-conformance/qasm3/README.md
index fd564b25a..dd5647379 100644
--- a/wai/quantum-conformance/qasm3/README.md
+++ b/wai/quantum-conformance/qasm3/README.md
@@ -35,7 +35,7 @@ WAI_QASM3_BUNDLE_WRITE=1 cargo test --release --features full --test qasm3_confo
 ## WASI (the edge)
 
 ```sh
-CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime \
+CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' \
   cargo test --target wasm32-wasip2 --features full --test qasm3_conformance
 ```
 
diff --git a/wai/tools/quantum-docs/check_docs.py b/wai/tools/quantum-docs/check_docs.py
index e52aec074..b1f1efc0f 100644
--- a/wai/tools/quantum-docs/check_docs.py
+++ b/wai/tools/quantum-docs/check_docs.py
@@ -35,8 +35,9 @@ fenced block is one of these, by its info string (the words after the fence):
 - ```` ```json ````: parses (`fragment` excepts one).
 - ```` ```<lang> norun=REASON ````: a block that cannot run here, for a stated
   reason: `ci:WORKFLOW` (the steps a CI workflow runs: each script it names
-  exists, and the workflow names it), `install` (each package it installs is
-  this repository's, at its version), `data` (it needs files the repository
+  exists, and the workflow names it; a leading `VAR=value` must be a value the
+  workflow gives VAR, where it sets VAR), `install` (each package it installs
+  is this repository's, at its version), `data` (it needs files the repository
   does not hold), `account` (it needs what only the person has: an account,
   a device, a program of their own), `server`
   (it starts a server that does not return), `usage` (a synopsis). Every
@@ -349,6 +350,17 @@ def static_line_checks(cli, block, line, report, pkgs, pys):
                         report.fail(block.where, f"pip install {t}: no Python package of this repository is named {name}")
 
 
+def workflow_values(text, var):
+    """Every value a workflow gives VAR: as an `env:` key (`VAR: value`) or in a script
+    (`VAR=value`, `VAR="value"`, `VAR='value'`), quotes off and blanks collapsed."""
+    vals = set()
+    for m in re.finditer(rf"^\s*{var}:[ \t]*(\S.*?)\s*$", text, re.M):
+        vals.add(" ".join(m.group(1).strip("'\"").split()))
+    for m in re.finditer(rf"\b{var}=(\"[^\"]*\"|'[^']*'|\S+)", text):
+        vals.add(" ".join(m.group(1).strip("'\"").split()))
+    return vals
+
+
 def check_norun(cli, block, report, pkgs, pys):
     reason = block.attrs.get("norun", "")
     kind, _, detail = reason.partition(":")
@@ -368,14 +380,23 @@ def check_norun(cli, block, report, pkgs, pys):
         if not detail or not wf.exists():
             report.fail(block.where, f"norun=ci:{detail}: no such workflow")
             return
-        flow = " ".join(wf.read_text(encoding="utf-8").split())
+        raw = wf.read_text(encoding="utf-8")
+        flow = " ".join(raw.split())
         for line in block.body:
             line = re.sub(r"\s+#.*$", "", line).strip()
             if not line or line.startswith("#"):
                 continue
             for seg in re.split(r"&&|;", line):
                 seg = seg.strip().strip("()").strip().rstrip("&").strip()
-                seg = re.sub(r"^([A-Z_][A-Z0-9_]*=\S*\s+)+", "", seg)
+                # A leading VAR=value (quoted or not) is not part of the command. When the
+                # workflow sets VAR, the value must be one it gives VAR, so a documented
+                # setting, such as the wasm runner and its flags, cannot drift from CI's.
+                while m := re.match(r"([A-Z_][A-Z0-9_]*)=('[^']*'|\"[^\"]*\"|\S*)\s+", seg):
+                    var, val = m.group(1), " ".join(m.group(2).strip("'\"").split())
+                    sets = workflow_values(raw, var)
+                    if sets and val not in sets:
+                        report.fail(block.where, f"{var}={val}: the workflow {detail} sets it to " + " or ".join(f"`{s}`" for s in sorted(sets)))
+                    seg = seg[m.end():]
                 if not seg or seg.startswith("cd "):
                     continue
                 # A placeholder (PATH/TO/PKG, FILE, $VAR, <x>) stands for any one word.
diff --git a/wai/wai-quantum/README.md b/wai/wai-quantum/README.md
index 46cbd376a..542c44253 100644
--- a/wai/wai-quantum/README.md
+++ b/wai/wai-quantum/README.md
@@ -46,7 +46,7 @@ fails the suite instead of quietly returning a different number
 (`.github/workflows/wai-conformance.yml`, its `quantum-wasip2` job):
 
 ```sh norun=ci:wai-conformance.yml
-CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime cargo test --locked --target wasm32-wasip2 --features full --test bit_identity
+CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' cargo test --locked --target wasm32-wasip2 --features full --test bit_identity
 ```
 
 
diff --git a/wai/wai-quantum/README.zh-CN.md b/wai/wai-quantum/README.zh-CN.md
index 6412c92d7..72a2b3b5c 100644
--- a/wai/wai-quantum/README.zh-CN.md
+++ b/wai/wai-quantum/README.zh-CN.md
@@ -30,7 +30,7 @@ let h: [u8; 32] = sv.statevector_hash();
 **跨架构可复现性是经过测试的，而不是仅凭声称。** CI 在 wasmtime 上的 `wasm32-wasip2` 下以及原生环境中，运行固定到精确 IEEE-754 位模式的浮点结果（熵、Holevo 界）以及线路核心的单元测试；因此，一个计算结果不同的平台会让测试套件失败，而不是悄悄返回另一个数字（`.github/workflows/wai-conformance.yml` 中的 `quantum-wasip2` 作业）：
 
 ```sh norun=ci:wai-conformance.yml
-CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime cargo test --locked --target wasm32-wasip2 --features full --test bit_identity
+CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' cargo test --locked --target wasm32-wasip2 --features full --test bit_identity
 ```
 
 
diff --git a/wai/wai-quantum/tests/bit_identity.rs b/wai/wai-quantum/tests/bit_identity.rs
index c295de9bc..fee481b64 100644
--- a/wai/wai-quantum/tests/bit_identity.rs
+++ b/wai/wai-quantum/tests/bit_identity.rs
@@ -4,7 +4,7 @@
 //! natively and confirmed under `wasm32-wasip2`; any platform whose math
 //! differs fails here, not in a user's diff.
 //!
-//! `CARGO_TARGET_WASM32_WASIP2_RUNNER=wasmtime cargo test --release --target wasm32-wasip2 --features full`
+//! `CARGO_TARGET_WASM32_WASIP2_RUNNER='wasmtime run -W nan-canonicalization=y -W relaxed-simd-deterministic=y' cargo test --release --target wasm32-wasip2 --features full`
 #![cfg(all(
     feature = "quantum",
     feature = "quantum_vml",
-- 
2.54.0 (Apple Git-157)


From d7ab19292f768b429297f47ebc9d42221af1e1b6 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:58:54 -0500
Subject: [PATCH 11/29] byte-equality: the disk check writes as well as
 measures, and covers the conformance matrix too

The disk pre-check read df under the leg's four directories (the workspace,
the runner's temporary directory, TMPDIR and HOME). Three gaps remained.

1. On linux-aarch64 df reads the runner VM's disk, not the host's. Run in a
   job container started as the runner starts one (its image and its two
   cache mounts), the step's loop put all four directories on the VM's data
   disk (84 GiB free), while the host volume that holds the disk image had
   50 GiB. The image is a 200 GiB file that was already allocated on the host
   but for 0.7 GiB, so a full host volume could take back at most that much.
   The step now also writes and syncs 32 MiB once on each filesystem it
   measures (dd conv=fsync, which the macOS host's dd and the job image's
   both take), and names a write refused for space, an I/O error, a read-only
   filesystem or a quota infrastructure. That also catches a disk out of
   inodes, which df's block count does not show. Another failed write is a
   warning, not a verdict.
2. The second determinism matrix, wai-conformance.yml's byte-exact job, runs
   on the same three labels with no disk check. It now has both steps, with
   the same floor.
3. The floor had no reading behind it. The step's loop, run on 2026-11-04 in
   each label's job environment: linux-x86_64 201 GiB (one host) and 304 GiB
   (the other); linux-aarch64 84 GiB; macos 78 GiB under the workspace and
   the runner's temporary directory, 85 GiB under TMPDIR and HOME. The floor
   stays at 4 GiB: it fails a leg only on a host within a few GiB of full,
   and the macOS runner's internal disk, where HOME and TMPDIR sit, has gone
   from 7 GiB free to 118 MiB within hours, with its jobs failing under HOME.
   The env comment says so.

The class of the failure is still not in the status context: the forge
reports `wai cross-architecture byte-equality / <label>` as failed for a full
disk as for a divergence. The comment now says so, and why the check is not a
job of its own per label: the `linux` label has two hosts, so such a job
could measure another host's disk than the leg's.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-byte-equality.yml | 60 ++++++++++++++++++----
 .github/workflows/wai-conformance.yml   | 67 +++++++++++++++++++++++++
 wai/CHANGELOG.md                        |  9 ++++
 3 files changed, 127 insertions(+), 9 deletions(-)

diff --git a/.github/workflows/wai-byte-equality.yml b/.github/workflows/wai-byte-equality.yml
index deab15f3f..5cb0a6ab3 100644
--- a/.github/workflows/wai-byte-equality.yml
+++ b/.github/workflows/wai-byte-equality.yml
@@ -173,7 +173,12 @@ jobs:
       # The floor of the disk check below, in GiB. The leg writes under 1 GiB (a
       # shallow checkout of about 0.3 GiB plus its pack, and a release target under
       # 0.1 GiB, measured 2026-11-03), and a cold toolchain install about 1 GiB more
-      # under HOME; the rest is room for the jobs that share the disk.
+      # under HOME; the rest is room for the jobs that share the disk. Measured with the
+      # step's own loop on 2026-11-04, every label had 50 GiB or more free under all
+      # four directories, so the floor fails a leg only on a host within a few GiB of
+      # full. The macOS runner's HOME and TMPDIR sit on its internal disk, which has
+      # gone from 7 GiB free to 118 MiB within hours, and at that level its jobs failed
+      # writing under HOME.
       DISK_FLOOR_GIB: '4'
     steps:
       # Infrastructure first, before the checkout writes anything. Main's run 17676
@@ -182,23 +187,48 @@ jobs:
       # full and no fixture was decoded, yet the matrix read as a byte-equality failure.
       # This step measures the free space under every directory the leg writes to (the
       # workspace, the runner's temporary directory, TMPDIR, and HOME, where the
-      # toolchain lives). Below the floor it fails, and its name and its error say
-      # infrastructure, so a full disk is not read as a divergence.
+      # toolchain lives), and writes and syncs 32 MiB once on each filesystem, which a
+      # disk out of inodes or remounted read-only fails though df shows room. Below the
+      # floor, or on a write refused for space, an I/O error or a read-only filesystem,
+      # it fails, and its name and its error say infrastructure.
+      #
+      # On linux-aarch64 the four directories are on the runner VM's data disk, so df
+      # reads the guest's free space, not the host volume's. The disk is an image file
+      # on that volume, and on 2026-11-04 all but 0.7 of its 200 GiB were already
+      # allocated there, so a full host volume could take back at most that much; the
+      # write probe catches the rest.
+      #
+      # The forge still reports the leg's own status context as failed (`wai
+      # cross-architecture byte-equality / <label>`), as for a divergence; the class is
+      # in the failed step's name and its error. A job of its own per label would put it
+      # in the context, but the `linux` label has two hosts, so that job could measure
+      # another host's disk than the leg's.
       - name: "infrastructure: this runner's disk has room for the leg"
         id: disk
         if: matrix.os != 'windows'
         shell: bash
         run: |
           set -euo pipefail
-          short=0
+          short=0; seen=""
           for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
             p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
             set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
             echo "$d: $(($1 / 1048576)) GiB free on $2"
             if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
               echo "::error::infrastructure: ${{ matrix.label }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The runner host's disk is full: no fixture was decoded, so this is not a byte-equality failure."
-              short=1
+              short=1; continue
             fi
+            case " $seen " in *" $2 "*) continue ;; esac
+            seen="$seen $2"; f="$p/.wai-disk-probe.$$"
+            if ! err=$(dd if=/dev/zero of="$f" bs=1048576 count=32 conv=fsync 2>&1); then
+              case "$err" in
+                *"No space left"*|*"Input/output error"*|*"Read-only file system"*|*"quota"*)
+                  echo "::error::infrastructure: ${{ matrix.label }}: a 32 MiB write under $p, on $2, failed: $(printf '%s' "$err" | head -n 1). The runner host's disk cannot take the leg's writes: no fixture was decoded, so this is not a byte-equality failure."
+                  short=1 ;;
+                *) echo "::warning::a 32 MiB write under $p failed for another reason: $(printf '%s' "$err" | head -n 1)" ;;
+              esac
+            fi
+            rm -f "$f"
           done
           exit $short
 
@@ -231,22 +261,34 @@ jobs:
 
       # A disk can also fill while the leg runs, from the jobs that share it (three
       # conformance jobs at once filled the arm64 VM's data disk mid-link). When a step
-      # after the check failed, the same measurement is taken again: below the floor,
-      # the failure is named infrastructure; with room, it stands as it is.
+      # after the check failed, the same measurement and write are made again: below the
+      # floor, or on a write refused for space, the failure is named infrastructure; with
+      # room, it stands as it is.
       - name: "infrastructure: was the disk full when the leg failed?"
         if: failure() && steps.disk.outcome == 'success'
         shell: bash
         run: |
           set -euo pipefail
-          short=0
+          short=0; seen=""
           for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
             p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
             set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
             echo "$d: $(($1 / 1048576)) GiB free on $2"
             if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
               echo "::error::infrastructure: ${{ matrix.label }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
-              short=1
+              short=1; continue
             fi
+            case " $seen " in *" $2 "*) continue ;; esac
+            seen="$seen $2"; f="$p/.wai-disk-probe.$$"
+            if ! err=$(dd if=/dev/zero of="$f" bs=1048576 count=32 conv=fsync 2>&1); then
+              case "$err" in
+                *"No space left"*|*"Input/output error"*|*"Read-only file system"*|*"quota"*)
+                  echo "::error::infrastructure: ${{ matrix.label }}: a 32 MiB write under $p, on $2, failed: $(printf '%s' "$err" | head -n 1). The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
+                  short=1 ;;
+                *) echo "::warning::a 32 MiB write under $p failed for another reason: $(printf '%s' "$err" | head -n 1)" ;;
+              esac
+            fi
+            rm -f "$f"
           done
           [ "$short" = 1 ] || echo "Every disk this leg writes to has room: the failure above is not the disk."
           exit $short
diff --git a/.github/workflows/wai-conformance.yml b/.github/workflows/wai-conformance.yml
index db7fe7c5e..6e657c777 100644
--- a/.github/workflows/wai-conformance.yml
+++ b/.github/workflows/wai-conformance.yml
@@ -2919,8 +2919,46 @@ jobs:
         # would add no arch coverage anyway.
         # Windows x86_64 leg removed 2026-09-30: runner offline. Restore by
         # re-adding `windows` here once the runner is back online.
+        # The two disk steps below are bash, and skip on Windows.
         os: [linux, linux-arm64, macos]
+    env:
+      # The floor of the disk check below, in GiB, as in wai-byte-equality.yml, whose
+      # comment gives the leg's writes and the readings behind it.
+      DISK_FLOOR_GIB: '4'
     steps:
+      # The two disk steps of wai-byte-equality.yml's legs, on the same three labels:
+      # a full runner disk fails here as infrastructure, before the checkout, and a
+      # failure later in the leg is named infrastructure when the disk filled meanwhile.
+      # wai-byte-equality.yml's comment says what they measure and why.
+      - name: "infrastructure: this runner's disk has room for the leg"
+        id: disk
+        if: matrix.os != 'windows'
+        shell: bash
+        run: |
+          set -euo pipefail
+          short=0; seen=""
+          for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
+            p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
+            set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
+            echo "$d: $(($1 / 1048576)) GiB free on $2"
+            if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
+              echo "::error::infrastructure: ${{ matrix.os }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The runner host's disk is full: no fixture was decoded, so this is not a byte-equality failure."
+              short=1; continue
+            fi
+            case " $seen " in *" $2 "*) continue ;; esac
+            seen="$seen $2"; f="$p/.wai-disk-probe.$$"
+            if ! err=$(dd if=/dev/zero of="$f" bs=1048576 count=32 conv=fsync 2>&1); then
+              case "$err" in
+                *"No space left"*|*"Input/output error"*|*"Read-only file system"*|*"quota"*)
+                  echo "::error::infrastructure: ${{ matrix.os }}: a 32 MiB write under $p, on $2, failed: $(printf '%s' "$err" | head -n 1). The runner host's disk cannot take the leg's writes: no fixture was decoded, so this is not a byte-equality failure."
+                  short=1 ;;
+                *) echo "::warning::a 32 MiB write under $p failed for another reason: $(printf '%s' "$err" | head -n 1)" ;;
+              esac
+            fi
+            rm -f "$f"
+          done
+          exit $short
+
       - uses: actions/checkout@v5
 
       - name: Install Rust 1.98.0 (unix)
@@ -2944,6 +2982,35 @@ jobs:
         working-directory: wai/byte-exact-conformance
         run: cargo test --release
 
+      - name: "infrastructure: was the disk full when the leg failed?"
+        if: failure() && steps.disk.outcome == 'success'
+        shell: bash
+        run: |
+          set -euo pipefail
+          short=0; seen=""
+          for d in "${GITHUB_WORKSPACE:-$PWD}" "${RUNNER_TEMP:-/tmp}" "${TMPDIR:-/tmp}" "$HOME"; do
+            p=$d; while [ ! -e "$p" ]; do p=$(dirname "$p"); done
+            set -- $(df -Pk "$p" | awk 'NR == 2 {print $4, $6}')
+            echo "$d: $(($1 / 1048576)) GiB free on $2"
+            if [ "$1" -lt $((DISK_FLOOR_GIB * 1048576)) ]; then
+              echo "::error::infrastructure: ${{ matrix.os }}: $d is on $2, which has $(($1 / 1024)) MiB free, below the floor of $DISK_FLOOR_GIB GiB. The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
+              short=1; continue
+            fi
+            case " $seen " in *" $2 "*) continue ;; esac
+            seen="$seen $2"; f="$p/.wai-disk-probe.$$"
+            if ! err=$(dd if=/dev/zero of="$f" bs=1048576 count=32 conv=fsync 2>&1); then
+              case "$err" in
+                *"No space left"*|*"Input/output error"*|*"Read-only file system"*|*"quota"*)
+                  echo "::error::infrastructure: ${{ matrix.os }}: a 32 MiB write under $p, on $2, failed: $(printf '%s' "$err" | head -n 1). The disk filled while the leg ran: read the failure above as the runner host's disk, not as a byte-equality failure."
+                  short=1 ;;
+                *) echo "::warning::a 32 MiB write under $p failed for another reason: $(printf '%s' "$err" | head -n 1)" ;;
+              esac
+            fi
+            rm -f "$f"
+          done
+          [ "$short" = 1 ] || echo "Every disk this leg writes to has room: the failure above is not the disk."
+          exit $short
+
   # The codec wrappers are behind `codecs`, which binds C and C++ libraries that no
   # container image of this fleet carries, so no other job compiles them. A feature no
   # workflow enables has no CI coverage at all: without this job the wrappers' tests,
diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index aaf219337..79b3e108a 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -6,6 +6,15 @@ wire format may change and such changes are recorded here.
 
 ## Unreleased
 
+### CI: a full runner disk fails a byte-equality leg as infrastructure, not as a divergence
+
+- **CI:** each leg of the two cross-architecture byte-equality matrices (`wai-byte-equality.yml`, and
+  `wai-conformance.yml`'s `byte-exact`) measures the free space under its workspace, the runner's temporary directory,
+  TMPDIR and HOME before the checkout, and writes and syncs 32 MiB on each filesystem. Below 4 GiB, or on a write
+  refused for space, an I/O error or a read-only filesystem, the leg fails in a step named for infrastructure, with an
+  error that says no fixture was decoded. When a later step fails, the measurement and the write are made again, and
+  the failure is named infrastructure if the disk filled meanwhile.
+
 ### CI: WebAssembly runs under one pinned wasmtime with both its determinism settings, and the browser packages hold no relaxed SIMD
 
 - **CI:** every workflow that runs a module under wasmtime runs wasmtime 50.0.1, pinned by version and SHA-256, with
-- 
2.54.0 (Apple Git-157)


From 9eb12b9cce9d3e70d3ca03581c1992a7b5352de4 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:41:26 -0500
Subject: [PATCH 12/29] =?UTF-8?q?determinism-tiers=20=C2=A71=20and=20?=
 =?UTF-8?q?=C2=A77.1:=20the=20published=20routes=20as=20their=20bodies=20s?=
 =?UTF-8?q?tate=20them,=20and=20a=20catalogue-prior=20record?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

§1 now records five published routes to a reproducible stage, each from the
paper section read. The explicit-parameter route keeps its entropy stage and
lets the reconstruction drift: its body says so, and its periodic I-frames
re-anchor the picture to one float decode (§3.2, Appendix B.8). Two integer
routes join the quantised decoder: a float model run in a 16-bit integer mode,
whose accumulators were observed not to overflow (§4.5), and an integer-trained
codec (abstract only). One more line reaches no tier: float synchronised on a
tested hardware class, whose streams decode only within the GPU groups its
tests found (Table IV, §III-B). §7.1 cites the two records of float decode
across devices beside WAI's record, and is not a comparison.

syfai-input §5 and the September sweep's row of arXiv 2608.00483, G7 and F5
are corrected against the bodies read on 2026-11-03.

research/catalogue-prior.md records a linear basis learned on openly
licensed films of one studio, measured against AV1: its readings, procedure,
decision rule and the time the rule was fixed. The rule refutes a linear or
spectral catalogue basis and exact recurrence across titles for a studio's
catalogue; a learned prior is not measured. tools/wai_catalogue_prior_verify.py
recomputes every BD-rate and verdict from the readings and checks every figure
the page quotes, in the conformance workflow.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-conformance.yml     |   16 +
 wai/CHANGELOG.md                          |   12 +
 wai/extensions/determinism-tiers.md       |   53 +-
 wai/research/2026-09-global-sweep.md      |    7 +-
 wai/research/catalogue-prior.md           |  326 ++++
 wai/research/catalogue-prior/a_basis.json | 1770 +++++++++++++++++++++
 wai/research/catalogue-prior/b_recur.json |  770 +++++++++
 wai/research/catalogue-prior/c_rd.json    | 1478 +++++++++++++++++
 wai/research/catalogue-prior/d_exact.json |  122 ++
 wai/research/catalogue-prior/run.json     |   91 ++
 wai/research/syfai-input.md               |   12 +-
 wai/tools/wai_catalogue_prior_verify.py   |  286 ++++
 12 files changed, 4937 insertions(+), 6 deletions(-)
 create mode 100644 wai/research/catalogue-prior.md
 create mode 100644 wai/research/catalogue-prior/a_basis.json
 create mode 100644 wai/research/catalogue-prior/b_recur.json
 create mode 100644 wai/research/catalogue-prior/c_rd.json
 create mode 100644 wai/research/catalogue-prior/d_exact.json
 create mode 100644 wai/research/catalogue-prior/run.json
 create mode 100644 wai/tools/wai_catalogue_prior_verify.py

diff --git a/.github/workflows/wai-conformance.yml b/.github/workflows/wai-conformance.yml
index 6e657c777..74a403f60 100644
--- a/.github/workflows/wai-conformance.yml
+++ b/.github/workflows/wai-conformance.yml
@@ -333,6 +333,11 @@ on:
       - 'wai/research/determinism-cost/**'
       - 'wai/tools/wai_determinism_cost_verify.py'
       - 'wai/extensions/determinism-tiers.md'
+      # The catalogue-prior record, its verifier, and the page whose quoted figures the
+      # verifier checks against the readings.
+      - 'wai/research/catalogue-prior.md'
+      - 'wai/research/catalogue-prior/**'
+      - 'wai/tools/wai_catalogue_prior_verify.py'
       # tests/stream_termination.rs requires this extension's §7 to list the codes the reference
       # reports and capability-constraints §4.2 to name the match's bits,
       # stream_termination_verify.py reads both, and the registry's wire identifiers test reads
@@ -1334,6 +1339,17 @@ jobs:
           python3 -I tools/wai_determinism_cost_verify.py research/determinism-cost/record.json \
             research/determinism-cost.md extensions/determinism-tiers.md
           (cd wai-rs && cargo test --no-default-features --features "$WAI_FEATURES" --test det_cost_bin)
+      # The catalogue-prior record (research/catalogue-prior.md): each reading checked against
+      # the SHA-256 its run.json states; every BD-rate the run stored recomputed from the
+      # rate and quality points, and the rate proxy, the recurrence and the decision rule's
+      # verdicts from the readings, by a second implementation that shares no code with the
+      # scripts that took them (tools/wai_catalogue_prior_verify.py, standard library only,
+      # run isolated); and every figure the page quotes checked (each is followed by a `cat`
+      # marker naming its value).
+      - name: catalogue prior · the record and the text quoting it
+        working-directory: wai
+        shell: bash
+        run: python3 -I tools/wai_catalogue_prior_verify.py research/catalogue-prior research/catalogue-prior.md
       # The staged-delivery corpus's second implementation (staged-conformance/README): it
       # replays every case from its stage-0 buffers through its own reading of the
       # extension (refines, WIR1, the sink's steps, holds, batches, conflicts, deadlines,
diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index 79b3e108a..fcf8d63cc 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -6,6 +6,18 @@ wire format may change and such changes are recorded here.
 
 ## Unreleased
 
+### A decode reports the path that ran, and no tier lies between the two
+
+- **determinism-tiers §1 and §7.1** record the five published routes to a reproducible stage, from the paper sections
+  read, and a line that reaches no tier: float synchronised on a tested hardware class (arXiv 2608.00483). The research
+  notes `syfai-input.md` §5 and `2026-09-global-sweep.md` (its row of arXiv 2608.00483, G7 and F5) are corrected against
+  the papers' bodies.
+- **A catalogue-prior record** ([research/catalogue-prior.md](research/catalogue-prior.md)): a linear basis learned on
+  openly licensed films of one studio, measured against AV1, with its readings, its procedure, its decision rule
+  and when that rule was fixed. The rule refutes a linear or spectral catalogue basis and exact recurrence across titles
+  for a studio's catalogue; a learned prior is not measured. `tools/wai_catalogue_prior_verify.py` recomputes every
+  BD-rate and verdict from the readings and checks every figure the page quotes, in the conformance workflow.
+
 ### CI: a full runner disk fails a byte-equality leg as infrastructure, not as a divergence
 
 - **CI:** each leg of the two cross-architecture byte-equality matrices (`wai-byte-equality.yml`, and
diff --git a/wai/extensions/determinism-tiers.md b/wai/extensions/determinism-tiers.md
index 778f75701..a6bf9d18f 100644
--- a/wai/extensions/determinism-tiers.md
+++ b/wai/extensions/determinism-tiers.md
@@ -28,7 +28,7 @@ from one diverging bit, select a different distribution from the encoder's and
 corrupt the entropy decode. A float stage whose operations, order, precision and
 environment are all fixed does reproduce, within the implementation that fixes
 them; §5 describes one, and §2–§3 say why that is not a tier and does not back
-`replicate`. Three published routes to reproducibility are below. Two make the
+`replicate`. Five published routes to reproducibility are below. Two make the
 **entropy stage** reproducible while leaving the reconstruction to pixels to the
 implementation, within whatever criterion a conformance specification sets on it:
 
@@ -54,14 +54,52 @@ implementation, within whatever criterion a conformance specification sets on it
   explicitly, through the hyperprior, so the decoder reads them instead of
   deriving them and the entropy decode is consistent across devices *without*
   bit-exact arithmetic. The side information costs bitrate. Same target: a
-  reproducible entropy stage, a reconstruction that may vary.
+  reproducible entropy stage, a reconstruction that may vary. The paper's body
+  states the second half: "floating-point differences still cause gradual
+  divergence in the features and reconstructed frames", and "without a
+  synchronization mechanism, the divergence can grow arbitrarily large" (§3.2).
+  It bounds the divergence with periodic I-frames that "fully resynchronize
+  encoder and decoder feature states" (§3.2). Those I-frames are decoded by the
+  same float model (Appendix B.8), so a resynchronisation moves the picture back
+  to one float decode; it does not make the picture reproducible.
 
-The third makes the reconstruction itself reproducible:
+The other three make the reconstruction itself reproducible, each by decoding in
+integer arithmetic:
 
 - **A quantised decoder.** A published learned image codec (arXiv 2312.11209)
   quantises the weights and activations of all its decoder subnetworks, without
   accumulator overflow, so that the reconstructed image is deterministic across
   devices, and reports its output as bit-exact.
+- **Integerisation of a float model.** A published learned video codec
+  (arXiv 2502.20762) runs its float model in a 16-bit integer mode without
+  retraining: each value is mapped to round(512 · v), and its accumulators are
+  32-bit, of which it reports that "no overflow issues have been observed"
+  (§4.5). That is an observation, not a bound. WAI's integer networks state their
+  bounds before they decode ([integer-operators](integer-operators.md) §8).
+- **Integer training.** A published learned video codec is trained as an
+  integer-only model from the start, and claims coding consistency across
+  platforms from integer decoding in the encoder's reconstruction loop
+  (OpenReview `KCQo0fXtFH`, abstract). Its body and its code were not read here.
+
+One more published line reaches neither tier: **float synchronised on a tested
+hardware class.** A streamable learned video codec (arXiv 2608.00483) decodes its
+P-frames in 16-bit float, and its I-frames and its periodic resets of a feature
+adapter in 32-bit float (§II-C, §II-D). It was tested on twelve GPUs of one
+vendor under one driver (Table II). In the mixed mode, a stream decoded only on
+GPUs of the generation that encoded it, and at 1080p not on every one of them; in
+pure 32-bit float, streams decoded across generations at 1080p and split into
+three groups at 4K (Table IV). A
+stream decoded outside its group fell from 26.3 dB to between 11.9 and 13.7 dB of
+luma PSNR at 1080p, and from 28.2 dB to between 7.3 and 10.6 dB at 4K, or the
+decoder did not complete (§II-F-3, Table IV). The authors give the cause SPEC §7
+gives (*Replicate-soundness*): pure 32-bit float "is strictly non-deterministic,
+with each GPU architecture generating a unique bitstream", because kernels
+"whose differing reduction orders break bit-exactness even under IEEE-754 FP32,
+which fixes per-operation rounding but not operation ordering" were chosen per
+generation (§III-B). Exactness found on the devices tested is a property of those
+devices and not of a capability (§2), and the frames between two resets have no
+portable hash, so this line is no tier of this document (§3, *No tier between the
+two*).
 
 WAI names the two outcomes as tiers, so a capability states which one it makes, an
 envelope's `intent` can demand the one it needs (`replicate` over a neural
@@ -937,6 +975,15 @@ record, recomputes each from the readings and checks it, in CI.
 Each figure holds for the pictures, models, platforms and builds the record names, and for
 nothing else (§7.2).
 
+Two published records of float decode across devices are cited in §1. Neither is part of
+this record. arXiv 2608.00483 found its streams to decode only within the groups of GPUs
+its tests established (Table IV). arXiv 2606.28027 found its float reconstruction to diverge
+across devices unless it is resynchronised (§3.2), and its 8-bit integer decode not to be
+bit-exact across neural processors, for "non-standard implementations of rounding and
+multiplication in quantize/dequantize and re-quantization" (Supplement D.1). The
+requantisation of WAI's integer networks states its rounding
+([integer-operators](integer-operators.md) §5.1).
+
 What the record does not cover is listed in its §1: `wai.image.jpegai`'s exact decoder
 (one configuration, so one rate point and no curve; and the reference set was not coded
 with the standard's reference encoder), `wai.neural.int_synth` and `wai.neural.int_mlicpp`
diff --git a/wai/research/2026-09-global-sweep.md b/wai/research/2026-09-global-sweep.md
index 8c6a8daac..f04cde830 100644
--- a/wai/research/2026-09-global-sweep.md
+++ b/wai/research/2026-09-global-sweep.md
@@ -277,7 +277,7 @@ work.
 | ARIB STD-B60 (MMT) | v2.0 (2025-03-25), the main deployed MMT profile | **R** ⚠ |
 | NHK STRL Open House 2026 | C2PA provenance across capture, edit and distribution, with hardened in-browser verification. Cloud-unified broadcast/online production on ST 2110 + CMAF low-latency (**no MoQ or QUIC**). Live 8K 360-degree coding on VVC | **R** ⚠ |
 | a learned image coding line with quantised arithmetic since 2020 | arXiv 2007.04684 (*fixed point weight quantization*), 2111.09348 (*quantized weights and activations*). arXiv metadata carries no affiliation, so the institution the sweep attached to this line is **not verified** | P✓ (titles) / ⚠ (affiliation) |
-| arXiv 2608.00483 (2026-08; GLOBECOM 2026), by authors overlapping the line above | mixed-precision cross-platform neural video coding. P-frames run in FP16. I-frames and periodic feature-adapter resets run in IEEE-754 FP32, keeping the arithmetic entropy coder synchronised across GPUs. Evaluated across 12 GPUs of four architecture generations. Claims "deterministic synchronization at critical boundaries" and cross-generation decodability "for recent architectures". **The abstract makes no bit-exactness or identical-reconstruction claim** | P✓ (abstract) |
+| arXiv 2608.00483 (2026-08; GLOBECOM 2026), by authors overlapping the line above | mixed-precision cross-platform neural video coding. P-frames run in FP16. I-frames and periodic feature-adapter resets run in IEEE-754 FP32, keeping the arithmetic entropy coder synchronised across GPUs. Evaluated across 12 GPUs of four architecture generations. Claims "deterministic synchronization at critical boundaries" and cross-generation decodability "for recent architectures". **The abstract makes no bit-exactness or identical-reconstruction claim.** *Corrected 2026-11-03, body read:* that holds of the abstract only. The body claims bit-exact bitstream hashes within a GPU generation and identical decoded output wherever a decode succeeds (§III-B), and "bit-exact synchronization at critical GOP boundaries" (§IV). The 12 GPUs are of one vendor, under one driver (Table II). In the mixed mode a stream never decodes across a generation; the cross-generation decodability at 1080p belongs to the pure-FP32 mode, which splits into three groups at 4K (Table IV) | P✓ (abstract; body 2026-11-03) |
 | METI Top Runner, TV target FY2026 | per-device annual kWh | **S** ⚠ |
 
 The 2030 landscape (§7.3) rejected a clause-level claim about a Japanese
@@ -690,7 +690,7 @@ review.
 | G4 | **WAI's `HwShunt` takes in IOReport, a reading the monorepo classes five ways; and neither WAI nor the family's receipt ladder separates platform-relayed counters from independently rooted instruments. Only `sandbox-tee` does.** Three layers. **(a) The family's receipt ladder.** Every standard that lists interfaces under `HwShunt` puts on-die counters and dedicated shunts in it together (JCR-1 §2.3, JCP §5.1, Sandbox §6.3, `mesh-schema`, `proof-schema`), and IOReport is classed inconsistently (§5.3). **(b) `sandbox-tee`'s `CounterClass`** reserves `HwShunt` for a dedicated shunt and gives RAPL, NVML, IOReport and Redfish classes of their own; its `AttestTrust` ladder is derived from evidence, and `efficiency/1` commits to the attestation (§5.1). **(c) WAI** has `HwShunt` cover "RAPL, NVML, SMC, IOReport, or an external shunt", and `wai_video_meter` stamps IOReport readings `HwShunt`. That places WAI with Joule Code's text and the Sandbox meter, and against JCR-1 §2.3. WAI's texts present the ladder as the family's own without noting the split: `video_receipt.rs` says it is "reused verbatim from `mesh-schema` / `proof-schema`", which list no IOReport; `energy_binding.rs` calls it "the family's ladder"; `energy-measurement.md` §2 says WAI reuses "the same three names the receipt families elsewhere in the ecosystem already use". At every layer but (b), a verifier cannot tell a platform-relayed counter, which may be deliberately filtered (§5.1), from an independently rooted instrument. | §5.3 (the per-standard survey); `wai/extensions/energy-measurement.md` §2; `wai-rs/src/energy_binding.rs`:30, :47–51 (`Acquisition`); `wai-rs/src/video_receipt.rs` `Provenance` doc comment; `wai-rs/src/bin/wai_video_meter.rs`:184–190; `jouleclaw/jouleclaw-rs/crates/jouleclaw-jcr1/docs/JCR-1-spec.md` §2.3; `jcp/SPEC.md` §5.1; `sandbox/SPEC.md` §6.3–§6.4; `mesh/mesh-rs/crates/mesh-schema/src/lib.rs` and `proof/proof-rs/crates/proof-schema/src/lib.rs` (`Provenance`); `sandbox/sandbox-rs/crates/sandbox-tee/src/energy.rs` (`AttestTrust`, `CounterClass`, `is_independently_rooted`); `jcp/jcp-rs/crates/jcp-efficiency/MAPPING.md` §4 and `src/training.rs` (`energy_attestation`). No mention of `sandbox-tee`, `CounterClass`, `AttestTrust`, `energy_attestation`, `PhysicallyRooted` or `IntegrityAttested` under `wai/` (grep) | **WAI's side closed at link 08 (PR #109); the family's side open.** energy-measurement §2 separates `CalibratedInstrument` (an external instrument with a current calibration) from `OnChipCounter` (a counter the measured silicon keeps, or any instrument whose uncertainty is not traceable). Each carries a declared uncertainty (§2.3); the counter carries its filtering state (§2.4), the instrument a calibration reference (§2.5). `HwShunt` is legacy, accepted and never emitted (§2.6), and IOReport is `OnChipCounter` (§2.9). `energy_class.rs` is the one definition, and the reference meters seal `OnChipCounter`. The "reused verbatim" sentence is gone; `video_receipt.rs` and `energy_binding.rs` still call the ladder the family's, and §2 says the three-name vocabulary is extended, not replaced, with both refinements mapping to `HwShunt`. §2.9's list of differing documents is partial (§5.3). No WAI text references `sandbox-tee`'s ladder (grep). The family's IOReport split and whether its receipt-level ladder should separate the two kinds of hardware evidence are unassigned (below) |
 | G5 | **The codec registry has no AV2, VVC, HEVC, LCEVC, MPEG-H Audio, AVS3 or Audio Vivid capability.** Brazil mandates three of these (§3.3). Two are Chinese standards: AVS3 (a group standard, with its national standard pending) and Audio Vivid (GB/T 46271-2025). 2030 §6.1 (*Honest scope*) records that WAI registers no VSEI-family codec. **The tree's codec survey is a second surface to reconcile.** `wai/glossary/codecs.json`, which `wai/README.md` describes as the "Reference catalog of media codecs surveyed for capability registration", has entries for HEVC / H.265 and VVC / H.266 (both `niche`), LCEVC (`niche`) and ECM (`academic`, JVET's exploration model beyond VVC), and none for AV2, MPEG-H Audio, AVS3 or Audio Vivid. `glossary/SCHEMA.md` defines `success` as "carries non-trivial production traffic today" and `niche` as "alive in specific contexts". §4 reports HEVC in production use at 65% in a vendor survey (**S** ⚠), §3.3 reports VVC as a national broadcast system's main codec (**R** ⚠), and the HEVC entry's own description calls it universal in broadcast UHD. The grades are descriptive and the contrary evidence is unverified, so this is a reconciliation item, not a correction. | capability strings in `wai-rs/src/codecs/*.rs`: none for these codecs. Other mentions, from a grep over `*.rs`, `*.md`, `*.json`, `*.toml`, `*.yml`, `*.yaml`, `*.html`, `*.ts` and `*.js` under `wai/`: the `glossary/codecs.json` entries above; the H.264/H.265 royalty sentence in `SPEC.md` and `CHANGELOG.md`; the BD-rate note that names VVC in `wai-rs/src/codecs/mod.rs` and `wai-rs/src/neural/mlicpp.rs`; `HEVC` as an example name in `glossary/SCHEMA.md`; the 2030 landscape. Royalty class per codec **not established by this sweep**: only AV2's royalty-free patent policy was seen at the issuing body (the AOMedia press release, §2.4) | **Closed in the registry at link 04 (PR #105); the glossary is not reconciled.** SPEC §5 *Broadcast and streaming codecs* registers `wai.video.{av2,vvc,hevc,avc,vp9,evc,evc.baseline,avs3,lcevc}` and `wai.audio.{mpegh,avs3,aac}`, each with one payload framing and a licence class whose basis is recorded in [`media-codec-registry.md`](media-codec-registry.md). `wai.audio.avs3` names AVS3 audio / Audio Vivid (GB/T 46271-2025). All are sink-supplied, none is a default, and `wai.video.lcevc` is companion-only. SPEC §7 classes AVC, HEVC, VVC and VP9 `StandardDefined` on the conformance clauses read, AV2 `StandardTolerance`, and EVC, AVS3, LCEVC, MPEG-H 3D Audio and AAC `StandardUnclassified`. The licence classes rest on the registry's recorded bases, not on this sweep. `glossary/codecs.json` is unchanged, so its grades still stand against §3.3 and §4 |
 | G6 | **No WAI receipt says who can check the energy figure, and the base JWP receipt does not carry even the acquisition class that `energy-measurement.md` requires.** The precedents in §5.1 differ exactly on who can check: instrument-signed (OCMF), peer-reviewed but unsigned, log-registered (SCITT), TEE-attested with a derived trust level (the family's `sandbox-tee` `AttestTrust`), or producer-asserted. **The base JWP receipt** has no acquisition-class field; its group signature covers `total_joules_micro` and no class, and the per-object `joules_micro` is in neither the Merkle leaf nor the signature. `energy-measurement.md` §2 requires a class inside the signature for any non-zero figure, while its header says it adds "no new wire format". **The video receipt** signs its class. **The energy binding** carries its class in a C2PA companion assertion signed as part of whatever manifest carries it. **The OER receipt** carries an evidential grade and a ledger identity, and no class. 2030 §8 item 4 queues typed receipt classes. | `extensions/energy-measurement.md` header, §1, §2 and §5; `extensions/jwp-receipts.md` §2.1–§2.4; `joule-code/joule-code-rs/crates/joule-moq/src/receipt.rs` (`ObjectReceipt`, `GroupReceipt::signing_payload`: no class); `wai-rs/src/energy_binding.rs`, `wai-rs/src/video_receipt.rs`; `extensions/oer-receipts.md` §2–§3, `wai-rs/src/oer.rs`; no reference to the family ladder under `wai/` (grep, as in G4) | **Partly closed at link 10 (PR #112); the rest open.** Every receipt within energy-measurement's scope that `wai-rs` and `wai-quantum` seal can carry the class (link 08, PR #109). jwp-receipts §2.4.3–§2.4.7 add a `claim` kind with typed classes. An `energy` claim (§2.4.5) restates a JWP object's figure beside its class, the class's evidence, a coverage statement and a grade; a verifier holding the object receipt rejects the pair if the figures differ, and one holding the OER receipt the claim names checks that grade and figure agree. `claim_receipt.rs` (feature `receipt_claims`) implements it, with vectors in `claim-conformance/`. Open, as energy-measurement §1 records: the JWP object receipt has no member of its own for the class; the group total has none, and nothing carries one for it; the `WAI_RECEIPT` property's class and evidence sit outside any signature; an OER/2 receipt has no member for the class. On who can check: a claim carries its signer's authority and no more (§2.4.4); key discovery and an authority class for who may sign which class are not done (CHANGELOG, v0.4.0 entry for typed receipt classes); no WAI receipt references `sandbox-tee`'s ladder or `efficiency/1`'s attestation (grep). *Status after the v0.4.0 change "Energy reports, labelled object leaves, and what the signatures cover":* the object's class rides in its labelled leaf, a group total can carry a signed class of its own, and a `WAI_RECEIPT` property's class counts only where a signed statement restates it; joule-moq verifies both labelled forms. *Status after the v0.4.0 change "Key discovery, signing authority and module revocation":* key discovery and an authority class for who may sign which class are jwp-receipts §7 (`wai-rs` `key_log`, feature `key_discovery`; `signer-conformance/`); a hardware-class energy authority is confirmed only under a pinned anchor, and a verifier reports whether the key also holds `delivery` |
-| G7 | **Explicit parameter transmission appears in `determinism-tiers.md` only as a generic route. Its first published cross-NPU instance (arXiv 2606.28027) and its cost are not recorded.** Also unrecorded: the mixed-precision entropy-synchronisation line (arXiv 2608.00483: FP16/FP32 across GPUs, no bit-exactness claim); the quantised learned-coding line (arXiv 2007.04684, 2111.09348); and IEEE 1857.11's use of *reproducibility cross-checks* rather than bit-exact decode (**R** ⚠), a reproducibility notion the tier vocabulary does not name. Both 2606.28027 and 2608.00483 target a synchronised entropy stage, so at most `entropy-consistency`, and 2608.00483 only over the GPUs it tested. On `main`, whether such a decode may back `replicate` depends on which of G10's conditions is read. | `extensions/determinism-tiers.md` §1–§3; `SPEC.md` §7 (read) | **Partly closed at link 02 (PR #103).** determinism-tiers §1 names arXiv 2606.28027 as the explicit-parameter instance and says the side information costs bitrate (no magnitude), and names a quantised decoder, arXiv 2312.11209, as the route to a reproducible reconstruction. Such a decode reaches `entropy-consistency` and does not back `replicate` (G10). Not in the tree at link 10 (grep): arXiv 2608.00483, the quantised line 2007.04684 / 2111.09348, and IEEE 1857.11's reproducibility cross-checks; the last is described in [`syfai-input.md`](syfai-input.md) §5, committed with this record |
+| G7 | **Explicit parameter transmission appears in `determinism-tiers.md` only as a generic route. Its first published cross-NPU instance (arXiv 2606.28027) and its cost are not recorded.** Also unrecorded: the mixed-precision entropy-synchronisation line (arXiv 2608.00483: FP16/FP32 across GPUs, no bit-exactness claim); the quantised learned-coding line (arXiv 2007.04684, 2111.09348); and IEEE 1857.11's use of *reproducibility cross-checks* rather than bit-exact decode (**R** ⚠), a reproducibility notion the tier vocabulary does not name. Both 2606.28027 and 2608.00483 target a synchronised entropy stage, so at most `entropy-consistency`, and 2608.00483 only over the GPUs it tested. On `main`, whether such a decode may back `replicate` depends on which of G10's conditions is read. *Corrected 2026-11-03, body read:* arXiv 2608.00483 makes no bit-exactness claim in its abstract, though its body claims bit-exact hashes within a GPU generation (§III-B; its row above). | `extensions/determinism-tiers.md` §1–§3; `SPEC.md` §7 (read) | **Partly closed at link 02 (PR #103).** determinism-tiers §1 names arXiv 2606.28027 as the explicit-parameter instance and says the side information costs bitrate (no magnitude), and names a quantised decoder, arXiv 2312.11209, as the route to a reproducible reconstruction. Such a decode reaches `entropy-consistency` and does not back `replicate` (G10). Not in the tree at link 10 (grep): arXiv 2608.00483, the quantised line 2007.04684 / 2111.09348, and IEEE 1857.11's reproducibility cross-checks; the last is described in [`syfai-input.md`](syfai-input.md) §5, committed with this record. *Since 2026-11-03:* determinism-tiers §1 records arXiv 2608.00483 as float synchronised on a tested hardware class, which reaches no tier (its Table IV and §III-B), and the integer routes of arXiv 2502.20762 (§4.5) and of an integer-trained codec (OpenReview `KCQo0fXtFH`, abstract); the quantised line 2007.04684 / 2111.09348 is still not in the tree |
 | G8 | **Standards inputs not tracked in the tree:** SyfAI; 3GPP TR 26.942; DVB S100 and CM-EE; the beyond-VVC CfP; Gaussian Splat Coding Exploration 45; IEEE 1857.11 / P1857.11.1; UWA 042; India D2M; the JPEG AI core experiments on energy consumption and bit-exact reconstruction. | a case-insensitive, word-bounded `git grep -P` over tracked `*.md`, `*.rs`, `*.json`, `*.toml`, `*.yml`, `*.yaml`, `*.html`, `*.ts` and `*.js` files finds none of these designations outside this record. The one other hit for `uwa` is the Hausa word in `aac/aac-rs/crates/aac-core/src/translations_data.json` (`"ha":"uwa"`), which is not the organisation. MPEG's splat work *is* tracked, though not Exploration 45 by name: `media-codec-registry.md` and `codecs/mod.rs` register `wai.splat.sog` against "MPEG WG4" work in progress, `3d-neural-coding-landscape.md` records splat coding under standardisation, and 2030 §6.4 records an NNC amendment covering Gaussian splats (G11) | **Partly closed by the two notes committed with this record.** [`syfai-input.md`](syfai-input.md) tracks SyfAI, IEEE 1857.11 and UWA 042; [`energy-exposure-mapping.md`](energy-exposure-mapping.md) tracks TR 26.942, DVB S100 and CM-EE. The same grep over the tree at link 10 finds the beyond-VVC CfP, Exploration 45, India D2M and the JPEG AI core experiments in no document but this record. Those four stay open |
 | G9 | **WAI's texts and code class `wai.image.jpegai` in ways that contradict one another, and SPEC contradicts itself on it.** SPEC §5, the capability's registration (`SPEC.md`:197), describes the decode as "integer entropy front end (mANS) + integer dequant/value path + fixed-order synthesis and post-filters" and declares "**decode-equivalence**: one pinned reconstruction … reproducible on every machine (a portable decoded-pixel hash)". `reconstruction-binding.md` §1 (:30–31) says the decode to RGB is "byte-identical on every machine", and §2 (:72) sets the binding's criterion at `decode-equivalence`; `determinism-tiers.md` §2–§3 list the capability there. SPEC §7 lets a neural capability back `replicate` only "with no float operation anywhere" (:445–448). `capability_replicate_class` returns `StandardDefined` for it through the default arm, and SPEC §7 defines that class as "reconstruction is fixed by the named codec/standard … the neural float hazard does not apply"; neither half holds for JPEG AI (§2.2; its decode runs the model). The strict `replicate` check reads only the envelope's prior (`neural/mod.rs`:282–292). **Loose ends:** (i) `jpegai_synthesis.rs` and `jpegai_mans.rs` say they back `wai.neural.jpeg_ai`, a string registered nowhere; (ii) the decode has f32 steps between its integer stages — its dequantisation among them, although SPEC §5 calls it integer; (iii) the byte-for-byte claim is made normatively, in SPEC §5 and `reconstruction-binding.md` §1, and rests on an `exp()` from the platform maths library and on no measurement on a second platform; (iv) the reference sink's dispatch (`neural/mod.rs`:205–216) uses the exact decoder only when the integer weight bundle is present, and otherwise falls back to a float ONNX Runtime decode. | `SPEC.md` §5 (:197) and §7 (:445–448, the `StandardDefined` bullet at :476–478, :492–495); `extensions/reconstruction-binding.md` §1 (:21–35) and §2 (:72); `extensions/determinism-tiers.md` §2–§3; `wai-rs/src/codecs/mod.rs` (the default arm and the test asserting `StandardDefined`); `wai-rs/src/jpegai_decode.rs`, `jpegai_synthesis.rs`, `jpegai_mans.rs`, `jpegai_pipeline.rs`; `wai-rs/src/neural/mod.rs`:205–216, :282–292; `git grep -n decode-equivalence` over `wai/` | **Closed at link 02 (PR #103), with the maths library removed at link 03 (PR #104).** SPEC §5's row registers `entropy-consistency` (registry class `NeuralEntropyExact`), describes dequantisation, latent refinement, post-filters and colour conversion as IEEE-754 float with a fixed evaluation order, and says "**Downgraded in v0.4.0:** this row previously claimed decode-equivalence". `reconstruction-binding.md` §1–§2 make the pinned reconstruction a property of the exact decoder, reproducible under that decoder's build and floating-point conditions. SPEC §7 states the new class and the reason. Loose ends: (i) fixed — the module docs name `wai.image.jpegai`, and `wai.neural.jpeg_ai` appears nowhere else in the tree (grep); (ii) the float steps remain and are now stated as float; (iii) link 03 builds `exp` and rounding from IEEE-754 basic operations (`float_repro`), guards the decode path with clippy's `disallowed_methods` and a linked-symbol check (`tests/jpegai_decode_symbols.rs`), and pins nine hashes for one codestream; determinism-tiers §6 records them identical on macOS arm64, wasm32-wasip2, Linux arm64 and Linux x86_64 under user-mode emulation, and not measured on Windows, macOS x86_64 or x86_64 hardware; (iv) the float fallback remains, and SPEC §5, determinism-tiers §5 and reconstruction-binding §1 now state it, and that the result does not record which path ran. Still true: a `norm.json` with a missing or mistyped key panics in the loaders instead of falling back, as the `try_jpegai_exact` doc comment says |
 | G10 | **WAI's texts give three conditions for a neural decode to back `replicate` or a portable reconstruction hash, and SPEC states all three.** (1) **All-integer:** SPEC §7 (`SPEC.md`:445–448, :492–495), enforced in code on the envelope's declared prior (`replicate_is_sound` requires `DeterminismClass::IntegerExact`, a substring heuristic over `Prior.determinism`). (2) **`decode-equivalence` with fixed-order float admitted:** `determinism-tiers.md` §2 defines the tier over "integer / fixed-order arithmetic end to end" (:57–58), and SPEC §5 registers `wai.image.jpegai`, whose float stages run in a fixed order, at that tier. (3) **`entropy-consistency` as the floor:** SPEC §7's closing paragraph (:497–505) calls it "the floor for a sound neural `replicate`", and `determinism-tiers.md` §2–§3 repeat it; §3 also says "a float neural decode meets neither tier", which does not square with the explicit-parameter route. A decode with integer entropy and fixed-order float synthesis meets (2) and (3) but not (1). | `SPEC.md` §5 (:197) and §7 (:445–448, :492–495, :497–505); `extensions/determinism-tiers.md` §1–§3; `extensions/reconstruction-binding.md` §1–§2; `wai-rs/src/container.rs` (`Prior::determinism_class` at :108–117, `is_cross_hardware_exact`, `replicate_is_sound`) | **Closed at link 02 (PR #103), on condition (1).** SPEC §7: a neural capability backs `replicate` only if its entire decode path is integer / fixed-point with no float operation, "the only construction by which a neural capability is registered at `decode-equivalence`"; `entropy-consistency` ("necessary for a sound neural `replicate` … and not sufficient") and a reconstruction pinned by one implementation's fixed-order float stages MUST NOT back it. determinism-tiers §2 defines `decode-equivalence` over "integer / fixed-point arithmetic end to end", and §3 says a decode that derives its entropy parameters in float meets neither tier while one that reads them from the codestream, or derives them in integer arithmetic, and runs later stages in float meets `entropy-consistency` only. The strict check reads the registry class before the prior. Remaining: the prior check still reads a sender-declared string by substring, which determinism-tiers §3 states |
@@ -863,6 +863,9 @@ conclusion is also exposed to the unreached areas in §1.
   PR #103) the consequence for `replicate` is settled either way: a decode that
   is not integer / fixed-point end to end does not back it, whatever its
   reconstructions do on the devices tested.
+  *Corrected 2026-11-03, body read:* arXiv 2608.00483 synchronises the entropy
+  stage by mixed precision, and the identical output it reports holds only
+  within the groups of GPUs its tests found (Table IV).
 - **F6. "SyfAI's AI update-framework thread covers the properties WAI's `Prior`
   and determinism tiers specify" (§2.1, G3).** This rests on a **secondary**
   report. It is falsified or narrowed if either of the following scopes bit
diff --git a/wai/research/catalogue-prior.md b/wai/research/catalogue-prior.md
new file mode 100644
index 000000000..554072dea
--- /dev/null
+++ b/wai/research/catalogue-prior.md
@@ -0,0 +1,326 @@
+# What a linear catalogue prior costs, measured on a studio's open films
+
+> **Status.** A measured record and its procedure, informative. Every measured figure below
+> is computed from the readings in [`catalogue-prior/`](catalogue-prior/), and
+> `tools/wai_catalogue_prior_verify.py`, which shares no code with the scripts that took
+> them, recomputes each from those readings and checks every figure this page quotes (each
+> is followed by a `cat` marker naming the value it is checked against). The readings were
+> taken on 4 November 2026 (UTC). No WAI capability is measured here: the record decides
+> whether a catalogue basis should become one, and §5 says it should not.
+
+## 1. The question, and what is measured
+
+A catalogue prior is a prior learned from a catalogue of titles and held by each sink, so
+that a title carries only what the prior does not predict (SPEC §3.1, the shared ambient
+prior). WAI already carries such a prior: pinned by its SHA-256, served by digest, and
+refused by name when a sink does not hold it (SPEC §3.1, [prior-carriage](../extensions/prior-carriage.md)).
+The claim splits into three, each of which can be refuted on its own (run.json, `rule`):
+
+- **H1a, a linear or spectral basis.** A basis learned on the catalogue (a mean frame, or
+  the principal components of its frames) captures a held-out title better than a generic
+  transform, and in front of a codec it saves bytes (§3.3, §3.5).
+- **H1b, exact recurrence.** Frames recur across titles, byte for byte or within a tight
+  bound, so a sink that holds them once never fetches them again (§3.4).
+- **H1c, a learned prior.** A neural codec's parameter set tuned on the catalogue saves
+  bytes against its generic set on a held-out title. Nothing here measures it (§6).
+
+A catalogue is a set of whole titles under one rights holder, each held out whole when it
+is tested (§3.1). This record uses openly licensed films of one studio: they share a
+production pipeline and a house style, but no characters or sets (§3.1). A negative result
+on such a catalogue refutes H1a and H1b for a studio's catalogue, not for a franchise, and
+says nothing of H1c (§6).
+
+The prior's bytes are charged once per sink, not once per catalogue, because the sink
+reconstructs and so every sink that decodes must hold it (SPEC §3.1). The break-even is the
+prior's bytes divided by the mean saving per title, in titles watched per sink (run.json,
+`rule.go`).
+
+## 2. The rule, and when it was fixed
+
+The rule below was written at 2026-11-04T01:58Z, before any result of measurements (a) or
+(c) had been seen (run.json, `timeline`). That is attested, not proved: no copy of the rule
+was committed anywhere before the results existed. The record is consistent with it
+(run.json, `timeline`):
+
+1. The rate proxy's four target PSNRs were recorded with the basis script at 01:43:38Z.
+2. Moment accumulation finished at 01:49:04Z, and printed no figure.
+3. The residual-coding script was last changed at 01:58:51Z.
+4. The first result file, `a_basis.json`, was written at 02:00:43Z. The basis script's only
+   later change, at 02:01:24Z, is a print format.
+
+The rule (run.json, `rule`):
+
+- **GO**, worth a capability and a demonstration, needs both: a BD-rate of −10 % or less
+  against per-title AV1 on both PSNR and VMAF, with the side information charged, on at
+  least three of the four held-out titles; and a break-even of at most five titles per
+  sink, with the prior's bytes counted.
+- **K1 (H1a)** holds when the catalogue basis's rate proxy is not at least 5 % below a
+  generic DCT's at both 35 dB and 40 dB on the median title.
+- **K2 (H1a)** holds when the catalogue-basis residual and its coefficients cost more bytes
+  than AV1 alone (a BD-rate above 0) on both metrics on the median title.
+- **K3 (H1b)** holds when non-flat material recurring across titles within a tight bound
+  (a thumbnail PSNR of 40 dB or more and a full-size PSNR of 35 dB or more) is under 1 % of
+  each title's runtime.
+- Anything between GO and a kill is inconclusive.
+
+### 2.1 Faults in the rule, found after the run
+
+Three faults were found after the results existed. None changes this run's verdicts, which
+rest on K2 and K3 (§5):
+
+1. No clause can kill H1c: K1 and K2 concern a linear basis, and K3 recurrence.
+2. GO and K1 can both hold: K1 judges an intra rate proxy with no prediction, and GO a
+   real codec. Here K2 holds as well, so the kill does not rest on K1 alone (§5).
+3. The per-title AV1 bar was run through a non-standard 10-bit pipeline, `base10`, which is
+   less efficient than plain 8-bit AV1 (§4.3). A −10 % GO against `base10` is
+   −3.1 %<!--cat:go_vs_base8_pct--> against plain AV1 (§4.3). Plain per-title AV1 is the stricter bar, and every residual arm is
+   further from it (§4.3).
+
+## 3. Procedure
+
+### 3.1 Corpus
+
+Openly licensed films of one studio, each named in run.json by its file's size and
+SHA-256, with the SHA-256 of the licence page fetched for it (run.json, `corpus`):
+
+| code | title (year) | licence, as its page states it | file SHA-256 |
+|---|---|---|---|
+| ED | Elephants Dream (2006) | Creative Commons, version not stated on the page fetched | `6fc295e0…f7558a547` |
+| BBB | Big Buck Bunny (2008) | Creative Commons Attribution 3.0 | `b0c9ade8…f88b33b67` (the archive) |
+| SIN | Sintel (2010) | Creative Commons Attribution 3.0 | `fae11ed4…19c3d9920` |
+| TOS | Tears of Steel (2012) | Creative Commons Attribution 3.0 | `99486359…e185987a15` |
+
+Another short film of the same studio was left out because no licence statement could be
+fetched for it (§6). The films' frames are not in this repository; only figures derived
+from them are (§3.2).
+
+### 3.2 Analysis raster
+
+Each title was decoded to luma at 640×360 by area scaling, kept as decoded: 8-bit, limited
+range (run.json, `analysis_raster`). The two wide films were stretched to 16:9, since no
+title has letterbox bars (§6). From that raster: 128×72 by a 5×5 area mean, in float, and
+64×36 by a 10×10 area mean, rounded to 8 bits (run.json, `analysis_raster`).
+
+### 3.3 Measurement (a): does a catalogue basis capture a held-out title better than a generic transform?
+
+- **Split.** Each title is cut in time into a first half A and a second half B; the test
+  set is half B of the held-out title (`a_basis.json`).
+- **Bases.** The catalogue basis is the principal components (the Karhunen–Loève transform)
+  of the other titles, leave one title out; the per-title basis is that of half A of
+  the held-out title; the generic basis is an orthonormal 2-D DCT-II whose coefficients are
+  ordered by the catalogue's training statistics (`a_basis.json`). Each vector has its own
+  mean removed first, and the principal-component bases also remove their training mean,
+  the catalogue's "central mean" (`a_basis.json`).
+- **Granularities.** Whole frames at 128×72; 8×8 and 16×16 blocks at 640×360, every second
+  frame; and 8×8 blocks of the difference of consecutive frames, an un-compensated stand-in
+  for inter-frame residuals (`a_basis.json`).
+- **Rate proxy.** The Gaussian reverse water-filling rate of the test coefficients' second
+  moments, at the mean squared error of a target PSNR: the classical yardstick for ranking
+  transforms, as an ideal intra coder with no prediction (`a_basis.json`). It ranks
+  transforms; its bits per pixel are not a codec's (§6). Every score comes from first and
+  second moments over every test vector, so nothing is sampled (`a_basis.json`).
+
+### 3.4 Measurement (b): what recurs across titles?
+
+- **Thumbnails.** 64×36, 8-bit. A frame is flat when its thumbnail's standard deviation is
+  under 2 code values, as black frames and fades to one tone are (`b_recur.json`).
+- **Recurrence.** Identical thumbnail bytes in another title (exact), or the nearest
+  non-flat frame of another title at a thumbnail PSNR of 35 dB or 40 dB, then measured again
+  on the two frames at 640×360 (`b_recur.json`). The nearest frame of the same title at
+  least 10 s away is measured the same way: a key frame every 10 s puts it out of the
+  encoder's reach in this run (`d_exact.json`).
+- **Bytes.** Each whole title was encoded once with SVT-AV1, preset 8, CRF 36, a key frame
+  every 240 frames and `tune=1`, at 640×360 luma, and each temporal unit's bytes were
+  charged to the frame it shows (`d_exact.json`). The attribution is approximate, since
+  hidden reference data rides in the unit that carries it (§6).
+
+### 3.5 Measurement (c): bytes at matched quality, residual coding against per-title AV1
+
+- **The clip.** For each title, 1,440 frames (60 s) starting at 60 % of the title, inside
+  half B, so neither basis saw it (`c_rd.json`).
+- **The arms** (`c_rd.json`):
+  - `base8`: AV1, 8-bit, on the clip.
+  - `base10`: AV1, 10-bit, on 2*x* + 256, the residual arms' own pipeline and scale.
+  - `catK`: AV1, 10-bit, on 2(*x* − *x̂*) + 512, where *x̂* is the bilinear ×5 upsampling of
+    the frame's DC, the catalogue mean and *K* catalogue components at 128×72. `cat0` is the
+    mean alone; `cat64` and `cat256` add 64 and 256 components. The coefficients are
+    quantised with a step of 2 (the DC with a step of 0.25) and charged at the larger of
+    two estimates of their bytes: their per-index empirical entropy over frame-to-frame
+    differences, and a signed Exp-Golomb code. The reconstruction is round(*x̂* + (*y* − 512)/2).
+  - `self64`: as `cat64`, with the per-title basis of half A.
+- **Coding.** SVT-AV1 v4.2.0 through FFmpeg, preset 8, CRF 20, 28, 36, 44 and 52, a key
+  frame every 240 frames, `tune=1`; decoded by dav1d; bytes are the sum of the IVF frame
+  payloads (`c_rd.json`).
+- **Quality.** Luma PSNR over the clip's pooled squared error, and VMAF (libvmaf, model
+  `vmaf_v0.6.1`, mean pooling) on the same luma (`c_rd.json`). VMAF is not reproducible to
+  the bit across builds, so it is a second check, and every verdict below holds on PSNR
+  alone (§5).
+- **BD-rate.** The Bjøntegaard delta rate: the natural log of bytes fitted as a
+  least-squares cubic in quality for each arm, both fits integrated over the quality range
+  the two arms share, and the mean difference turned into a percentage (`c_rd.json`,
+  `bd_vs_base10`). A positive figure is the extra bytes an arm spends for the same quality.
+  The verifier recomputes every figure the run stored, and also integrates a
+  piecewise-linear interpolant of the same points (§4.3).
+
+### 3.6 Host and tools
+
+A shared Linux x86_64 build host, each step in a user scope with a 64 GB memory cap, a low
+CPU weight and `nice 10` (run.json, `host`). FFmpeg 9.0.2 with SVT-AV1 4.2.0, dav1d 1.5.4
+and libvmaf 3.2.1; Python 3.14.4 with NumPy 2.3.5 (run.json, `tools`). The measurement
+scripts are not part of this record; §3.3–§3.5 state what they did.
+
+## 4. Results
+
+### 4.1 Measurement (a)
+
+The rate proxy of the catalogue basis against the DCT's, the median over the titles; a
+negative figure is fewer bits (`a_basis.json`, §3.3):
+
+| granularity | at 35 dB | at 40 dB |
+|---|---|---|
+| whole frames, 128×72 | +18.1 %<!--cat:proxy.frame128x72.catalogue.35.median--> | +10.4 %<!--cat:proxy.frame128x72.catalogue.40.median--> |
+| 8×8 blocks | −0.6 %<!--cat:proxy.block8.catalogue.35.median--> | −0.3 %<!--cat:proxy.block8.catalogue.40.median--> |
+| 16×16 blocks | +0.2 %<!--cat:proxy.block16.catalogue.35.median--> | +0.1 %<!--cat:proxy.block16.catalogue.40.median--> |
+| 8×8 frame differences | −0.9 %<!--cat:proxy.block8_diff.catalogue.35.median--> | −0.6 %<!--cat:proxy.block8_diff.catalogue.40.median--> |
+
+- Over whole frames, a basis learned on the title's own first half is further from the DCT
+  still, by +38.7 %<!--cat:proxy.frame128x72.per_title.35.SIN--> to
+  +55.7 %<!--cat:proxy.frame128x72.per_title.35.ED--> at 35 dB: a film's frame statistics
+  are not stationary from its first half to its second (`a_basis.json`).
+- The share of a held-out title's detail energy that the catalogue mean alone captures is
+  −3.05 %<!--cat:mean_frame.ED.captured_ac_pct--> on ED, so that subtracting it adds energy
+  there, and 1.72 %<!--cat:mean_frame.SIN.captured_ac_pct--> to
+  2.87 %<!--cat:mean_frame.TOS.captured_ac_pct--> on the other titles
+  (`a_basis.json`).
+- Counting each frame's DC, the mean alone captures
+  78.4 %<!--cat:mean_frame.captured_incl_dc_pct.min--> to
+  92.5 %<!--cat:mean_frame.captured_incl_dc_pct.max--> of a title's energy, and sixteen
+  components capture 89.8 %<!--cat:mean_frame.k16_incl_dc_pct.min--> to
+  96.5 %<!--cat:mean_frame.k16_incl_dc_pct.max-->, as sixteen DCT components do, within
+  0.3<!--cat:mean_frame.k16_vs_dct_points_max--> of a point on every title (`a_basis.json`). A test of the form "a given share of spectral mass in a given
+  number of components" is met by any of these transforms, and so does not discriminate.
+- Over 8×8 and 16×16 blocks the catalogue basis and the DCT are within
+  1.2 %<!--cat:proxy_block_title_abs_max--> of each other on every title (`a_basis.json`), as the DCT's closeness to the Karhunen–Loève transform for natural
+  images predicts (Ahmed, Natarajan and Rao, IEEE Trans. Computers C-23(1), 1974).
+
+### 4.2 Measurement (b)
+
+Shares of each title's runtime and of its AV1 bytes at CRF 36 (`d_exact.json`,
+`b_recur.json`, §3.4):
+
+| title | flat frames: runtime / bytes | non-flat, exact in another title | another title's frame at 35 dB, both sizes: runtime / bytes | its own frame 10 s or more away at 35 dB: runtime / bytes |
+|---|---|---|---|---|
+| ED | 1.45 %<!--cat:recur.ED.flat.runtime_pct--> / 0.211 %<!--cat:recur.ED.flat.bytes_pct--> | 0<!--cat:recur.ED.exact_nonflat--> | 0.083 %<!--cat:recur.ED.cross35.runtime_pct--> / 0.002 %<!--cat:recur.ED.cross35.bytes_pct--> | 0.070 %<!--cat:recur.ED.self_far35.runtime_pct--> / 0.002 %<!--cat:recur.ED.self_far35.bytes_pct--> |
+| BBB | 0.66 %<!--cat:recur.BBB.flat.runtime_pct--> / 0.649 %<!--cat:recur.BBB.flat.bytes_pct--> | 0<!--cat:recur.BBB.exact_nonflat--> | 0.035 %<!--cat:recur.BBB.cross35.runtime_pct--> / 0.001 %<!--cat:recur.BBB.cross35.bytes_pct--> | 0.049 %<!--cat:recur.BBB.self_far35.runtime_pct--> / 0.002 %<!--cat:recur.BBB.self_far35.bytes_pct--> |
+| SIN | 2.26 %<!--cat:recur.SIN.flat.runtime_pct--> / 0.173 %<!--cat:recur.SIN.flat.bytes_pct--> | 0<!--cat:recur.SIN.exact_nonflat--> | 0.169 %<!--cat:recur.SIN.cross35.runtime_pct--> / 0.010 %<!--cat:recur.SIN.cross35.bytes_pct--> | 1.253 %<!--cat:recur.SIN.self_far35.runtime_pct--> / 0.984 %<!--cat:recur.SIN.self_far35.bytes_pct--> |
+| TOS | 1.53 %<!--cat:recur.TOS.flat.runtime_pct--> / 0.278 %<!--cat:recur.TOS.flat.bytes_pct--> | 0<!--cat:recur.TOS.exact_nonflat--> | 0 %<!--cat:recur.TOS.cross35.runtime_pct--> / 0 %<!--cat:recur.TOS.cross35.bytes_pct--> | 0 %<!--cat:recur.TOS.self_far35.runtime_pct--> / 0 %<!--cat:recur.TOS.self_far35.bytes_pct--> |
+
+- No non-flat frame of one title matches another title's at 40 dB at full size
+  (`b_recur.json`, `cross_ge35.full_psnr_ge40`).
+- The frames that match another title's at 35 dB are dark, nearly flat end-credit frames,
+  their thumbnails spreading a median of 2.6<!--cat:recur.cross35_spread.min--> to
+  3.0<!--cat:recur.cross35_spread.max--> code values against
+  30<!--cat:recur.ordinary_spread.min--> to 37<!--cat:recur.ordinary_spread.max--> for the
+  titles' ordinary non-flat frames (`d_exact.json`, `thumb_std_median`).
+- The one recurrence of size is inside one title: SIN shows earlier shots again minutes
+  later, 0.98 %<!--cat:recur.SIN.self_far35.bytes_pct--> of its bytes (`d_exact.json`,
+  `self_far35`). It needs no catalogue.
+
+### 4.3 Measurement (c)
+
+The BD-rate of each arm against AV1 through the same 10-bit pipeline (`base10`), the median
+over the titles, PSNR / VMAF; GO needs −10 % or less on both (`c_rd.json`, §3.5):
+
+| arm | median BD-rate, PSNR / VMAF | AV1's bytes only, PSNR / VMAF | against plain 8-bit AV1 (`base8`), PSNR |
+|---|---|---|---|
+| `cat0`, the mean alone | +2.1 %<!--cat:bd.cat0.base10.psnr.median--> / +2.4 %<!--cat:bd.cat0.base10.vmaf.median--> | +2.1 %<!--cat:bd_av1only.cat0.base10.psnr.median--> / +2.4 %<!--cat:bd_av1only.cat0.base10.vmaf.median--> | +10.4 %<!--cat:bd.cat0.base8.psnr.median--> |
+| `cat64` | +38.8 %<!--cat:bd.cat64.base10.psnr.median--> / +39.9 %<!--cat:bd.cat64.base10.vmaf.median--> | +31.8 %<!--cat:bd_av1only.cat64.base10.psnr.median--> / +29.6 %<!--cat:bd_av1only.cat64.base10.vmaf.median--> | +46.6 %<!--cat:bd.cat64.base8.psnr.median--> |
+| `cat256` | +77.7 %<!--cat:bd.cat256.base10.psnr.median--> / +80.5 %<!--cat:bd.cat256.base10.vmaf.median--> | +50.4 %<!--cat:bd_av1only.cat256.base10.psnr.median--> / +48.5 %<!--cat:bd_av1only.cat256.base10.vmaf.median--> | +80.3 %<!--cat:bd.cat256.base8.psnr.median--> |
+| `self64`, the title's own basis | +52.1 %<!--cat:bd.self64.base10.psnr.median--> / +52.0 %<!--cat:bd.self64.base10.vmaf.median--> | +44.9 %<!--cat:bd_av1only.self64.base10.psnr.median--> / +43.1 %<!--cat:bd_av1only.self64.base10.vmaf.median--> | +60.9 %<!--cat:bd.self64.base8.psnr.median--> |
+
+- **The coefficients are not the main cost.** With their bytes left out, AV1 still spends
+  more on the residual than on the picture: the middle column (`c_rd.json`, `av1_bytes`).
+  The coefficients are 1.5 %<!--cat:side_pct.cat64.min--> to
+  15.2 %<!--cat:side_pct.cat64.max--> of `cat64`'s bytes across the ladder (`c_rd.json`,
+  `side_bytes`).
+- **The basis's own bytes**, at 16 bits per value, are 18,432 B<!--cat:basis_bytes.cat0-->
+  for the mean alone, 1,198,080 B<!--cat:basis_bytes.cat64--> for 64 components and
+  4,737,024 B<!--cat:basis_bytes.cat256--> for 256 (§3.5). Plain AV1 spends
+  0.9 MB<!--cat:base8_mb.min--> to 11.6 MB<!--cat:base8_mb.max--> on one 60 s clip across
+  the ladder (`c_rd.json`, `base8`). No arm saves bytes on PSNR on any title against either
+  bar, so no number of titles watched recovers the basis's bytes, and there is no break-even
+  (`c_rd.json`; the verifier's `no_psnr_saving_any_title`).
+- **The reconstruction from the basis alone** reaches 14.1 dB<!--cat:basis_only_psnr.cat0.min-->
+  to 22.0 dB<!--cat:basis_only_psnr.cat64.max--> for the mean and for 64 components
+  (`c_rd.json`, `basis_only_psnr`). The share of residual samples clipped was
+  0<!--cat:clipped_max--> in every arm (`c_rd.json`, `clipped_frac`).
+- **The pipeline.** `base10` against `base8` costs a median of
+  +7.7 %<!--cat:bd.base10.base8.psnr.median--> on PSNR and
+  +1.5 %<!--cat:bd.base10.base8.vmaf.median--> on VMAF (`c_rd.json`): the ×2 scaling into
+  10 bits is not the usual ×4, and it is less efficient than plain 8-bit AV1.
+- **The interpolation.** Integrated piecewise-linearly instead, no PSNR figure above moves
+  by more than 0.48<!--cat:linear_shift_psnr_max--> of a point (the verifier's
+  `bd_linear`). VMAF near saturation is not stable under the cubic fit: on TOS, `cat0`
+  against `base8` gives −5.0 %<!--cat:bd.cat0.base8.vmaf.TOS--> by the cubic and
+  +10.2 %<!--cat:bd_linear.cat0.base8.vmaf.TOS--> by the linear interpolant, and `base10`
+  against `base8` −8.3 %<!--cat:bd.base10.base8.vmaf.TOS--> and
+  +6.6 %<!--cat:bd_linear.base10.base8.vmaf.TOS--> (the verifier's `bd`, `bd_linear`).
+  That VMAF "saving" is an artefact of the fit, and no verdict rests on VMAF (§5).
+
+## 5. Verdict against the rule
+
+The verifier applies the rule run.json records to the figures above and requires the
+verdicts it states (run.json, `verdicts`):
+
+| claim | result | rule |
+|---|---|---|
+| H1a, a linear or spectral catalogue basis | the rate proxy is +10.4 %<!--cat:proxy.frame128x72.catalogue.40.median--> to +18.1 %<!--cat:proxy.frame128x72.catalogue.35.median--> from the DCT's over whole frames, and its medians are within 0.9 %<!--cat:proxy_block_median_abs_max--> of it over blocks and block differences (§4.1); in front of AV1, +38.8 %<!--cat:bd.cat64.base10.psnr.median--> on PSNR at 64 components (§4.3) | **K1 and K2 hold: refuted for a studio's catalogue** |
+| H1b, exact recurrence across titles | non-flat recurrence across titles is at most 0.169 %<!--cat:recur.cross35_runtime_pct_max--> of a title's runtime and 0.010 %<!--cat:recur.cross35_bytes_pct_max--> of its bytes, even at the looser bound of 35 dB at both sizes (§4.2) | **K3 holds: refuted for a studio's catalogue; untested for an episodic series** |
+| H1c, a learned prior tuned on the catalogue | not measured | **open** |
+
+GO is missed by 48.8<!--cat:go_gap_points.cat64--> points at 64 components (§4.3).
+
+Why a linear basis fails here, on the mechanism the figures show:
+
+1. A predictive codec already holds a better prior, the frames it has just decoded, which
+   it refreshes every frame and never sends. The basis alone reaches only the coarse
+   structure that prediction already gets (§4.3).
+2. Subtracting a reconstruction that changes with its coefficients, not with the motion,
+   makes the residual less predictable in time than the picture. That is the extra cost
+   before any coefficient is charged (§4.3, AV1's bytes only).
+3. A film's statistics change from its first half to its second, so a basis fitted to some
+   titles fits others no better than a fixed transform, and more titles make it more
+   generic (§4.1).
+
+The mechanism is general, so the result probably extends to a franchise. That is an
+inference: it was measured on films of one studio only (§6).
+
+## 6. What this does not show
+
+- **The corpus.** Films of one studio, not a franchise: they share no characters or
+  sets (§3.1). TOS is live action with computer graphics. ED's source is 720×405, softer
+  than the others' (run.json, `corpus`). Another short film was left out for want of a
+  fetched licence statement (§3.1).
+- **The signal.** Luma only, at 640×360, with the two wide films stretched to 16:9; no
+  chroma and no 1080p (§3.2).
+- **The codec side.** One encoder at one preset with `tune=1`, one 60 s clip per title, and
+  a ×2 10-bit pipeline for the residual arms, compared with the same pipeline (§3.5, §4.3).
+- **The simplifications.** Linear bases only. The rate proxy is a Gaussian high-rate rate
+  with no prediction (§3.3). The coefficients' bytes are estimates, the larger of two, and a
+  better coefficient coder cannot rescue `cat64`, whose AV1 bytes alone cost more (§4.3).
+  Recurrence is matched pixel-aligned only, so an element drawn at another size or place
+  would be missed (§3.4). The byte attribution to frames in measurement (b) is approximate
+  (§3.4).
+- **Not measured.** H1c, a learned prior; an episodic series, whose opening could recur
+  across episodes; energy; decode cost (§1).
+- **The rule** was attested, not committed, before the results (§2).
+
+## 7. Verify
+
+```sh
+python3 -I tools/wai_catalogue_prior_verify.py research/catalogue-prior research/catalogue-prior.md
+```
+
+It checks each reading against the SHA-256 in run.json, recomputes every stored BD-rate,
+the rate proxy, the recurrence and the verdicts, and every quoted figure (the module's
+docstring). The conformance workflow runs it (`.github/workflows/wai-conformance.yml`).
diff --git a/wai/research/catalogue-prior/a_basis.json b/wai/research/catalogue-prior/a_basis.json
new file mode 100644
index 000000000..284f6ebc0
--- /dev/null
+++ b/wai/research/catalogue-prior/a_basis.json
@@ -0,0 +1,1770 @@
+{
+ "frame128x72": {
+  "ED": {
+   "n_test": 7847,
+   "E_ac": 11632755.668506913,
+   "E_raw": 61810750.43416485,
+   "catalogue": {
+    "captured_ac": {
+     "0": -0.03049958150026133,
+     "1": 0.10305805352319419,
+     "4": 0.3000898049108234,
+     "16": 0.5233924902996061,
+     "64": 0.6795552181080031,
+     "256": 0.8129509102157015,
+     "1024": 0.9192570066064154,
+     "4096": 0.9791000100262187
+    },
+    "captured_incl_dc": {
+     "0": 0.8060604382912451,
+     "1": 0.8311959256454438,
+     "4": 0.8682772133944339,
+     "16": 0.9103026792074713,
+     "64": 0.9396924349435306,
+     "256": 0.9647974447131988,
+     "1024": 0.9848042046489655,
+     "4096": 0.9960666312068449
+    },
+    "psnr_k": {
+     "0": 16.98892400828313,
+     "1": 17.591758858217315,
+     "4": 18.668979009554295,
+     "16": 20.337793395142644,
+     "64": 22.06187016624337,
+     "256": 24.399846213892047,
+     "1024": 28.048353745524643,
+     "4096": 33.917941428009655
+    },
+    "bpp_at_psnr": {
+     "30": 0.4579595539363131,
+     "35": 1.2155350241112564,
+     "40": 2.0447703872141227,
+     "45": 2.87516238802275
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": 0.018687678385231665,
+     "1": 0.10735776996200053,
+     "4": 0.2760836239189578,
+     "16": 0.4646558962089753,
+     "64": 0.6499308721267906,
+     "256": 0.7918023969646596,
+     "1024": 0.8907977464687253,
+     "4096": 0.9620100129900897
+    },
+    "captured_incl_dc": {
+     "0": 0.8153174586676718,
+     "1": 0.832005130361859,
+     "4": 0.8637592608368159,
+     "16": 0.8992484783919621,
+     "64": 0.9341171462402199,
+     "256": 0.960817303950087,
+     "1024": 0.9794481845818601,
+     "4096": 0.9928503013855698
+    },
+    "psnr_k": {
+     "0": 17.20132968576503,
+     "1": 17.612627915295647,
+     "4": 18.522518194345135,
+     "16": 19.83307196794907,
+     "64": 21.677864079647907,
+     "256": 23.934644953699006,
+     "1024": 26.737086198751186,
+     "4096": 31.32271075516301
+    },
+    "bpp_at_psnr": {
+     "30": 0.7041283482517924,
+     "35": 1.53385533079303,
+     "40": 2.3643373545148703,
+     "45": 3.1948193782367103
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": -8.881784197001252e-16,
+     "1": 0.0918184368738254,
+     "4": 0.30024290409309407,
+     "16": 0.5286039399675658,
+     "64": 0.7010933355947647,
+     "256": 0.8323496665465081,
+     "1024": 0.9305415471856563,
+     "4096": 0.9854019694133626
+    },
+    "captured_incl_dc": {
+     "0": 0.8118004459289477,
+     "1": 0.8290806348041029,
+     "4": 0.8683060265922659,
+     "16": 0.911283471711045,
+     "64": 0.9437458990500691,
+     "256": 0.9684482820041896,
+     "1024": 0.9869279501538752,
+     "4096": 0.9972526571532793
+    },
+    "psnr_k": {
+     "0": 17.11940220570751,
+     "1": 17.537675394711634,
+     "4": 18.669929094345687,
+     "16": 20.385542722367596,
+     "64": 22.364046224195505,
+     "256": 24.87535799001155,
+     "1024": 28.702151153892213,
+     "4096": 35.47645951296436
+    },
+    "bpp_at_psnr": {
+     "30": 0.3315661233115439,
+     "35": 0.9853460453352721,
+     "40": 1.8057783282425852,
+     "45": 2.636170328724157
+    }
+   }
+  },
+  "BBB": {
+   "n_test": 7158,
+   "E_ac": 11594320.698514603,
+   "E_raw": 151487520.70526186,
+   "catalogue": {
+    "captured_ac": {
+     "0": 0.019175155252679543,
+     "1": 0.12714595901920012,
+     "4": 0.3767708676825742,
+     "16": 0.5384435118670717,
+     "64": 0.6686502626272893,
+     "256": 0.7970777338199853,
+     "1024": 0.9025186556089457,
+     "4096": 0.9708764781234492
+    },
+    "captured_incl_dc": {
+     "0": 0.9249311247149071,
+     "1": 0.933194829336366,
+     "4": 0.9523002396824214,
+     "16": 0.9646740938198783,
+     "64": 0.9746396594215674,
+     "256": 0.9844690452387956,
+     "1024": 0.9925391216139101,
+     "4096": 0.9977709883234283
+    },
+    "psnr_k": {
+     "0": 17.21786059981695,
+     "1": 17.724358913043112,
+     "4": 19.18729772309257,
+     "16": 20.491526579206344,
+     "64": 21.930908874695003,
+     "256": 24.06047814421934,
+     "1024": 27.244560081513363,
+     "4096": 32.491336255886736
+    },
+    "bpp_at_psnr": {
+     "30": 0.5820816527981432,
+     "35": 1.3903456983162985,
+     "40": 2.2206976500362017,
+     "45": 3.0510895675482876
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -0.04417380007951155,
+     "1": 0.0882466098771385,
+     "4": 0.2896921279632365,
+     "16": 0.4581303125033652,
+     "64": 0.5902997726290017,
+     "256": 0.7341721170069183,
+     "1024": 0.8544755808425298,
+     "4096": 0.9405580430292593
+    },
+    "captured_incl_dc": {
+     "0": 0.9200826190385463,
+     "1": 0.9302176103098956,
+     "4": 0.9456355399789201,
+     "16": 0.9585272047202918,
+     "64": 0.96864298917641,
+     "256": 0.9796544711295024,
+     "1024": 0.988862074068401,
+     "4096": 0.9954505222023735
+    },
+    "psnr_k": {
+     "0": 16.946047262781722,
+     "1": 17.535001313733318,
+     "4": 18.619308903778652,
+     "16": 19.794826611679397,
+     "64": 21.009113132023572,
+     "256": 22.887769855308086,
+     "1024": 25.50441643647382,
+     "4096": 29.39284419179435
+    },
+    "bpp_at_psnr": {
+     "30": 0.9423044847785802,
+     "35": 1.7727865085004209,
+     "40": 2.603268532222261,
+     "45": 3.433750555944102
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 3.3306690738754696e-16,
+     "1": 0.101027670398399,
+     "4": 0.36653509932177886,
+     "16": 0.5387714193285753,
+     "64": 0.6853927884022322,
+     "256": 0.8128285346425311,
+     "1024": 0.9170681331936085,
+     "4096": 0.9789293729113822
+    },
+    "captured_incl_dc": {
+     "0": 0.9234635259423591,
+     "1": 0.9311958276169101,
+     "4": 0.9515168300628153,
+     "16": 0.964699190700799,
+     "64": 0.9759210733112007,
+     "256": 0.9856745559973374,
+     "1024": 0.9936526873276209,
+     "4096": 0.9983873284964537
+    },
+    "psnr_k": {
+     "0": 17.133775181756185,
+     "1": 17.596311938353576,
+     "4": 19.116549618828625,
+     "16": 20.494613070566338,
+     "64": 22.156088446397177,
+     "256": 24.411378777391924,
+     "1024": 27.946560766617456,
+     "4096": 33.897000572199275
+    },
+    "bpp_at_psnr": {
+     "30": 0.4442539979433081,
+     "35": 1.1828763241007931,
+     "40": 2.0122211991726875,
+     "45": 2.842613116129323
+    }
+   }
+  },
+  "SIN": {
+   "n_test": 10656,
+   "E_ac": 10871058.247009454,
+   "E_raw": 49514801.0074625,
+   "catalogue": {
+    "captured_ac": {
+     "0": 0.01717390951382558,
+     "1": 0.2011987825883823,
+     "4": 0.44950296422963265,
+     "16": 0.6096736098553203,
+     "64": 0.7198888157299055,
+     "256": 0.823618880468989,
+     "1024": 0.915771208832764,
+     "4096": 0.9748972353936047
+    },
+    "captured_incl_dc": {
+     "0": 0.7842188707423964,
+     "1": 0.8246218426495342,
+     "4": 0.8791374453920602,
+     "16": 0.9143031813463567,
+     "64": 0.9385011160727464,
+     "256": 0.9612752271021876,
+     "1024": 0.9815074265426974,
+     "4096": 0.9944886456040496
+    },
+    "psnr_k": {
+     "0": 17.48874327340568,
+     "1": 18.389122861680576,
+     "4": 20.005960194593605,
+     "16": 21.499230891317204,
+     "64": 22.94020554587521,
+     "256": 24.94898909411512,
+     "1024": 28.158904364139463,
+     "4096": 33.4162945061319
+    },
+    "bpp_at_psnr": {
+     "30": 0.4517228704595164,
+     "35": 1.2298850437397215,
+     "40": 2.059935322204682,
+     "45": 2.890327232859712
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -0.002522984486078572,
+     "1": 0.194248525966202,
+     "4": 0.421938493465689,
+     "16": 0.6029445141241806,
+     "64": 0.7203144512888213,
+     "256": 0.8221805035384642,
+     "1024": 0.8994231031641615,
+     "4096": 0.9615642112660967
+    },
+    "captured_incl_dc": {
+     "0": 0.7798943843746564,
+     "1": 0.8230959020613857,
+     "4": 0.8730856192486478,
+     "16": 0.9128257970056239,
+     "64": 0.9385945650790803,
+     "256": 0.9609594290160652,
+     "1024": 0.9779181723129399,
+     "4096": 0.9915613576224807
+    },
+    "psnr_k": {
+     "0": 17.40256665932289,
+     "1": 18.351498955738435,
+     "4": 19.793769538474923,
+     "16": 21.42499803469967,
+     "64": 22.94680977199051,
+     "256": 24.913716281504836,
+     "1024": 27.388527723303948,
+     "4096": 31.56615206677351
+    },
+    "bpp_at_psnr": {
+     "30": 0.608150897347458,
+     "35": 1.436937620461093,
+     "40": 2.2673638630468584,
+     "45": 3.09783919923912
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 0.0,
+     "1": 0.05549927217536288,
+     "4": 0.4737804354065689,
+     "16": 0.6241773312815655,
+     "64": 0.745390083498388,
+     "256": 0.8469625248188017,
+     "1024": 0.9295371226987404,
+     "4096": 0.9812700218279408
+    },
+    "captured_incl_dc": {
+     "0": 0.7804483098827146,
+     "1": 0.7926332688890948,
+     "4": 0.8844676052207302,
+     "16": 0.917487497898479,
+     "64": 0.9440999625114502,
+     "256": 0.9664003636726858,
+     "1024": 0.9845297561979816,
+     "4096": 0.9958878016364645
+    },
+    "psnr_k": {
+     "0": 17.413510042529833,
+     "1": 17.661487073311797,
+     "4": 20.201840133405994,
+     "16": 21.66368032101005,
+     "64": 23.35475689781926,
+     "256": 25.56553212183551,
+     "1024": 28.933906309407384,
+     "4096": 34.68813733004802
+    },
+    "bpp_at_psnr": {
+     "30": 0.33345095061118823,
+     "35": 1.0358337600765688,
+     "40": 1.8635722329297526,
+     "45": 2.693964143584783
+    }
+   }
+  },
+  "TOS": {
+   "n_test": 8810,
+   "E_ac": 14063003.21605006,
+   "E_raw": 78225338.71747088,
+   "catalogue": {
+    "captured_ac": {
+     "0": 0.028687251898302124,
+     "1": 0.0843958002160704,
+     "4": 0.26814772475787785,
+     "16": 0.4307496003028296,
+     "64": 0.5799407532148573,
+     "256": 0.7303867541780437,
+     "1024": 0.8768942335938527,
+     "4096": 0.9644243701462901
+    },
+    "captured_incl_dc": {
+     "0": 0.8253817174294464,
+     "1": 0.8353967522889795,
+     "4": 0.8684308554089406,
+     "16": 0.8976626968584648,
+     "64": 0.9244836182837558,
+     "256": 0.951530130706394,
+     "1024": 0.977868593255483,
+     "4096": 0.9936043716109344
+    },
+    "psnr_k": {
+     "0": 16.421856621470774,
+     "1": 16.678369747943144,
+     "4": 17.651213235260855,
+     "16": 18.74241406640651,
+     "64": 20.062342015702793,
+     "256": 21.988035260368598,
+     "1024": 25.392663546391375,
+     "4096": 30.78392153014555
+    },
+    "bpp_at_psnr": {
+     "30": 0.8633671634501457,
+     "35": 1.6881740219118297,
+     "40": 2.5185659325668603,
+     "45": 3.348957843221891
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": 0.022599939864782814,
+     "1": 0.09543179664531054,
+     "4": 0.2473600854825755,
+     "16": 0.3925290623451123,
+     "64": 0.5342546392429248,
+     "256": 0.672588985683999,
+     "1024": 0.7941890125293335,
+     "4096": 0.9170139996437741
+    },
+    "captured_incl_dc": {
+     "0": 0.8242873675664989,
+     "1": 0.8373807546061476,
+     "4": 0.8646937461451699,
+     "16": 0.8907915786628654,
+     "64": 0.91627037206137,
+     "256": 0.9411395051426248,
+     "1024": 0.9630002167308992,
+     "4096": 0.9850811462240932
+    },
+    "psnr_k": {
+     "0": 16.39472389453862,
+     "1": 16.731034331542645,
+     "4": 17.52957504468135,
+     "16": 18.460192454157337,
+     "64": 19.61396213240328,
+     "256": 21.14451465600378,
+     "1024": 23.160761944801813,
+     "4096": 27.105399172956655
+    },
+    "bpp_at_psnr": {
+     "30": 1.3075346663102485,
+     "35": 2.1380040947015195,
+     "40": 2.96848611842336,
+     "45": 3.798968142145201
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 4.440892098500626e-16,
+     "1": 0.08283886285495523,
+     "4": 0.2695252609497928,
+     "16": 0.4374284258770502,
+     "64": 0.6008655674525064,
+     "256": 0.7563999936538276,
+     "1024": 0.88844480353178,
+     "4096": 0.9751811820897526
+    },
+    "captured_incl_dc": {
+     "0": 0.8202244509692456,
+     "1": 0.8351168530200785,
+     "4": 0.8686785027341519,
+     "16": 0.8988633863929509,
+     "64": 0.9282453882516957,
+     "256": 0.9562066751152216,
+     "1024": 0.979945103307692,
+     "4096": 0.995538183383891
+    },
+    "psnr_k": {
+     "0": 16.295447508662853,
+     "1": 16.67099106772009,
+     "4": 17.65939548913276,
+     "16": 18.793669667154408,
+     "64": 20.284255557298803,
+     "256": 22.428674555913766,
+     "1024": 25.820549454538746,
+     "4096": 32.34763658122679
+    },
+    "bpp_at_psnr": {
+     "30": 0.6907528949722059,
+     "35": 1.4870032592993783,
+     "40": 2.3173948521772214,
+     "45": 3.147786762832252
+    }
+   }
+  }
+ },
+ "block8": {
+  "ED": {
+   "n_test": 14126400,
+   "E_ac": 13144.987549977348,
+   "E_raw": 438498.20152799564,
+   "catalogue": {
+    "captured_ac": {
+     "0": -0.00012598746570624542,
+     "1": 0.31305251169376036,
+     "2": 0.4697790763389622,
+     "4": 0.6235407225261249,
+     "8": 0.7495005342104548,
+     "16": 0.8706134177767083,
+     "32": 0.9647156315698868
+    },
+    "captured_incl_dc": {
+     "0": 0.9700189336972544,
+     "1": 0.9794071762442629,
+     "2": 0.9841054138512427,
+     "4": 0.9887147712391435,
+     "8": 0.9924907049844096,
+     "16": 0.9961213409621004,
+     "32": 0.9989422702713342
+    },
+    "psnr_k": {
+     "0": 25.004454435282128,
+     "1": 26.63576615787378,
+     "2": 27.76043293810879,
+     "4": 29.24782151367928,
+     "8": 31.016933517069056,
+     "16": 33.88610914715677,
+     "32": 39.52917807524446
+    },
+    "bpp_at_psnr": {
+     "30": 0.16932019610566595,
+     "35": 0.6404959137002135,
+     "40": 1.2975428396053785,
+     "45": 2.061639479346424
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": 9.633856979274213e-06,
+     "1": 0.1560918845251601,
+     "2": 0.4685854288885952,
+     "4": 0.5767272379910637,
+     "8": 0.7450145277463348,
+     "16": 0.8695058520690988,
+     "32": 0.9690788013127495
+    },
+    "captured_incl_dc": {
+     "0": 0.9700229992569124,
+     "1": 0.9747019220772485,
+     "2": 0.9840696315360145,
+     "4": 0.9873114298588593,
+     "8": 0.9923562266697563,
+     "16": 0.9960881391441988,
+     "32": 0.9990730662740281
+    },
+    "psnr_k": {
+     "0": 25.00504339693983,
+     "1": 25.74204992428376,
+     "2": 27.750666973763604,
+     "4": 28.738798336185805,
+     "8": 30.93984718445421,
+     "16": 33.84909119764423,
+     "32": 40.10243834365702
+    },
+    "bpp_at_psnr": {
+     "30": 0.17216654033973175,
+     "35": 0.639574752463091,
+     "40": 1.2904710640006107,
+     "45": 2.0495577324027567
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 2.220446049250313e-16,
+     "1": 0.31273728937267076,
+     "2": 0.4678862726481875,
+     "4": 0.6202041423994015,
+     "8": 0.7491583453800501,
+     "16": 0.8678650686190337,
+     "32": 0.9639101491145982
+    },
+    "captured_incl_dc": {
+     "0": 0.9700227104599923,
+     "1": 0.979397726733474,
+     "2": 0.9840486727269621,
+     "4": 0.9886147496106114,
+     "8": 0.9924804470907632,
+     "16": 0.9960389529036437,
+     "32": 0.9989181240905526
+    },
+    "psnr_k": {
+     "0": 25.00500155742904,
+     "1": 26.63377375064045,
+     "2": 27.744956928410282,
+     "4": 29.209499321270886,
+     "8": 31.01100498905323,
+     "16": 33.79482512234556,
+     "32": 39.43115068080208
+    },
+    "bpp_at_psnr": {
+     "30": 0.17067368603858002,
+     "35": 0.64103391747354,
+     "40": 1.2962214963254604,
+     "45": 2.058729772415838
+    }
+   }
+  },
+  "BBB": {
+   "n_test": 12884400,
+   "E_ac": 19182.39972338642,
+   "E_raw": 1067205.8204713815,
+   "catalogue": {
+    "captured_ac": {
+     "0": -1.1235342238569146e-05,
+     "1": 0.2256203431420999,
+     "2": 0.36208838471072935,
+     "4": 0.4883569963494865,
+     "8": 0.5983229049451074,
+     "16": 0.7416044469741845,
+     "32": 0.8938457359875721
+    },
+    "captured_incl_dc": {
+     "0": 0.9820253835986957,
+     "1": 0.986080979103969,
+     "2": 0.9885339122426591,
+     "4": 0.9908035166006093,
+     "8": 0.9927800894173644,
+     "16": 0.9953554912372036,
+     "32": 0.9980919392627292
+    },
+    "psnr_k": {
+     "0": 23.363525188417867,
+     "1": 24.474034628330827,
+     "2": 25.315968882726597,
+     "4": 26.273903584220044,
+     "8": 27.324803307192415,
+     "16": 29.240723630598595,
+     "32": 33.10419954635846
+    },
+    "bpp_at_psnr": {
+     "30": 0.5538625699034593,
+     "35": 1.3436016459227904,
+     "40": 2.161107388023978,
+     "45": 2.9786131301251686
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -4.707474309251225e-06,
+     "1": 0.22423403109721518,
+     "2": 0.3608126181975032,
+     "4": 0.45476272115661376,
+     "8": 0.5936541929454331,
+     "16": 0.7100929955276643,
+     "32": 0.8882348839644416
+    },
+    "captured_incl_dc": {
+     "0": 0.9820255009332994,
+     "1": 0.9860560609567136,
+     "2": 0.9885109810866056,
+     "4": 0.9901996791750577,
+     "8": 0.9926961720529224,
+     "16": 0.9947890913489006,
+     "32": 0.9979910874830332
+    },
+    "psnr_k": {
+     "0": 23.36355353836208,
+     "1": 24.46626674095568,
+     "2": 25.307292052779133,
+     "4": 25.997718565968725,
+     "8": 27.27461615487932,
+     "16": 28.740986897470297,
+     "32": 32.880511249195365
+    },
+    "bpp_at_psnr": {
+     "30": 0.5616340962889035,
+     "35": 1.351121983954208,
+     "40": 2.1686277260553957,
+     "45": 2.9861334681565865
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 3.3306690738754696e-16,
+     "1": 0.22403517338754753,
+     "2": 0.3596120985167459,
+     "4": 0.48447880518567044,
+     "8": 0.5950433197109449,
+     "16": 0.7380106559011609,
+     "32": 0.8923328548955758
+    },
+    "captured_incl_dc": {
+     "0": 0.9820255855473936,
+     "1": 0.9860524866058229,
+     "2": 0.9884894024483052,
+     "4": 0.9907338083853044,
+     "8": 0.9927211407931329,
+     "16": 0.995290894947001,
+     "32": 0.9980647461109642
+    },
+    "psnr_k": {
+     "0": 23.363573982615122,
+     "1": 24.465153625114773,
+     "2": 25.299142798373218,
+     "4": 26.24110872964604,
+     "8": 27.289488306460058,
+     "16": 29.18073770656217,
+     "32": 33.04274200782427
+    },
+    "bpp_at_psnr": {
+     "30": 0.5616210954110952,
+     "35": 1.3522118345526992,
+     "40": 2.1697175766538868,
+     "45": 2.9872233187550776
+    }
+   }
+  },
+  "SIN": {
+   "n_test": 19180800,
+   "E_ac": 17852.906596492325,
+   "E_raw": 357908.9512909435,
+   "catalogue": {
+    "captured_ac": {
+     "0": 2.1552878865405134e-05,
+     "1": 0.16911384998175538,
+     "2": 0.3200276536422608,
+     "4": 0.44421288428034034,
+     "8": 0.5583428312102383,
+     "16": 0.7371770352984364,
+     "32": 0.8976753770068288
+    },
+    "captured_incl_dc": {
+     "0": 0.9501199348310047,
+     "1": 0.9585544514181862,
+     "2": 0.9660822040244132,
+     "4": 0.972276704930995,
+     "8": 0.9779696368206637,
+     "16": 0.9868900908364933,
+     "32": 0.9948959311293296
+    },
+    "psnr_k": {
+     "0": 23.675607624168215,
+     "1": 24.48009882214036,
+     "2": 25.35060151178716,
+     "4": 26.22642927294524,
+     "8": 27.224661179076676,
+     "16": 29.478880920929193,
+     "32": 33.575712488223594
+    },
+    "bpp_at_psnr": {
+     "30": 0.5172372666496439,
+     "35": 1.2961575047739062,
+     "40": 2.1136632468750944,
+     "45": 2.931168988976285
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -2.5177368825124447e-05,
+     "1": 0.1509325193502531,
+     "2": 0.3192692156303746,
+     "4": 0.420765340871488,
+     "8": 0.5544571141779762,
+     "16": 0.6949574210094256,
+     "32": 0.8760282255390268
+    },
+    "captured_incl_dc": {
+     "0": 0.9501176038729656,
+     "1": 0.9576475459150033,
+     "2": 0.9660443722715705,
+     "4": 0.9711071147296727,
+     "8": 0.9777758128244011,
+     "16": 0.9847841283347929,
+     "32": 0.9938161465309259
+    },
+    "psnr_k": {
+     "0": 23.675404677648864,
+     "1": 24.386091943325194,
+     "2": 25.34576011074823,
+     "4": 26.04696861575807,
+     "8": 27.186618885300895,
+     "16": 28.831909379842372,
+     "32": 32.74228584528303
+    },
+    "bpp_at_psnr": {
+     "30": 0.5219450944830686,
+     "35": 1.302929545664895,
+     "40": 2.1204352877660835,
+     "45": 2.9379410298672735
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 2.220446049250313e-16,
+     "1": 0.16787742781796477,
+     "2": 0.31818565102328333,
+     "4": 0.44094730563330886,
+     "8": 0.5551924252994078,
+     "16": 0.7330797031823748,
+     "32": 0.8949356784647846
+    },
+    "captured_incl_dc": {
+     "0": 0.9501188597488311,
+     "1": 0.9584927772708245,
+     "2": 0.965990322833433,
+     "4": 0.9721138141445013,
+     "8": 0.9778124909815775,
+     "16": 0.9866857112385564,
+     "32": 0.994759271842108
+    },
+    "psnr_k": {
+     "0": 23.67551402019589,
+     "1": 24.473640991645436,
+     "2": 25.338852651797602,
+     "4": 26.200986571132674,
+     "8": 27.19379227711192,
+     "16": 29.411698029014538,
+     "32": 33.460961416623604
+    },
+    "bpp_at_psnr": {
+     "30": 0.5225141580308293,
+     "35": 1.302557275666151,
+     "40": 2.1200630177673387,
+     "45": 2.9375687598685296
+    }
+   }
+  },
+  "TOS": {
+   "n_test": 15858000,
+   "E_ac": 27433.95603241266,
+   "E_raw": 562542.6054273327,
+   "catalogue": {
+    "captured_ac": {
+     "0": 2.052061978163877e-05,
+     "1": 0.2581447142738461,
+     "2": 0.43117374827497346,
+     "4": 0.5752073943890434,
+     "8": 0.6934169232034333,
+     "16": 0.8198383140138931,
+     "32": 0.9377533711320726
+    },
+    "captured_incl_dc": {
+     "0": 0.9512332171715382,
+     "1": 0.9638213619827712,
+     "2": 0.9722596044648883,
+     "4": 0.9792838061458966,
+     "8": 0.9850486228634547,
+     "16": 0.9912139138895736,
+     "32": 0.9969643698751823
+    },
+    "psnr_k": {
+     "0": 21.809808086647486,
+     "1": 23.10652701194727,
+     "2": 24.259922654063665,
+     "4": 25.52794948230907,
+     "8": 26.944237181984086,
+     "16": 29.25309459183672,
+     "32": 33.86856060551771
+    },
+    "bpp_at_psnr": {
+     "30": 0.5928856922892405,
+     "35": 1.3463178060206469,
+     "40": 2.1625269692994724,
+     "45": 2.980032711400663
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": 1.5562447179817696e-05,
+     "1": 0.17291821654059603,
+     "2": 0.4306927062771383,
+     "4": 0.5724291725998514,
+     "8": 0.6911378531602098,
+     "16": 0.8060256531579615,
+     "32": 0.927991340381091
+    },
+    "captured_incl_dc": {
+     "0": 0.9512329753724498,
+     "1": 0.9596650545865796,
+     "2": 0.9722361451128481,
+     "4": 0.9791483184269618,
+     "8": 0.9849374776688378,
+     "16": 0.9905403010343758,
+     "32": 0.9964882972722796
+    },
+    "psnr_k": {
+     "0": 21.809786553188975,
+     "1": 22.63423441020333,
+     "2": 24.2562514873803,
+     "4": 25.499638303798505,
+     "8": 26.91207210765974,
+     "16": 28.932275984120487,
+     "32": 33.23587169614494
+    },
+    "bpp_at_psnr": {
+     "30": 0.5994270122667162,
+     "35": 1.3537073258649635,
+     "40": 2.1697630467098152,
+     "45": 2.9872687888110057
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 1.1102230246251565e-16,
+     "1": 0.25646063542217323,
+     "2": 0.4286665013041947,
+     "4": 0.5710358073133768,
+     "8": 0.6882062682552115,
+     "16": 0.8160165669586802,
+     "32": 0.9357844239944428
+    },
+    "captured_incl_dc": {
+     "0": 0.9512322164263939,
+     "1": 0.9637392331898119,
+     "2": 0.9721373315872518,
+     "4": 0.979080367090232,
+     "8": 0.9847945107706632,
+     "16": 0.9910275357563119,
+     "32": 0.9968683486873066
+    },
+    "psnr_k": {
+     "0": 21.809718965813715,
+     "1": 23.096679306280656,
+     "2": 24.240822080149066,
+     "4": 25.485508551370692,
+     "8": 26.87104516633017,
+     "16": 29.161931782602288,
+     "32": 33.7333151410759
+    },
+    "bpp_at_psnr": {
+     "30": 0.6033097649024612,
+     "35": 1.358702275961316,
+     "40": 2.174897465178968,
+     "45": 2.9924032072801587
+    }
+   }
+  }
+ },
+ "block16": {
+  "ED": {
+   "n_test": 3453120,
+   "E_ac": 80389.23872063526,
+   "E_raw": 1747809.4507380123,
+   "catalogue": {
+    "captured_ac": {
+     "0": -0.0002589618527468751,
+     "1": 0.14767887390735568,
+     "4": 0.5102895877206634,
+     "16": 0.7542465768485815,
+     "64": 0.9135963216755354,
+     "128": 0.9797083236258057
+    },
+    "captured_incl_dc": {
+     "0": 0.9539938084023507,
+     "1": 0.9607981027661964,
+     "4": 0.9774761217705503,
+     "16": 0.9886967480397874,
+     "64": 0.9960259249540998,
+     "128": 0.9990666989382635
+    },
+    "psnr_k": {
+     "0": 23.16009959086354,
+     "1": 23.85519156996329,
+     "4": 26.261830720945625,
+     "16": 29.256228343043595,
+     "64": 33.795901788392854,
+     "128": 40.088044829610396
+    },
+    "bpp_at_psnr": {
+     "30": 0.17841581192715456,
+     "35": 0.6511987977617966,
+     "40": 1.3130473399074474,
+     "45": 2.091451185957633
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": 8.907993382267776e-06,
+     "1": 0.1479924435440083,
+     "4": 0.5014287111919149,
+     "16": 0.7518748996343279,
+     "64": 0.9174318651370525,
+     "128": 0.9790033126900115
+    },
+    "captured_incl_dc": {
+     "0": 0.9540061288832803,
+     "1": 0.9608125251761255,
+     "4": 0.9770685721270563,
+     "16": 0.9885876644519486,
+     "64": 0.9962023380172178,
+     "128": 0.9990342724672918
+    },
+    "psnr_k": {
+     "0": 23.16126278939977,
+     "1": 23.85678963688338,
+     "4": 26.18395145030209,
+     "16": 29.21451710452661,
+     "64": 33.993099358304015,
+     "128": 39.93971629614468
+    },
+    "bpp_at_psnr": {
+     "30": 0.18256025339967802,
+     "35": 0.6522405802460732,
+     "40": 1.3111446813654175,
+     "45": 2.0770866934905365
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 0.0,
+     "1": 0.14688697190164723,
+     "4": 0.5038589758852698,
+     "16": 0.7514719993864678,
+     "64": 0.9123850503584556,
+     "128": 0.9794146005416108
+    },
+    "captured_incl_dc": {
+     "0": 0.9540057191665311,
+     "1": 0.9607616798029532,
+     "4": 0.9771803504038622,
+     "16": 0.9885691333448007,
+     "64": 0.9959702134009766,
+     "128": 0.9990531893562417
+    },
+    "psnr_k": {
+     "0": 23.16122410230375,
+     "1": 23.851158360059994,
+     "4": 26.205172714626052,
+     "16": 29.207470842560582,
+     "64": 33.735441945544224,
+     "128": 40.02563111357838
+    },
+    "bpp_at_psnr": {
+     "30": 0.17849239825202723,
+     "35": 0.6436727529116719,
+     "40": 1.2999317072184429,
+     "45": 2.07327965171011
+    }
+   }
+  },
+  "BBB": {
+   "n_test": 3149520,
+   "E_ac": 105981.2557151566,
+   "E_raw": 4276538.009773941,
+   "catalogue": {
+    "captured_ac": {
+     "0": -4.76850897244141e-05,
+     "1": 0.13303659332363227,
+     "4": 0.4024860974544443,
+     "16": 0.6336028454212617,
+     "64": 0.8125758637627987,
+     "128": 0.9257404519204362
+    },
+    "captured_incl_dc": {
+     "0": 0.9752167970450362,
+     "1": 0.9785148944569984,
+     "4": 0.9851923977866411,
+     "16": 0.9909199379395279,
+     "64": 0.9953552510782437,
+     "128": 0.9981596983035518
+    },
+    "psnr_k": {
+     "0": 21.96070556370582,
+     "1": 22.580904983531806,
+     "4": 24.19743250642928,
+     "16": 26.32139172947638,
+     "64": 29.232657472147817,
+     "128": 33.25338963330678
+    },
+    "bpp_at_psnr": {
+     "30": 0.5564171829411355,
+     "35": 1.3546244117896262,
+     "40": 2.1817799598095666,
+     "45": 3.0090253422059687
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -2.2609951574770193e-05,
+     "1": 0.15512097184751783,
+     "4": 0.3896405753980803,
+     "16": 0.6317413276878308,
+     "64": 0.7880743232968203,
+     "128": 0.9193035139656446
+    },
+    "captured_incl_dc": {
+     "0": 0.975217418457642,
+     "1": 0.9790621899942532,
+     "4": 0.9848740597864319,
+     "16": 0.9908738057675528,
+     "64": 0.9947480533824428,
+     "128": 0.9980001779705511
+    },
+    "psnr_k": {
+     "0": 21.96081445981971,
+     "1": 22.692967351941938,
+     "4": 24.105056103164173,
+     "16": 26.29938282189788,
+     "64": 28.69907686664133,
+     "128": 32.89236641667559
+    },
+    "bpp_at_psnr": {
+     "30": 0.5797794320466126,
+     "35": 1.3788532172052057,
+     "40": 2.2060489321440655,
+     "45": 3.033294314547298
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 0.0,
+     "1": 0.13179988624992167,
+     "4": 0.3942926882837616,
+     "16": 0.6302206382794244,
+     "64": 0.8094621470857942,
+     "128": 0.9249296503538836
+    },
+    "captured_incl_dc": {
+     "0": 0.9752179787779417,
+     "1": 0.9784842463560521,
+     "4": 0.9849893485466923,
+     "16": 0.9908361200103616,
+     "64": 0.9952780868854747,
+     "128": 0.9981396050019226
+    },
+    "psnr_k": {
+     "0": 21.960912652481696,
+     "1": 22.574714268449725,
+     "4": 24.138284490244427,
+     "16": 26.28148596772747,
+     "64": 29.16109998209183,
+     "128": 33.20622826616028
+    },
+    "bpp_at_psnr": {
+     "30": 0.5559052653161101,
+     "35": 1.3495280163563308,
+     "40": 2.1766448075870013,
+     "45": 3.0038901899831902
+    }
+   }
+  },
+  "SIN": {
+   "n_test": 4688640,
+   "E_ac": 93402.34059172809,
+   "E_raw": 1447403.9394459343,
+   "catalogue": {
+    "captured_ac": {
+     "0": 5.60015355263177e-05,
+     "1": 0.13787334732292322,
+     "4": 0.35216609190578985,
+     "16": 0.5792889239308272,
+     "64": 0.7979356940188536,
+     "128": 0.9276185616295737
+    },
+    "captured_incl_dc": {
+     "0": 0.9354726711929597,
+     "1": 0.944366154428614,
+     "4": 0.9581946672372271,
+     "16": 0.9728511176826231,
+     "64": 0.9869605998613581,
+     "128": 0.9953291575523957
+    },
+    "psnr_k": {
+     "0": 22.509868885622716,
+     "1": 23.153914951929924,
+     "4": 24.394988911677544,
+     "16": 26.269786208760934,
+     "64": 29.454729631759555,
+     "128": 33.913353575266015
+    },
+    "bpp_at_psnr": {
+     "30": 0.5149778441800431,
+     "35": 1.2976487664116247,
+     "40": 2.1248864300960055,
+     "45": 2.9521308379762154
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -0.00011726405881407054,
+     "1": 0.13846135432776363,
+     "4": 0.32613849549579654,
+     "16": 0.5779232171279136,
+     "64": 0.7614672133777981,
+     "128": 0.9072011476298496
+    },
+    "captured_incl_dc": {
+     "0": 0.93546149020083,
+     "1": 0.9444040990749146,
+     "4": 0.9565150819062649,
+     "16": 0.9727629873352628,
+     "64": 0.9846072544289817,
+     "128": 0.9940116025807445
+    },
+    "psnr_k": {
+     "0": 22.50911642575437,
+     "1": 23.156878035397153,
+     "4": 24.223919194612268,
+     "16": 26.25571103097405,
+     "64": 28.734144849385643,
+     "128": 32.83419961327714
+    },
+    "bpp_at_psnr": {
+     "30": 0.5431380082355126,
+     "35": 1.3345246452297417,
+     "40": 2.161763825911266,
+     "45": 2.9890082338026427
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": -4.440892098500626e-16,
+     "1": 0.13792072456406512,
+     "4": 0.3445374307641117,
+     "16": 0.5751501275895635,
+     "64": 0.7928716752438977,
+     "128": 0.9258976005166406
+    },
+    "captured_incl_dc": {
+     "0": 0.935469057361083,
+     "1": 0.9443692117266446,
+     "4": 0.9577023825426817,
+     "16": 0.9725840372533309,
+     "64": 0.986633813956269,
+     "128": 0.9952181023095332
+    },
+    "psnr_k": {
+     "0": 22.509625667233777,
+     "1": 23.154153620340296,
+     "4": 24.344146707559386,
+     "16": 26.227270745602127,
+     "64": 29.347230740871005,
+     "128": 33.811302957816935
+    },
+    "bpp_at_psnr": {
+     "30": 0.514363460240496,
+     "35": 1.2963399017370294,
+     "40": 2.123576934462687,
+     "45": 2.950821342342521
+    }
+   }
+  },
+  "TOS": {
+   "n_test": 3876400,
+   "E_ac": 154511.10271485787,
+   "E_raw": 2259549.8000883744,
+   "catalogue": {
+    "captured_ac": {
+     "0": 6.104865553313576e-05,
+     "1": 0.1590037427895049,
+     "4": 0.4468662022264146,
+     "16": 0.7054093779567226,
+     "64": 0.8757404335017762,
+     "128": 0.9584516752467819
+    },
+    "captured_incl_dc": {
+     "0": 0.9316228082188189,
+     "1": 0.942491526818493,
+     "4": 0.9621759551218959,
+     "16": 0.9798554907444029,
+     "64": 0.9915029610581011,
+     "128": 0.9971588687829214
+    },
+    "psnr_k": {
+     "0": 20.32387148072939,
+     "1": 21.075665711615407,
+     "4": 22.895304384770718,
+     "16": 25.63141716712611,
+     "64": 29.38030800003812,
+     "128": 34.13807116641328
+    },
+    "bpp_at_psnr": {
+     "30": 0.584375674660791,
+     "35": 1.3376840983945357,
+     "40": 2.163074813844502,
+     "45": 2.990324123436083
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": 5.3124907082313655e-05,
+     "1": 0.15887444137536588,
+     "4": 0.4394567154637291,
+     "16": 0.702162893738006,
+     "64": 0.8640467193794393,
+     "128": 0.9514309734602544
+    },
+    "captured_incl_dc": {
+     "0": 0.931622266382073,
+     "1": 0.9424826850111203,
+     "4": 0.9616692842929507,
+     "16": 0.979633491717623,
+     "64": 0.9907033289084561,
+     "128": 0.9966787836018707
+    },
+    "psnr_k": {
+     "0": 20.323837066362508,
+     "1": 21.07499804426522,
+     "4": 22.837514802010244,
+     "16": 25.583818304747407,
+     "64": 28.989709423766765,
+     "128": 33.46001235095229
+    },
+    "bpp_at_psnr": {
+     "30": 0.604116032458154,
+     "35": 1.3597240996964353,
+     "40": 2.184709605298412,
+     "45": 3.0119589149371873
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 0.0,
+     "1": 0.15779140782374845,
+     "4": 0.43601003504176095,
+     "16": 0.7016297932216005,
+     "64": 0.8733178896939792,
+     "128": 0.9578644564476596
+    },
+    "captured_incl_dc": {
+     "0": 0.9316186336283384,
+     "1": 0.9424086256970344,
+     "4": 0.9614335955762501,
+     "16": 0.9795970375758978,
+     "64": 0.9913373042024287,
+     "128": 0.9971187139590783
+    },
+    "psnr_k": {
+     "0": 20.323606341693864,
+     "1": 21.069409664086987,
+     "4": 22.81089257494616,
+     "16": 25.57605178899987,
+     "64": 29.29645344810321,
+     "128": 34.07712033437713
+    },
+    "bpp_at_psnr": {
+     "30": 0.5910189079691692,
+     "35": 1.3437099184598535,
+     "40": 2.1686876376444344,
+     "45": 2.9959369472359665
+    }
+   }
+  }
+ },
+ "block8_diff": {
+  "ED": {
+   "n_test": 14122800,
+   "E_ac": 10185.503768630253,
+   "E_raw": 22467.01266484661,
+   "catalogue": {
+    "captured_ac": {
+     "0": -5.312089434994505e-09,
+     "1": 0.22947894473474506,
+     "2": 0.2924072202045669,
+     "4": 0.4531997972407469,
+     "8": 0.6068094852143384,
+     "16": 0.76018795944459,
+     "32": 0.9176730465725789
+    },
+    "captured_incl_dc": {
+     "0": 0.5466462776035428,
+     "1": 0.6506814132662618,
+     "2": 0.679210181042948,
+     "4": 0.7521060939887896,
+     "8": 0.8217456174578442,
+     "16": 0.8912803193162433,
+     "32": 0.9626767694083832
+    },
+    "psnr_k": {
+     "0": 26.112778187446146,
+     "1": 27.244933103001372,
+     "2": 27.61494428287936,
+     "4": 28.7344915411486,
+     "8": 30.16674788559721,
+     "16": 32.31406836552773,
+     "32": 36.95735776745909
+    },
+    "bpp_at_psnr": {
+     "30": 0.19730087967027532,
+     "35": 0.7848759311802959,
+     "40": 1.5084937219995411,
+     "45": 2.298611583818653
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -1.3039867852171483e-08,
+     "1": 0.0712127201152194,
+     "2": 0.2889177760738394,
+     "4": 0.3554671644975651,
+     "8": 0.5345857938246328,
+     "16": 0.7378527634672578,
+     "32": 0.9281609508424444
+    },
+    "captured_incl_dc": {
+     "0": 0.5466462741001257,
+     "1": 0.5789308315865116,
+     "2": 0.6776282285655917,
+     "4": 0.7077986413704274,
+     "8": 0.7890027382950413,
+     "16": 0.8811545751332542,
+     "32": 0.9674314998240069
+    },
+    "psnr_k": {
+     "0": 26.11277815388483,
+     "1": 26.433615621179186,
+     "2": 27.593579990380952,
+     "4": 28.020327738164017,
+     "8": 29.434381856131417,
+     "16": 31.92732537162886,
+     "32": 37.54917245559967
+    },
+    "bpp_at_psnr": {
+     "30": 0.2081621962373322,
+     "35": 0.7804418115010353,
+     "40": 1.504618865354372,
+     "45": 2.293725414462422
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 3.3306690738754696e-16,
+     "1": 0.22559969871550145,
+     "2": 0.2881688672754291,
+     "4": 0.4487174813564685,
+     "8": 0.6044673369928915,
+     "16": 0.7670335151089417,
+     "32": 0.9138973200946711
+    },
+    "captured_incl_dc": {
+     "0": 0.5466462800117984,
+     "1": 0.6489227426526883,
+     "2": 0.6772887079759005,
+     "4": 0.7500740194084898,
+     "8": 0.8206837958488875,
+     "16": 0.8943837774420635,
+     "32": 0.9609650297639657
+    },
+    "psnr_k": {
+     "0": 26.112778210516257,
+     "1": 27.223123077867093,
+     "2": 27.589008426172764,
+     "4": 28.699035999799797,
+     "8": 30.140954677867768,
+     "16": 32.43984374162142,
+     "32": 36.762611521765564
+    },
+    "bpp_at_psnr": {
+     "30": 0.19710262494714753,
+     "35": 0.7630644665968758,
+     "40": 1.4879761695838425,
+     "45": 2.277899763671396
+    }
+   }
+  },
+  "BBB": {
+   "n_test": 12880800,
+   "E_ac": 18168.999749239178,
+   "E_raw": 25753.305597189712,
+   "catalogue": {
+    "captured_ac": {
+     "0": -5.5088797967783876e-09,
+     "1": 0.17037047961007445,
+     "2": 0.21528767379410918,
+     "4": 0.3344891995802962,
+     "8": 0.4446236223296387,
+     "16": 0.582696689514964,
+     "32": 0.8273336912259881
+    },
+    "captured_incl_dc": {
+     "0": 0.29449833999901454,
+     "1": 0.414694999403461,
+     "2": 0.44638415428831235,
+     "4": 0.530481028141838,
+     "8": 0.6081810457867094,
+     "16": 0.7055918233507597,
+     "32": 0.8781836332047637
+    },
+    "psnr_k": {
+     "0": 23.599293135356156,
+     "1": 24.410451189377152,
+     "2": 24.65218841398182,
+     "4": 25.36774207996258,
+     "8": 26.153419123640283,
+     "16": 27.3947748592599,
+     "32": 31.227217110818618
+    },
+    "bpp_at_psnr": {
+     "30": 0.7117460746254448,
+     "35": 1.4947980453912015,
+     "40": 2.309756848971706,
+     "45": 3.1272625910728964
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -1.4048873619998403e-08,
+     "1": 0.13422500361340184,
+     "2": 0.21113490487600162,
+     "4": 0.2884913592050715,
+     "8": 0.39964043379884995,
+     "16": 0.5586333677772389,
+     "32": 0.7948434650922921
+    },
+    "captured_incl_dc": {
+     "0": 0.2944983339740347,
+     "1": 0.3891943062267629,
+     "2": 0.4434543689391268,
+     "4": 0.49802947557942834,
+     "8": 0.5764453317809288,
+     "16": 0.6886151100131795,
+     "32": 0.8552617248599077
+    },
+    "psnr_k": {
+     "0": 23.59929309826743,
+     "1": 24.225242767102856,
+     "2": 24.629265756497077,
+     "4": 25.077491372607472,
+     "8": 25.815178808029657,
+     "16": 27.151298189767154,
+     "32": 30.478439607028797
+    },
+    "bpp_at_psnr": {
+     "30": 0.7282754513124126,
+     "35": 1.5078695840281278,
+     "40": 2.322231012245793,
+     "45": 3.139736754346984
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 4.440892098500626e-16,
+     "1": 0.16607064960768192,
+     "2": 0.2104151581808248,
+     "4": 0.32220708023106237,
+     "8": 0.4389422312395721,
+     "16": 0.5767902226955935,
+     "32": 0.8165181815608543
+    },
+    "captured_incl_dc": {
+     "0": 0.29449834388553864,
+     "1": 0.4116614622157624,
+     "2": 0.44294658645369667,
+     "4": 0.5218159726003581,
+     "8": 0.6041728149636334,
+     "16": 0.7014248012279087,
+     "32": 0.8705532732242898
+    },
+    "psnr_k": {
+     "0": 23.599293159280915,
+     "1": 24.38800056698545,
+     "2": 24.62530513624006,
+     "4": 25.28832288047182,
+     "8": 26.109217356661873,
+     "16": 27.333736233936335,
+     "32": 30.9633628027056
+    },
+    "bpp_at_psnr": {
+     "30": 0.7193031849342499,
+     "35": 1.501685347601065,
+     "40": 2.316702825894996,
+     "45": 3.134208567996187
+    }
+   }
+  },
+  "SIN": {
+   "n_test": 19177200,
+   "E_ac": 18788.44572480928,
+   "E_raw": 25091.490426384426,
+   "catalogue": {
+    "captured_ac": {
+     "0": -8.290271136246474e-09,
+     "1": 0.1551342708968645,
+     "2": 0.23877126893121603,
+     "4": 0.3209646556466311,
+     "8": 0.44445858310948005,
+     "16": 0.5873360587755155,
+     "32": 0.8423272435117617
+    },
+    "captured_incl_dc": {
+     "0": 0.25120247696350484,
+     "1": 0.36736663999385166,
+     "2": 0.4299938164369861,
+     "4": 0.49154002030923494,
+     "8": 0.5840119665368473,
+     "16": 0.6909982655263364,
+     "32": 0.8819350314700612
+    },
+    "psnr_k": {
+     "0": 23.453694766994516,
+     "1": 24.18581786430734,
+     "2": 24.638543087916204,
+     "4": 25.13477100042725,
+     "8": 26.00653038163393,
+     "16": 27.29772958627217,
+     "32": 31.476128201284727
+    },
+    "bpp_at_psnr": {
+     "30": 0.7088876597329858,
+     "35": 1.4796886188546259,
+     "40": 2.2963740637652434,
+     "45": 3.113879805866434
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -1.4062927711222528e-08,
+     "1": 0.05231382195442413,
+     "2": 0.20514237135977076,
+     "4": 0.27147853366237773,
+     "8": 0.3950095834569125,
+     "16": 0.5599085344253663,
+     "32": 0.789132530217944
+    },
+    "captured_incl_dc": {
+     "0": 0.25120247264095386,
+     "1": 0.290374943146534,
+     "2": 0.4048125814417991,
+     "4": 0.4544849350499409,
+     "8": 0.5469846783873384,
+     "16": 0.6704606034001843,
+     "32": 0.8421029622472327
+    },
+    "psnr_k": {
+     "0": 23.453694741924185,
+     "1": 23.687049337916683,
+     "2": 24.450801335717706,
+     "4": 24.829271272485975,
+     "8": 25.636209851109047,
+     "16": 27.01826533663263,
+     "32": 30.213598933712973
+    },
+    "bpp_at_psnr": {
+     "30": 0.7324888137318432,
+     "35": 1.5115879204357288,
+     "40": 2.3284986649752515,
+     "45": 3.1460044070764424
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 3.3306690738754696e-16,
+     "1": 0.15194786970651553,
+     "2": 0.2347115043201926,
+     "4": 0.31537806756925935,
+     "8": 0.4365716644860611,
+     "16": 0.5805083470149411,
+     "32": 0.8298252694556957
+    },
+    "captured_incl_dc": {
+     "0": 0.2512024831712396,
+     "1": 0.36498067069489815,
+     "2": 0.42695387477734237,
+     "4": 0.48735679702935386,
+     "8": 0.5781062614562007,
+     "16": 0.6858856919143957,
+     "32": 0.8725735843414215
+    },
+    "psnr_k": {
+     "0": 23.453694802998708,
+     "1": 24.169469308698766,
+     "2": 24.615442955217457,
+     "4": 25.09918672291657,
+     "8": 25.94530795889556,
+     "16": 27.226461566097516,
+     "32": 31.144744086871945
+    },
+    "bpp_at_psnr": {
+     "30": 0.7244092768405925,
+     "35": 1.498360627130964,
+     "40": 2.3151646405302997,
+     "45": 3.1326703826314906
+    }
+   }
+  },
+  "TOS": {
+   "n_test": 15854400,
+   "E_ac": 29896.963322480668,
+   "E_raw": 43363.72977157958,
+   "catalogue": {
+    "captured_ac": {
+     "0": 4.0887648822263145e-10,
+     "1": 0.2517079397370563,
+     "2": 0.31479380969145787,
+     "4": 0.46569649597851726,
+     "8": 0.5831646259388876,
+     "16": 0.7215369579016715,
+     "32": 0.9048612683444246
+    },
+    "captured_incl_dc": {
+     "0": 0.31055369388795384,
+     "1": 0.48409280294779855,
+     "2": 0.5275871229735094,
+     "4": 0.6316264226590467,
+     "8": 0.712614390979228,
+     "16": 0.808014684158086,
+     "32": 0.9344069528650589
+    },
+    "psnr_k": {
+     "0": 21.436332563793393,
+     "1": 22.69562119026687,
+     "2": 23.07811978375461,
+     "4": 24.158452338582684,
+     "8": 25.236686886462458,
+     "16": 26.988656928781538,
+     "32": 31.652758988872396
+    },
+    "bpp_at_psnr": {
+     "30": 0.8100566909416409,
+     "35": 1.5850847234731282,
+     "40": 2.402238364328855,
+     "45": 3.2197442822987203
+    }
+   },
+   "per_title": {
+    "captured_ac": {
+     "0": -4.25853328156478e-08,
+     "1": 0.06291655750920211,
+     "2": 0.31143234132381137,
+     "4": 0.3701489962504322,
+     "8": 0.5483397140731396,
+     "16": 0.6999462142967081,
+     "32": 0.8728537236019943
+    },
+    "captured_incl_dc": {
+     "0": 0.31055366424575515,
+     "1": 0.353931281791797,
+     "2": 0.5252695710233755,
+     "4": 0.565751551886342,
+     "8": 0.6886044841228933,
+     "16": 0.7931290257273452,
+     "32": 0.9123394693656515
+    },
+    "psnr_k": {
+     "0": 21.43633237707192,
+     "1": 21.718549918883426,
+     "2": 23.05686635724949,
+     "4": 23.443954304028015,
+     "8": 24.888213510755712,
+     "16": 26.664341456806277,
+     "32": 30.393296102388273
+    },
+    "bpp_at_psnr": {
+     "30": 0.8277772587106704,
+     "35": 1.604906237111571,
+     "40": 2.4217693137762306,
+     "45": 3.2392752317887594
+    }
+   },
+   "dct": {
+    "captured_ac": {
+     "0": 1.1102230246251565e-16,
+     "1": 0.24742743085881735,
+     "2": 0.3098167387262194,
+     "4": 0.4588592302892287,
+     "8": 0.5766068219027826,
+     "16": 0.7156071755278977,
+     "32": 0.8963978809664195
+    },
+    "captured_incl_dc": {
+     "0": 0.31055369360605556,
+     "1": 0.4811416219122102,
+     "2": 0.5241556997798653,
+     "4": 0.6269124950837326,
+     "8": 0.70809313720848,
+     "16": 0.8039264176027676,
+     "32": 0.9285719016977121
+    },
+    "psnr_k": {
+     "0": 21.436332562017668,
+     "1": 22.67084871725931,
+     "2": 23.046688338811627,
+     "4": 24.103230011577267,
+     "8": 25.16889400011531,
+     "16": 26.89714621709031,
+     "32": 31.28264617827501
+    },
+    "bpp_at_psnr": {
+     "30": 0.8313018692438423,
+     "35": 1.607575216581481,
+     "40": 2.4244244352667375,
+     "45": 3.2419303532366137
+    }
+   }
+  }
+ }
+}
\ No newline at end of file
diff --git a/wai/research/catalogue-prior/b_recur.json b/wai/research/catalogue-prior/b_recur.json
new file mode 100644
index 000000000..9371676bb
--- /dev/null
+++ b/wai/research/catalogue-prior/b_recur.json
@@ -0,0 +1,770 @@
+{
+ "ED": {
+  "frames": 15693,
+  "flat": 228,
+  "exact_thumb_cross_flat": 14,
+  "exact_thumb_cross_nonflat": 0,
+  "cross_thumb_psnr_quantiles": {
+   "0.5": 19.26856803894043,
+   "0.9": 24.309589385986328,
+   "0.99": 32.53339385986328,
+   "0.999": 38.60044479370117,
+   "1.0": 40.34019470214844
+  },
+  "cross_ge35": {
+   "frames": 83,
+   "seconds": 3.4583333333333335,
+   "full_psnr_median": 33.070213317871094,
+   "full_psnr_ge35": 13,
+   "full_psnr_ge40": 0,
+   "runs": [
+    {
+     "start_s": 0.5,
+     "len": 5,
+     "with": "SIN",
+     "at_s": 103.58333333333333
+    },
+    {
+     "start_s": 159.16666666666666,
+     "len": 1,
+     "with": "BBB",
+     "at_s": 490.25
+    },
+    {
+     "start_s": 163.375,
+     "len": 1,
+     "with": "BBB",
+     "at_s": 299.2083333333333
+    },
+    {
+     "start_s": 165.66666666666666,
+     "len": 1,
+     "with": "BBB",
+     "at_s": 299.2083333333333
+    },
+    {
+     "start_s": 428.5833333333333,
+     "len": 2,
+     "with": "SIN",
+     "at_s": 103.58333333333333
+    },
+    {
+     "start_s": 564.9166666666666,
+     "len": 10,
+     "with": "SIN",
+     "at_s": 437.6666666666667
+    },
+    {
+     "start_s": 566.25,
+     "len": 4,
+     "with": "SIN",
+     "at_s": 778.0
+    },
+    {
+     "start_s": 573.5833333333334,
+     "len": 1,
+     "with": "SIN",
+     "at_s": 778.0
+    },
+    {
+     "start_s": 574.6666666666666,
+     "len": 6,
+     "with": "SIN",
+     "at_s": 771.125
+    },
+    {
+     "start_s": 579.375,
+     "len": 1,
+     "with": "SIN",
+     "at_s": 771.125
+    },
+    {
+     "start_s": 580.4583333333334,
+     "len": 10,
+     "with": "SIN",
+     "at_s": 778.0
+    },
+    {
+     "start_s": 584.0833333333334,
+     "len": 2,
+     "with": "SIN",
+     "at_s": 778.0
+    },
+    {
+     "start_s": 584.625,
+     "len": 4,
+     "with": "SIN",
+     "at_s": 778.0
+    },
+    {
+     "start_s": 589.0833333333334,
+     "len": 3,
+     "with": "SIN",
+     "at_s": 778.0
+    },
+    {
+     "start_s": 590.0416666666666,
+     "len": 10,
+     "with": "SIN",
+     "at_s": 147.33333333333334
+    },
+    {
+     "start_s": 590.5833333333334,
+     "len": 2,
+     "with": "SIN",
+     "at_s": 147.33333333333334
+    },
+    {
+     "start_s": 641.4166666666666,
+     "len": 2,
+     "with": "SIN",
+     "at_s": 103.58333333333333
+    },
+    {
+     "start_s": 641.7083333333334,
+     "len": 3,
+     "with": "SIN",
+     "at_s": 774.5
+    },
+    {
+     "start_s": 646.0416666666666,
+     "len": 3,
+     "with": "SIN",
+     "at_s": 774.5
+    },
+    {
+     "start_s": 646.75,
+     "len": 5,
+     "with": "SIN",
+     "at_s": 771.125
+    },
+    {
+     "start_s": 653.2916666666666,
+     "len": 7,
+     "with": "SIN",
+     "at_s": 764.2916666666666
+    }
+   ]
+  },
+  "cross_ge40": {
+   "frames": 1,
+   "seconds": 0.041666666666666664,
+   "full_psnr_median": 34.72438430786133,
+   "full_psnr_ge35": 0,
+   "full_psnr_ge40": 0,
+   "runs": [
+    {
+     "start_s": 574.6666666666666,
+     "len": 1,
+     "with": "SIN",
+     "at_s": 771.125
+    }
+   ]
+  },
+  "self_far_ge35": {
+   "frames": 80,
+   "seconds": 3.3333333333333335,
+   "full_psnr_ge35": 11,
+   "full_psnr_ge40": 0,
+   "full_psnr_median": 32.833478927612305,
+   "runs": [
+    {
+     "start_s": 0.5,
+     "len": 2
+    },
+    {
+     "start_s": 428.5833333333333,
+     "len": 1
+    },
+    {
+     "start_s": 564.875,
+     "len": 11
+    },
+    {
+     "start_s": 566.25,
+     "len": 4
+    },
+    {
+     "start_s": 573.5416666666666,
+     "len": 2
+    },
+    {
+     "start_s": 574.6666666666666,
+     "len": 6
+    },
+    {
+     "start_s": 579.375,
+     "len": 1
+    },
+    {
+     "start_s": 580.4583333333334,
+     "len": 11
+    },
+    {
+     "start_s": 584.0416666666666,
+     "len": 3
+    },
+    {
+     "start_s": 584.625,
+     "len": 4
+    },
+    {
+     "start_s": 589.0833333333334,
+     "len": 3
+    },
+    {
+     "start_s": 590.0416666666666,
+     "len": 9
+    },
+    {
+     "start_s": 641.375,
+     "len": 3
+    },
+    {
+     "start_s": 641.7083333333334,
+     "len": 3
+    },
+    {
+     "start_s": 646.0416666666666,
+     "len": 3
+    },
+    {
+     "start_s": 646.75,
+     "len": 6
+    },
+    {
+     "start_s": 653.25,
+     "len": 8
+    }
+   ]
+  },
+  "self_far_ge40": {
+   "frames": 5,
+   "seconds": 0.20833333333333334,
+   "full_psnr_ge35": 5,
+   "full_psnr_ge40": 0,
+   "full_psnr_median": 36.25993347167969,
+   "runs": [
+    {
+     "start_s": 0.5,
+     "len": 1
+    },
+    {
+     "start_s": 574.6666666666666,
+     "len": 1
+    },
+    {
+     "start_s": 641.4583333333334,
+     "len": 1
+    },
+    {
+     "start_s": 646.75,
+     "len": 1
+    },
+    {
+     "start_s": 653.5416666666666,
+     "len": 1
+    }
+   ]
+  }
+ },
+ "BBB": {
+  "frames": 14315,
+  "flat": 94,
+  "exact_thumb_cross_flat": 73,
+  "exact_thumb_cross_nonflat": 0,
+  "cross_thumb_psnr_quantiles": {
+   "0.5": 17.602630615234375,
+   "0.9": 22.566377639770508,
+   "0.99": 23.89752197265625,
+   "0.999": 32.356964111328125,
+   "1.0": 39.9736328125
+  },
+  "cross_ge35": {
+   "frames": 12,
+   "seconds": 0.5,
+   "full_psnr_median": 34.779043197631836,
+   "full_psnr_ge35": 5,
+   "full_psnr_ge40": 0,
+   "runs": [
+    {
+     "start_s": 299.1666666666667,
+     "len": 2,
+     "with": "SIN",
+     "at_s": 103.75
+    },
+    {
+     "start_s": 300.3333333333333,
+     "len": 2,
+     "with": "SIN",
+     "at_s": 146.54166666666666
+    },
+    {
+     "start_s": 460.8333333333333,
+     "len": 1,
+     "with": "SIN",
+     "at_s": 676.8333333333334
+    },
+    {
+     "start_s": 461.875,
+     "len": 2,
+     "with": "SIN",
+     "at_s": 680.375
+    },
+    {
+     "start_s": 490.25,
+     "len": 5,
+     "with": "SIN",
+     "at_s": 3.3333333333333335
+    }
+   ]
+  },
+  "cross_ge40": {
+   "frames": 0,
+   "seconds": 0.0,
+   "full_psnr_median": null,
+   "full_psnr_ge35": 0,
+   "full_psnr_ge40": 0,
+   "runs": []
+  },
+  "self_far_ge35": {
+   "frames": 8,
+   "seconds": 0.3333333333333333,
+   "full_psnr_ge35": 7,
+   "full_psnr_ge40": 0,
+   "full_psnr_median": 35.861019134521484,
+   "runs": [
+    {
+     "start_s": 299.1666666666667,
+     "len": 2
+    },
+    {
+     "start_s": 300.3333333333333,
+     "len": 1
+    },
+    {
+     "start_s": 460.8333333333333,
+     "len": 1
+    },
+    {
+     "start_s": 461.875,
+     "len": 1
+    },
+    {
+     "start_s": 490.25,
+     "len": 3
+    }
+   ]
+  },
+  "self_far_ge40": {
+   "frames": 0,
+   "seconds": 0.0,
+   "full_psnr_ge35": 0,
+   "full_psnr_ge40": 0,
+   "full_psnr_median": null,
+   "runs": []
+  }
+ },
+ "SIN": {
+  "frames": 21312,
+  "flat": 482,
+  "exact_thumb_cross_flat": 349,
+  "exact_thumb_cross_nonflat": 0,
+  "cross_thumb_psnr_quantiles": {
+   "0.5": 19.00314712524414,
+   "0.9": 26.15960693359375,
+   "0.99": 32.81053161621094,
+   "0.999": 38.6514892578125,
+   "1.0": 40.34019470214844
+  },
+  "cross_ge35": {
+   "frames": 112,
+   "seconds": 4.666666666666667,
+   "full_psnr_median": 34.31572341918945,
+   "full_psnr_ge35": 36,
+   "full_psnr_ge40": 0,
+   "runs": [
+    {
+     "start_s": 3.2916666666666665,
+     "len": 5,
+     "with": "BBB",
+     "at_s": 490.25
+    },
+    {
+     "start_s": 103.58333333333333,
+     "len": 14,
+     "with": "BBB",
+     "at_s": 299.2083333333333
+    },
+    {
+     "start_s": 146.375,
+     "len": 9,
+     "with": "BBB",
+     "at_s": 490.3333333333333
+    },
+    {
+     "start_s": 147.0,
+     "len": 9,
+     "with": "BBB",
+     "at_s": 461.875
+    },
+    {
+     "start_s": 150.5,
+     "len": 2,
+     "with": "ED",
+     "at_s": 0.5833333333333334
+    },
+    {
+     "start_s": 437.6666666666667,
+     "len": 5,
+     "with": "ED",
+     "at_s": 565.2916666666666
+    },
+    {
+     "start_s": 676.7916666666666,
+     "len": 5,
+     "with": "BBB",
+     "at_s": 490.2916666666667
+    },
+    {
+     "start_s": 680.25,
+     "len": 11,
+     "with": "ED",
+     "at_s": 589.1666666666666
+    },
+    {
+     "start_s": 750.3333333333334,
+     "len": 7,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 753.875,
+     "len": 5,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 757.3333333333334,
+     "len": 4,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 760.625,
+     "len": 6,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 764.125,
+     "len": 5,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 767.5,
+     "len": 6,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 770.9166666666666,
+     "len": 6,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 774.3333333333334,
+     "len": 5,
+     "with": "ED",
+     "at_s": 641.7083333333334
+    },
+    {
+     "start_s": 777.875,
+     "len": 4,
+     "with": "ED",
+     "at_s": 584.625
+    },
+    {
+     "start_s": 781.25,
+     "len": 4,
+     "with": "ED",
+     "at_s": 584.625
+    }
+   ]
+  },
+  "cross_ge40": {
+   "frames": 3,
+   "seconds": 0.125,
+   "full_psnr_median": 34.72438430786133,
+   "full_psnr_ge35": 1,
+   "full_psnr_ge40": 0,
+   "runs": [
+    {
+     "start_s": 750.5833333333334,
+     "len": 1,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 767.7083333333334,
+     "len": 1,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    },
+    {
+     "start_s": 771.125,
+     "len": 1,
+     "with": "ED",
+     "at_s": 574.6666666666666
+    }
+   ]
+  },
+  "self_far_ge35": {
+   "frames": 368,
+   "seconds": 15.333333333333334,
+   "full_psnr_ge35": 267,
+   "full_psnr_ge40": 108,
+   "full_psnr_median": 36.434078216552734,
+   "runs": [
+    {
+     "start_s": 3.2916666666666665,
+     "len": 3
+    },
+    {
+     "start_s": 39.708333333333336,
+     "len": 65
+    },
+    {
+     "start_s": 42.5,
+     "len": 16
+    },
+    {
+     "start_s": 90.625,
+     "len": 4
+    },
+    {
+     "start_s": 100.70833333333333,
+     "len": 3
+    },
+    {
+     "start_s": 103.58333333333333,
+     "len": 5
+    },
+    {
+     "start_s": 147.0,
+     "len": 9
+    },
+    {
+     "start_s": 150.5,
+     "len": 1
+    },
+    {
+     "start_s": 431.5833333333333,
+     "len": 82
+    },
+    {
+     "start_s": 437.6666666666667,
+     "len": 5
+    },
+    {
+     "start_s": 676.8333333333334,
+     "len": 4
+    },
+    {
+     "start_s": 680.25,
+     "len": 10
+    },
+    {
+     "start_s": 687.9583333333334,
+     "len": 56
+    },
+    {
+     "start_s": 698.7916666666666,
+     "len": 36
+    },
+    {
+     "start_s": 750.25,
+     "len": 9
+    },
+    {
+     "start_s": 753.7083333333334,
+     "len": 9
+    },
+    {
+     "start_s": 757.25,
+     "len": 6
+    },
+    {
+     "start_s": 760.5416666666666,
+     "len": 8
+    },
+    {
+     "start_s": 763.9583333333334,
+     "len": 9
+    },
+    {
+     "start_s": 767.375,
+     "len": 9
+    },
+    {
+     "start_s": 770.75,
+     "len": 10
+    },
+    {
+     "start_s": 774.375,
+     "len": 4
+    },
+    {
+     "start_s": 777.875,
+     "len": 4
+    },
+    {
+     "start_s": 781.2916666666666,
+     "len": 3
+    }
+   ]
+  },
+  "self_far_ge40": {
+   "frames": 180,
+   "seconds": 7.5,
+   "full_psnr_ge35": 160,
+   "full_psnr_ge40": 108,
+   "full_psnr_median": 40.80683517456055,
+   "runs": [
+    {
+     "start_s": 39.708333333333336,
+     "len": 55
+    },
+    {
+     "start_s": 147.25,
+     "len": 3
+    },
+    {
+     "start_s": 431.5833333333333,
+     "len": 55
+    },
+    {
+     "start_s": 680.25,
+     "len": 2
+    },
+    {
+     "start_s": 687.9583333333334,
+     "len": 10
+    },
+    {
+     "start_s": 689.6666666666666,
+     "len": 15
+    },
+    {
+     "start_s": 699.4166666666666,
+     "len": 21
+    },
+    {
+     "start_s": 750.5,
+     "len": 3
+    },
+    {
+     "start_s": 753.9583333333334,
+     "len": 3
+    },
+    {
+     "start_s": 757.4166666666666,
+     "len": 2
+    },
+    {
+     "start_s": 760.7916666666666,
+     "len": 2
+    },
+    {
+     "start_s": 764.2083333333334,
+     "len": 3
+    },
+    {
+     "start_s": 767.625,
+     "len": 3
+    },
+    {
+     "start_s": 771.0,
+     "len": 4
+    }
+   ]
+  }
+ },
+ "TOS": {
+  "frames": 17620,
+  "flat": 270,
+  "exact_thumb_cross_flat": 245,
+  "exact_thumb_cross_nonflat": 0,
+  "cross_thumb_psnr_quantiles": {
+   "0.5": 17.279958724975586,
+   "0.9": 21.737445831298828,
+   "0.99": 29.325151443481445,
+   "0.999": 33.93708801269531,
+   "1.0": 36.36323547363281
+  },
+  "cross_ge35": {
+   "frames": 10,
+   "seconds": 0.4166666666666667,
+   "full_psnr_median": 33.064083099365234,
+   "full_psnr_ge35": 0,
+   "full_psnr_ge40": 0,
+   "runs": [
+    {
+     "start_s": 65.16666666666667,
+     "len": 8,
+     "with": "BBB",
+     "at_s": 490.375
+    },
+    {
+     "start_s": 707.6666666666666,
+     "len": 2,
+     "with": "BBB",
+     "at_s": 490.4166666666667
+    }
+   ]
+  },
+  "cross_ge40": {
+   "frames": 0,
+   "seconds": 0.0,
+   "full_psnr_median": null,
+   "full_psnr_ge35": 0,
+   "full_psnr_ge40": 0,
+   "runs": []
+  },
+  "self_far_ge35": {
+   "frames": 19,
+   "seconds": 0.7916666666666666,
+   "full_psnr_ge35": 0,
+   "full_psnr_ge40": 0,
+   "full_psnr_median": 32.27806854248047,
+   "runs": [
+    {
+     "start_s": 65.16666666666667,
+     "len": 14
+    },
+    {
+     "start_s": 604.9583333333334,
+     "len": 1
+    },
+    {
+     "start_s": 616.75,
+     "len": 1
+    },
+    {
+     "start_s": 707.625,
+     "len": 3
+    }
+   ]
+  },
+  "self_far_ge40": {
+   "frames": 0,
+   "seconds": 0.0,
+   "full_psnr_ge35": 0,
+   "full_psnr_ge40": 0,
+   "full_psnr_median": null,
+   "runs": []
+  }
+ }
+}
\ No newline at end of file
diff --git a/wai/research/catalogue-prior/c_rd.json b/wai/research/catalogue-prior/c_rd.json
new file mode 100644
index 000000000..bc710b33a
--- /dev/null
+++ b/wai/research/catalogue-prior/c_rd.json
@@ -0,0 +1,1478 @@
+{
+ "ED": {
+  "clip_start_frame": 9415,
+  "clip_start_s": 392.2916666666667,
+  "frames": 1440,
+  "arms": {
+   "base8": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 5415480,
+      "side_bytes": 0.0,
+      "total_bytes": 5415480.0,
+      "ivf_frames": 1440,
+      "psnr": 44.869189269834536,
+      "psnr_framemean": 45.12320713963563,
+      "vmaf": 97.340717
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3467665,
+      "side_bytes": 0.0,
+      "total_bytes": 3467665.0,
+      "ivf_frames": 1440,
+      "psnr": 43.295862574370965,
+      "psnr_framemean": 43.56051469404182,
+      "vmaf": 96.027134
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 2163034,
+      "side_bytes": 0.0,
+      "total_bytes": 2163034.0,
+      "ivf_frames": 1440,
+      "psnr": 41.55102374604969,
+      "psnr_framemean": 41.84135396459378,
+      "vmaf": 93.681386
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1370501,
+      "side_bytes": 0.0,
+      "total_bytes": 1370501.0,
+      "ivf_frames": 1440,
+      "psnr": 39.78276027948923,
+      "psnr_framemean": 40.10981344549731,
+      "vmaf": 90.113656
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 864516,
+      "side_bytes": 0.0,
+      "total_bytes": 864516.0,
+      "ivf_frames": 1440,
+      "psnr": 37.9218327207931,
+      "psnr_framemean": 38.305413276417,
+      "vmaf": 84.804058
+     }
+    ]
+   },
+   "base10": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 2768547,
+      "side_bytes": 0.0,
+      "total_bytes": 2768547.0,
+      "ivf_frames": 1440,
+      "psnr": 42.0666664718805,
+      "psnr_framemean": 42.347922082477226,
+      "vmaf": 94.862108
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 1816307,
+      "side_bytes": 0.0,
+      "total_bytes": 1816307.0,
+      "ivf_frames": 1440,
+      "psnr": 40.54653895998685,
+      "psnr_framemean": 40.83814647736295,
+      "vmaf": 92.316925
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1132632,
+      "side_bytes": 0.0,
+      "total_bytes": 1132632.0,
+      "ivf_frames": 1440,
+      "psnr": 38.71067500629147,
+      "psnr_framemean": 39.03783127121822,
+      "vmaf": 87.640369
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 710419,
+      "side_bytes": 0.0,
+      "total_bytes": 710419.0,
+      "ivf_frames": 1440,
+      "psnr": 36.7980985135572,
+      "psnr_framemean": 37.18115508284545,
+      "vmaf": 81.098441
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 439002,
+      "side_bytes": 0.0,
+      "total_bytes": 439002.0,
+      "ivf_frames": 1440,
+      "psnr": 34.81974617145788,
+      "psnr_framemean": 35.25803659326303,
+      "vmaf": 72.021287
+     }
+    ]
+   },
+   "cat0": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 2791409,
+      "side_bytes": 966.25,
+      "total_bytes": 2792375.25,
+      "ivf_frames": 1440,
+      "psnr": 42.031185253744965,
+      "psnr_framemean": 42.329100602069175,
+      "vmaf": 94.812859
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 1838016,
+      "side_bytes": 966.25,
+      "total_bytes": 1838982.25,
+      "ivf_frames": 1440,
+      "psnr": 40.5260180113524,
+      "psnr_framemean": 40.83361790855102,
+      "vmaf": 92.267706
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1146309,
+      "side_bytes": 966.25,
+      "total_bytes": 1147275.25,
+      "ivf_frames": 1440,
+      "psnr": 38.698672810354196,
+      "psnr_framemean": 39.03741530346891,
+      "vmaf": 87.593717
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 720927,
+      "side_bytes": 966.25,
+      "total_bytes": 721893.25,
+      "ivf_frames": 1440,
+      "psnr": 36.813755474011955,
+      "psnr_framemean": 37.21086455340776,
+      "vmaf": 81.171251
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 444411,
+      "side_bytes": 966.25,
+      "total_bytes": 445377.25,
+      "ivf_frames": 1440,
+      "psnr": 34.8392915360901,
+      "psnr_framemean": 35.292325808495825,
+      "vmaf": 72.289543
+     }
+    ],
+    "ent": 6687.77484967633,
+    "eg": 7730.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 14.063588952900574
+   },
+   "cat64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 3249986,
+      "side_bytes": 89341.5,
+      "total_bytes": 3339327.5,
+      "ivf_frames": 1440,
+      "psnr": 41.65932390471218,
+      "psnr_framemean": 41.962647880823674,
+      "vmaf": 94.433466
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 2130058,
+      "side_bytes": 89341.5,
+      "total_bytes": 2219399.5,
+      "ivf_frames": 1440,
+      "psnr": 40.029087119485105,
+      "psnr_framemean": 40.33314030059839,
+      "vmaf": 91.493937
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1319656,
+      "side_bytes": 89341.5,
+      "total_bytes": 1408997.5,
+      "ivf_frames": 1440,
+      "psnr": 38.14939970499018,
+      "psnr_framemean": 38.48349546527657,
+      "vmaf": 86.294758
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 814074,
+      "side_bytes": 89341.5,
+      "total_bytes": 903415.5,
+      "ivf_frames": 1440,
+      "psnr": 36.21312837544313,
+      "psnr_framemean": 36.59671019798449,
+      "vmaf": 78.876807
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 497280,
+      "side_bytes": 89341.5,
+      "total_bytes": 586621.5,
+      "ivf_frames": 1440,
+      "psnr": 34.27440820843604,
+      "psnr_framemean": 34.72024553897723,
+      "vmaf": 69.31851
+     }
+    ],
+    "ent": 551529.5288197354,
+    "eg": 714732.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 19.883853800696528
+   },
+   "cat256": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 3509166,
+      "side_bytes": 315438.25,
+      "total_bytes": 3824604.25,
+      "ivf_frames": 1440,
+      "psnr": 41.42163188166242,
+      "psnr_framemean": 41.735336703386764,
+      "vmaf": 94.066684
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 2288543,
+      "side_bytes": 315438.25,
+      "total_bytes": 2603981.25,
+      "ivf_frames": 1440,
+      "psnr": 39.743820935811314,
+      "psnr_framemean": 40.064703024808395,
+      "vmaf": 90.920158
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1393617,
+      "side_bytes": 315438.25,
+      "total_bytes": 1709055.25,
+      "ivf_frames": 1440,
+      "psnr": 37.782409251427474,
+      "psnr_framemean": 38.13182835948933,
+      "vmaf": 85.115616
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 845624,
+      "side_bytes": 315438.25,
+      "total_bytes": 1161062.25,
+      "ivf_frames": 1440,
+      "psnr": 35.81904115009913,
+      "psnr_framemean": 36.23074157562024,
+      "vmaf": 77.020217
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 506289,
+      "side_bytes": 315438.25,
+      "total_bytes": 821727.25,
+      "ivf_frames": 1440,
+      "psnr": 33.850560310547536,
+      "psnr_framemean": 34.33017710039257,
+      "vmaf": 66.68899
+     }
+    ],
+    "ent": 2021233.6826639387,
+    "eg": 2523506.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 22.34223743885449
+   },
+   "self64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 3403150,
+      "side_bytes": 90158.25,
+      "total_bytes": 3493308.25,
+      "ivf_frames": 1440,
+      "psnr": 41.497607350787355,
+      "psnr_framemean": 41.79090288996509,
+      "vmaf": 94.297886
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 2233324,
+      "side_bytes": 90158.25,
+      "total_bytes": 2323482.25,
+      "ivf_frames": 1440,
+      "psnr": 39.864791189368404,
+      "psnr_framemean": 40.160710505510195,
+      "vmaf": 91.318955
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1375685,
+      "side_bytes": 90158.25,
+      "total_bytes": 1465843.25,
+      "ivf_frames": 1440,
+      "psnr": 37.94069161439629,
+      "psnr_framemean": 38.262444772016835,
+      "vmaf": 85.818262
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 849278,
+      "side_bytes": 90158.25,
+      "total_bytes": 939436.25,
+      "ivf_frames": 1440,
+      "psnr": 36.00201371295039,
+      "psnr_framemean": 36.37420668171531,
+      "vmaf": 78.163113
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 517609,
+      "side_bytes": 90158.25,
+      "total_bytes": 607767.25,
+      "ivf_frames": 1440,
+      "psnr": 34.028754512288636,
+      "psnr_framemean": 34.469487930876575,
+      "vmaf": 68.156363
+     }
+    ],
+    "ent": 554730.2472278435,
+    "eg": 721266.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 19.40441525259992
+   }
+  },
+  "bd_vs_base10": {
+   "base8": {
+    "psnr": -8.230699686868414,
+    "vmaf": -2.391187773126624
+   },
+   "cat0": {
+    "psnr": 1.494584361998097,
+    "vmaf": 1.219346810978461
+   },
+   "cat64": {
+    "psnr": 42.89160416143525,
+    "vmaf": 40.728692179681424
+   },
+   "cat256": {
+    "psnr": 88.54766650498895,
+    "vmaf": 88.14562230738039
+   },
+   "self64": {
+    "psnr": 56.11133665811578,
+    "vmaf": 51.323576165278205
+   }
+  }
+ },
+ "BBB": {
+  "clip_start_frame": 8589,
+  "clip_start_s": 357.875,
+  "frames": 1440,
+  "arms": {
+   "base8": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 5235603,
+      "side_bytes": 0.0,
+      "total_bytes": 5235603.0,
+      "ivf_frames": 1440,
+      "psnr": 46.03158148483384,
+      "psnr_framemean": 47.006917155740084,
+      "vmaf": 97.215047
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3473876,
+      "side_bytes": 0.0,
+      "total_bytes": 3473876.0,
+      "ivf_frames": 1440,
+      "psnr": 44.1773378334363,
+      "psnr_framemean": 45.241170552117566,
+      "vmaf": 96.486438
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 2247165,
+      "side_bytes": 0.0,
+      "total_bytes": 2247165.0,
+      "ivf_frames": 1440,
+      "psnr": 42.10625750498732,
+      "psnr_framemean": 43.21400898285007,
+      "vmaf": 95.134891
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1474835,
+      "side_bytes": 0.0,
+      "total_bytes": 1474835.0,
+      "ivf_frames": 1440,
+      "psnr": 40.062279237314435,
+      "psnr_framemean": 41.16812438212843,
+      "vmaf": 92.818246
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 966594,
+      "side_bytes": 0.0,
+      "total_bytes": 966594.0,
+      "ivf_frames": 1440,
+      "psnr": 37.97041374294856,
+      "psnr_framemean": 39.022041445502985,
+      "vmaf": 89.07917
+     }
+    ]
+   },
+   "base10": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 2899001,
+      "side_bytes": 0.0,
+      "total_bytes": 2899001.0,
+      "ivf_frames": 1440,
+      "psnr": 42.91184364846271,
+      "psnr_framemean": 43.94442888566958,
+      "vmaf": 95.859135
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 1948146,
+      "side_bytes": 0.0,
+      "total_bytes": 1948146.0,
+      "ivf_frames": 1440,
+      "psnr": 41.07542498212071,
+      "psnr_framemean": 42.14766050038985,
+      "vmaf": 94.266249
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1254709,
+      "side_bytes": 0.0,
+      "total_bytes": 1254709.0,
+      "ivf_frames": 1440,
+      "psnr": 38.95685073685845,
+      "psnr_framemean": 40.002281312874885,
+      "vmaf": 91.138471
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 801267,
+      "side_bytes": 0.0,
+      "total_bytes": 801267.0,
+      "ivf_frames": 1440,
+      "psnr": 36.79231011240814,
+      "psnr_framemean": 37.81268959957981,
+      "vmaf": 86.419929
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 501179,
+      "side_bytes": 0.0,
+      "total_bytes": 501179.0,
+      "ivf_frames": 1440,
+      "psnr": 34.58629035698691,
+      "psnr_framemean": 35.58756189583957,
+      "vmaf": 79.587598
+     }
+    ]
+   },
+   "cat0": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 2927166,
+      "side_bytes": 411.0,
+      "total_bytes": 2927577.0,
+      "ivf_frames": 1440,
+      "psnr": 42.86025793323965,
+      "psnr_framemean": 43.90866244643448,
+      "vmaf": 95.809387
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 1976634,
+      "side_bytes": 411.0,
+      "total_bytes": 1977045.0,
+      "ivf_frames": 1440,
+      "psnr": 41.04620401290852,
+      "psnr_framemean": 42.133349448011316,
+      "vmaf": 94.193291
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1266929,
+      "side_bytes": 411.0,
+      "total_bytes": 1267340.0,
+      "ivf_frames": 1440,
+      "psnr": 38.90767251711119,
+      "psnr_framemean": 39.95223056860102,
+      "vmaf": 91.013853
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 807858,
+      "side_bytes": 411.0,
+      "total_bytes": 808269.0,
+      "ivf_frames": 1440,
+      "psnr": 36.7510290622706,
+      "psnr_framemean": 37.766492631551806,
+      "vmaf": 86.249764
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 504694,
+      "side_bytes": 411.0,
+      "total_bytes": 505105.0,
+      "ivf_frames": 1440,
+      "psnr": 34.55803811446037,
+      "psnr_framemean": 35.54785186513876,
+      "vmaf": 79.460526
+     }
+    ],
+    "ent": 2985.9181878901754,
+    "eg": 3288.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 15.468116198264832
+   },
+   "cat64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 3232326,
+      "side_bytes": 57802.5,
+      "total_bytes": 3290128.5,
+      "ivf_frames": 1440,
+      "psnr": 42.41144785743891,
+      "psnr_framemean": 43.517434218914396,
+      "vmaf": 95.407098
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 2162757,
+      "side_bytes": 57802.5,
+      "total_bytes": 2220559.5,
+      "ivf_frames": 1440,
+      "psnr": 40.546417997258615,
+      "psnr_framemean": 41.669959196330346,
+      "vmaf": 93.440621
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1379500,
+      "side_bytes": 57802.5,
+      "total_bytes": 1437302.5,
+      "ivf_frames": 1440,
+      "psnr": 38.435001176374406,
+      "psnr_framemean": 39.516260642535,
+      "vmaf": 89.861803
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 869798,
+      "side_bytes": 57802.5,
+      "total_bytes": 927600.5,
+      "ivf_frames": 1440,
+      "psnr": 36.293173405349236,
+      "psnr_framemean": 37.31492716585215,
+      "vmaf": 84.521484
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 539371,
+      "side_bytes": 57802.5,
+      "total_bytes": 597173.5,
+      "ivf_frames": 1440,
+      "psnr": 34.12421470152814,
+      "psnr_framemean": 35.110356976474264,
+      "vmaf": 77.097059
+     }
+    ],
+    "ent": 415953.90036967705,
+    "eg": 462420.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 19.72412828483119
+   },
+   "cat256": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 3445910,
+      "side_bytes": 217128.75,
+      "total_bytes": 3663038.75,
+      "ivf_frames": 1440,
+      "psnr": 42.11807807465411,
+      "psnr_framemean": 43.225513476345576,
+      "vmaf": 95.047316
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 2297681,
+      "side_bytes": 217128.75,
+      "total_bytes": 2514809.75,
+      "ivf_frames": 1440,
+      "psnr": 40.25352710653237,
+      "psnr_framemean": 41.40162596279262,
+      "vmaf": 92.924888
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1449203,
+      "side_bytes": 217128.75,
+      "total_bytes": 1666331.75,
+      "ivf_frames": 1440,
+      "psnr": 38.109030797665135,
+      "psnr_framemean": 39.20880573124025,
+      "vmaf": 88.862543
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 901284,
+      "side_bytes": 217128.75,
+      "total_bytes": 1118412.75,
+      "ivf_frames": 1440,
+      "psnr": 35.962559951313985,
+      "psnr_framemean": 37.01742715527696,
+      "vmaf": 83.08026
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 551541,
+      "side_bytes": 217128.75,
+      "total_bytes": 768669.75,
+      "ivf_frames": 1440,
+      "psnr": 33.82251029078642,
+      "psnr_framemean": 34.83852274945522,
+      "vmaf": 75.187802
+     }
+    ],
+    "ent": 1585632.1331176655,
+    "eg": 1737030.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 21.53346950225071
+   },
+   "self64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 3515344,
+      "side_bytes": 57114.75,
+      "total_bytes": 3572458.75,
+      "ivf_frames": 1440,
+      "psnr": 42.05709166934332,
+      "psnr_framemean": 43.16800806082902,
+      "vmaf": 95.180352
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 2349173,
+      "side_bytes": 57114.75,
+      "total_bytes": 2406287.75,
+      "ivf_frames": 1440,
+      "psnr": 40.16802303539208,
+      "psnr_framemean": 41.310483807532215,
+      "vmaf": 93.063156
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1486044,
+      "side_bytes": 57114.75,
+      "total_bytes": 1543158.75,
+      "ivf_frames": 1440,
+      "psnr": 38.01153120853521,
+      "psnr_framemean": 39.11492483409908,
+      "vmaf": 89.065428
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 931315,
+      "side_bytes": 57114.75,
+      "total_bytes": 988429.75,
+      "ivf_frames": 1440,
+      "psnr": 35.85878660042172,
+      "psnr_framemean": 36.892943426472925,
+      "vmaf": 83.151032
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 569960,
+      "side_bytes": 57114.75,
+      "total_bytes": 627074.75,
+      "ivf_frames": 1440,
+      "psnr": 33.70221947840986,
+      "psnr_framemean": 34.67693751600591,
+      "vmaf": 75.343871
+     }
+    ],
+    "ent": 412495.5183635528,
+    "eg": 456918.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 18.997518038390144
+   }
+  },
+  "bd_vs_base10": {
+   "base8": {
+    "psnr": -6.684973762917911,
+    "vmaf": -2.6388118772678215
+   },
+   "cat0": {
+    "psnr": 1.9191650574068353,
+    "vmaf": 2.1969926576168364
+   },
+   "cat64": {
+    "psnr": 27.961684347309166,
+    "vmaf": 32.07818047515638
+   },
+   "cat256": {
+    "psnr": 58.97000667382035,
+    "vmaf": 69.30818024884458
+   },
+   "self64": {
+    "psnr": 49.60010303360654,
+    "vmaf": 52.696257577043305
+   }
+  }
+ },
+ "SIN": {
+  "clip_start_frame": 12787,
+  "clip_start_s": 532.7916666666666,
+  "frames": 1440,
+  "arms": {
+   "base8": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 9551822,
+      "side_bytes": 0.0,
+      "total_bytes": 9551822.0,
+      "ivf_frames": 1440,
+      "psnr": 44.400439725430026,
+      "psnr_framemean": 44.83439544407736,
+      "vmaf": 99.019028
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 6056691,
+      "side_bytes": 0.0,
+      "total_bytes": 6056691.0,
+      "ivf_frames": 1440,
+      "psnr": 42.30970282222789,
+      "psnr_framemean": 42.81374100840456,
+      "vmaf": 98.07761
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 3590978,
+      "side_bytes": 0.0,
+      "total_bytes": 3590978.0,
+      "ivf_frames": 1440,
+      "psnr": 39.98532742203559,
+      "psnr_framemean": 40.568837133983045,
+      "vmaf": 95.199064
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 2108471,
+      "side_bytes": 0.0,
+      "total_bytes": 2108471.0,
+      "ivf_frames": 1440,
+      "psnr": 37.79175556205321,
+      "psnr_framemean": 38.432366672499775,
+      "vmaf": 89.142737
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 1243721,
+      "side_bytes": 0.0,
+      "total_bytes": 1243721.0,
+      "ivf_frames": 1440,
+      "psnr": 35.744970106597805,
+      "psnr_framemean": 36.42060937334419,
+      "vmaf": 79.523535
+     }
+    ]
+   },
+   "base10": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 4881481,
+      "side_bytes": 0.0,
+      "total_bytes": 4881481.0,
+      "ivf_frames": 1440,
+      "psnr": 40.89818292115763,
+      "psnr_framemean": 41.44581436684374,
+      "vmaf": 96.693371
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3025514,
+      "side_bytes": 0.0,
+      "total_bytes": 3025514.0,
+      "ivf_frames": 1440,
+      "psnr": 38.93604383872607,
+      "psnr_framemean": 39.51739859003909,
+      "vmaf": 93.024047
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1730304,
+      "side_bytes": 0.0,
+      "total_bytes": 1730304.0,
+      "ivf_frames": 1440,
+      "psnr": 36.745007522279096,
+      "psnr_framemean": 37.370381532258286,
+      "vmaf": 84.916236
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 993923,
+      "side_bytes": 0.0,
+      "total_bytes": 993923.0,
+      "ivf_frames": 1440,
+      "psnr": 34.679034983897445,
+      "psnr_framemean": 35.32665541381422,
+      "vmaf": 73.290644
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 569364,
+      "side_bytes": 0.0,
+      "total_bytes": 569364.0,
+      "ivf_frames": 1440,
+      "psnr": 32.69413102461901,
+      "psnr_framemean": 33.358219780815865,
+      "vmaf": 58.865099
+     }
+    ]
+   },
+   "cat0": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 4946859,
+      "side_bytes": 767.75,
+      "total_bytes": 4947626.75,
+      "ivf_frames": 1440,
+      "psnr": 40.83877490739848,
+      "psnr_framemean": 41.389685646845535,
+      "vmaf": 96.536524
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3062146,
+      "side_bytes": 767.75,
+      "total_bytes": 3062913.75,
+      "ivf_frames": 1440,
+      "psnr": 38.871385953564385,
+      "psnr_framemean": 39.44875510876222,
+      "vmaf": 92.78271
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1743765,
+      "side_bytes": 767.75,
+      "total_bytes": 1744532.75,
+      "ivf_frames": 1440,
+      "psnr": 36.67970842285908,
+      "psnr_framemean": 37.29481930337524,
+      "vmaf": 84.546086
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1000552,
+      "side_bytes": 767.75,
+      "total_bytes": 1001319.75,
+      "ivf_frames": 1440,
+      "psnr": 34.61490578342566,
+      "psnr_framemean": 35.25825555794419,
+      "vmaf": 72.858099
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 570303,
+      "side_bytes": 767.75,
+      "total_bytes": 571070.75,
+      "ivf_frames": 1440,
+      "psnr": 32.63779899462632,
+      "psnr_framemean": 33.30247572565728,
+      "vmaf": 58.509489
+     }
+    ],
+    "ent": 5647.569971912102,
+    "eg": 6142.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 18.248728688614293
+   },
+   "cat64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 5756836,
+      "side_bytes": 106189.75,
+      "total_bytes": 5863025.75,
+      "ivf_frames": 1440,
+      "psnr": 40.39446395648882,
+      "psnr_framemean": 40.88294400450343,
+      "vmaf": 96.04376
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3544920,
+      "side_bytes": 106189.75,
+      "total_bytes": 3651109.75,
+      "ivf_frames": 1440,
+      "psnr": 38.402415411628944,
+      "psnr_framemean": 38.897985154928996,
+      "vmaf": 91.807445
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1998913,
+      "side_bytes": 106189.75,
+      "total_bytes": 2105102.75,
+      "ivf_frames": 1440,
+      "psnr": 36.181066543650964,
+      "psnr_framemean": 36.70128273053039,
+      "vmaf": 82.651321
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1125740,
+      "side_bytes": 106189.75,
+      "total_bytes": 1231929.75,
+      "ivf_frames": 1440,
+      "psnr": 34.08677678072393,
+      "psnr_framemean": 34.63152379500046,
+      "vmaf": 69.939279
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 625911,
+      "side_bytes": 106189.75,
+      "total_bytes": 732100.75,
+      "ivf_frames": 1440,
+      "psnr": 32.074002177709296,
+      "psnr_framemean": 32.63966338298702,
+      "vmaf": 54.797392
+     }
+    ],
+    "ent": 625965.5759421862,
+    "eg": 849518.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 22.00446851191781
+   },
+   "cat256": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 6334140,
+      "side_bytes": 405183.5,
+      "total_bytes": 6739323.5,
+      "ivf_frames": 1440,
+      "psnr": 40.19883925469521,
+      "psnr_framemean": 40.65626910653122,
+      "vmaf": 95.866731
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3894020,
+      "side_bytes": 405183.5,
+      "total_bytes": 4299203.5,
+      "ivf_frames": 1440,
+      "psnr": 38.15256579841126,
+      "psnr_framemean": 38.60830629139634,
+      "vmaf": 91.437754
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 2182948,
+      "side_bytes": 405183.5,
+      "total_bytes": 2588131.5,
+      "ivf_frames": 1440,
+      "psnr": 35.88029145106651,
+      "psnr_framemean": 36.35854893785764,
+      "vmaf": 81.647449
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1203557,
+      "side_bytes": 405183.5,
+      "total_bytes": 1608740.5,
+      "ivf_frames": 1440,
+      "psnr": 33.719125856211406,
+      "psnr_framemean": 34.21736638474951,
+      "vmaf": 68.033088
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 645515,
+      "side_bytes": 405183.5,
+      "total_bytes": 1050698.5,
+      "ivf_frames": 1440,
+      "psnr": 31.66143932502858,
+      "psnr_framemean": 32.18878291478436,
+      "vmaf": 52.41805
+     }
+    ],
+    "ent": 2401503.230583039,
+    "eg": 3241468.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 23.869539593770273
+   },
+   "self64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 5673805,
+      "side_bytes": 107080.0,
+      "total_bytes": 5780885.0,
+      "ivf_frames": 1440,
+      "psnr": 40.44399502277548,
+      "psnr_framemean": 40.938090520908474,
+      "vmaf": 96.048498
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3502082,
+      "side_bytes": 107080.0,
+      "total_bytes": 3609162.0,
+      "ivf_frames": 1440,
+      "psnr": 38.455775577639756,
+      "psnr_framemean": 38.959416087216574,
+      "vmaf": 91.879793
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 1983741,
+      "side_bytes": 107080.0,
+      "total_bytes": 2090821.0,
+      "ivf_frames": 1440,
+      "psnr": 36.23825194191703,
+      "psnr_framemean": 36.76989504447683,
+      "vmaf": 82.774189
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1122589,
+      "side_bytes": 107080.0,
+      "total_bytes": 1229669.0,
+      "ivf_frames": 1440,
+      "psnr": 34.14476301139776,
+      "psnr_framemean": 34.698540197312006,
+      "vmaf": 70.185892
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 625033,
+      "side_bytes": 107080.0,
+      "total_bytes": 732113.0,
+      "ivf_frames": 1440,
+      "psnr": 32.11594864242234,
+      "psnr_framemean": 32.687722229328344,
+      "vmaf": 54.978638
+     }
+    ],
+    "ent": 629601.3128140335,
+    "eg": 856640.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 22.045897018301716
+   }
+  },
+  "bd_vs_base10": {
+   "base8": {
+    "psnr": -7.425448161319426,
+    "vmaf": -0.6077191012131222
+   },
+   "cat0": {
+    "psnr": 2.600956783925912,
+    "vmaf": 2.677198246071111
+   },
+   "cat64": {
+    "psnr": 41.356719247048005,
+    "vmaf": 39.00435665927691
+   },
+   "cat256": {
+    "psnr": 86.59690287133594,
+    "vmaf": 83.53383018313959
+   },
+   "self64": {
+    "psnr": 38.383308640352574,
+    "vmaf": 37.357117144287514
+   }
+  }
+ },
+ "TOS": {
+  "clip_start_frame": 10572,
+  "clip_start_s": 440.5,
+  "frames": 1440,
+  "arms": {
+   "base8": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 11584448,
+      "side_bytes": 0.0,
+      "total_bytes": 11584448.0,
+      "ivf_frames": 1440,
+      "psnr": 43.62404399113377,
+      "psnr_framemean": 44.08847765846276,
+      "vmaf": 99.466534
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 7489055,
+      "side_bytes": 0.0,
+      "total_bytes": 7489055.0,
+      "ivf_frames": 1440,
+      "psnr": 41.48581324615925,
+      "psnr_framemean": 42.06127827647209,
+      "vmaf": 99.008561
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 4624135,
+      "side_bytes": 0.0,
+      "total_bytes": 4624135.0,
+      "ivf_frames": 1440,
+      "psnr": 39.14333463989206,
+      "psnr_framemean": 39.813931790381304,
+      "vmaf": 97.525807
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 2857959,
+      "side_bytes": 0.0,
+      "total_bytes": 2857959.0,
+      "ivf_frames": 1440,
+      "psnr": 36.93068352453639,
+      "psnr_framemean": 37.64475178056617,
+      "vmaf": 93.800733
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 1780836,
+      "side_bytes": 0.0,
+      "total_bytes": 1780836.0,
+      "ivf_frames": 1440,
+      "psnr": 34.79327599708622,
+      "psnr_framemean": 35.53694297678101,
+      "vmaf": 86.2764
+     }
+    ]
+   },
+   "base10": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 6152006,
+      "side_bytes": 0.0,
+      "total_bytes": 6152006.0,
+      "ivf_frames": 1440,
+      "psnr": 40.12251219382975,
+      "psnr_framemean": 40.74386771873605,
+      "vmaf": 98.249812
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 3985892,
+      "side_bytes": 0.0,
+      "total_bytes": 3985892.0,
+      "ivf_frames": 1440,
+      "psnr": 38.11199449071471,
+      "psnr_framemean": 38.77262027017178,
+      "vmaf": 96.200745
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 2412543,
+      "side_bytes": 0.0,
+      "total_bytes": 2412543.0,
+      "ivf_frames": 1440,
+      "psnr": 35.87189219836006,
+      "psnr_framemean": 36.55132920050559,
+      "vmaf": 90.699956
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1458608,
+      "side_bytes": 0.0,
+      "total_bytes": 1458608.0,
+      "ivf_frames": 1440,
+      "psnr": 33.670204823428286,
+      "psnr_framemean": 34.375851036127806,
+      "vmaf": 81.024012
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 862600,
+      "side_bytes": 0.0,
+      "total_bytes": 862600.0,
+      "ivf_frames": 1440,
+      "psnr": 31.5090345462869,
+      "psnr_framemean": 32.23579938777524,
+      "vmaf": 68.140692
+     }
+    ]
+   },
+   "cat0": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 6237955,
+      "side_bytes": 930.75,
+      "total_bytes": 6238885.75,
+      "ivf_frames": 1440,
+      "psnr": 40.06669239080406,
+      "psnr_framemean": 40.662450215528764,
+      "vmaf": 98.205142
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 4035468,
+      "side_bytes": 930.75,
+      "total_bytes": 4036398.75,
+      "ivf_frames": 1440,
+      "psnr": 38.06009288586888,
+      "psnr_framemean": 38.69117496375121,
+      "vmaf": 96.044519
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 2439780,
+      "side_bytes": 930.75,
+      "total_bytes": 2440710.75,
+      "ivf_frames": 1440,
+      "psnr": 35.81373251834147,
+      "psnr_framemean": 36.46989068428275,
+      "vmaf": 90.343993
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1474430,
+      "side_bytes": 930.75,
+      "total_bytes": 1475360.75,
+      "ivf_frames": 1440,
+      "psnr": 33.63498893697723,
+      "psnr_framemean": 34.30839955611832,
+      "vmaf": 80.618757
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 869669,
+      "side_bytes": 930.75,
+      "total_bytes": 870599.75,
+      "ivf_frames": 1440,
+      "psnr": 31.443425107204696,
+      "psnr_framemean": 32.144237365444965,
+      "vmaf": 67.492317
+     }
+    ],
+    "ent": 6445.0589524148445,
+    "eg": 7446.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 14.958910281420778
+   },
+   "cat64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 7239511,
+      "side_bytes": 111682.25,
+      "total_bytes": 7351193.25,
+      "ivf_frames": 1440,
+      "psnr": 39.68984129495519,
+      "psnr_framemean": 40.20589735485601,
+      "vmaf": 97.718237
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 4676991,
+      "side_bytes": 111682.25,
+      "total_bytes": 4788673.25,
+      "ivf_frames": 1440,
+      "psnr": 37.6164827629708,
+      "psnr_framemean": 38.15258195365614,
+      "vmaf": 94.964198
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 2795074,
+      "side_bytes": 111682.25,
+      "total_bytes": 2906756.25,
+      "ivf_frames": 1440,
+      "psnr": 35.32429191759978,
+      "psnr_framemean": 35.86272134593087,
+      "vmaf": 88.05427
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1661631,
+      "side_bytes": 111682.25,
+      "total_bytes": 1773313.25,
+      "ivf_frames": 1440,
+      "psnr": 33.113951724289585,
+      "psnr_framemean": 33.668574788656336,
+      "vmaf": 76.843042
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 961678,
+      "side_bytes": 111682.25,
+      "total_bytes": 1073360.25,
+      "ivf_frames": 1440,
+      "psnr": 30.921172475836695,
+      "psnr_framemean": 31.50278714339536,
+      "vmaf": 62.869991
+     }
+    ],
+    "ent": 656457.2309044809,
+    "eg": 893458.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 19.010959164343404
+   },
+   "cat256": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 7785432,
+      "side_bytes": 418489.25,
+      "total_bytes": 8203921.25,
+      "ivf_frames": 1440,
+      "psnr": 39.5055172456427,
+      "psnr_framemean": 39.997669498203365,
+      "vmaf": 97.505422
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 5025349,
+      "side_bytes": 418489.25,
+      "total_bytes": 5443838.25,
+      "ivf_frames": 1440,
+      "psnr": 37.39793874527949,
+      "psnr_framemean": 37.90385535604739,
+      "vmaf": 94.536224
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 2979644,
+      "side_bytes": 418489.25,
+      "total_bytes": 3398133.25,
+      "ivf_frames": 1440,
+      "psnr": 35.0618282710334,
+      "psnr_framemean": 35.56666362945194,
+      "vmaf": 87.020993
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1745856,
+      "side_bytes": 418489.25,
+      "total_bytes": 2164345.25,
+      "ivf_frames": 1440,
+      "psnr": 32.80497125995932,
+      "psnr_framemean": 33.31885733122704,
+      "vmaf": 75.011164
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 989190,
+      "side_bytes": 418489.25,
+      "total_bytes": 1407679.25,
+      "ivf_frames": 1440,
+      "psnr": 30.59102099131275,
+      "psnr_framemean": 31.131313507201188,
+      "vmaf": 60.362969
+     }
+    ],
+    "ent": 2494567.3391360487,
+    "eg": 3347914.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 20.85612457718682
+   },
+   "self64": {
+    "points": [
+     {
+      "crf": 20,
+      "av1_bytes": 7816592,
+      "side_bytes": 110629.75,
+      "total_bytes": 7927221.75,
+      "ivf_frames": 1440,
+      "psnr": 39.53205635634789,
+      "psnr_framemean": 40.01025694358808,
+      "vmaf": 97.594729
+     },
+     {
+      "crf": 28,
+      "av1_bytes": 5063981,
+      "side_bytes": 110629.75,
+      "total_bytes": 5174610.75,
+      "ivf_frames": 1440,
+      "psnr": 37.423683171234444,
+      "psnr_framemean": 37.909592538050866,
+      "vmaf": 94.615053
+     },
+     {
+      "crf": 36,
+      "av1_bytes": 3021433,
+      "side_bytes": 110629.75,
+      "total_bytes": 3132062.75,
+      "ivf_frames": 1440,
+      "psnr": 35.08027188830487,
+      "psnr_framemean": 35.561379121979954,
+      "vmaf": 87.209707
+     },
+     {
+      "crf": 44,
+      "av1_bytes": 1786190,
+      "side_bytes": 110629.75,
+      "total_bytes": 1896819.75,
+      "ivf_frames": 1440,
+      "psnr": 32.810221269667736,
+      "psnr_framemean": 33.29717294132781,
+      "vmaf": 75.082227
+     },
+     {
+      "crf": 52,
+      "av1_bytes": 1028724,
+      "side_bytes": 110629.75,
+      "total_bytes": 1139353.75,
+      "ivf_frames": 1440,
+      "psnr": 30.5961645311899,
+      "psnr_framemean": 31.103185741326765,
+      "vmaf": 60.437779
+     }
+    ],
+    "ent": 651730.2423338838,
+    "eg": 885038.0,
+    "clipped_frac": 0.0,
+    "basis_only_psnr": 18.42171459430036
+   }
+  },
+  "bd_vs_base10": {
+   "base8": {
+    "psnr": -6.829339873954798,
+    "vmaf": 9.082657246734826
+   },
+   "cat0": {
+    "psnr": 2.3752150669950955,
+    "vmaf": 3.1995567310901363
+   },
+   "cat64": {
+    "psnr": 36.18165799822559,
+    "vmaf": 41.63729272478061
+   },
+   "cat256": {
+    "psnr": 68.7479076403819,
+    "vmaf": 77.53775838031953
+   },
+   "self64": {
+    "psnr": 54.51745909094539,
+    "vmaf": 59.65191972969153
+   }
+  }
+ }
+}
\ No newline at end of file
diff --git a/wai/research/catalogue-prior/d_exact.json b/wai/research/catalogue-prior/d_exact.json
new file mode 100644
index 000000000..f012e8802
--- /dev/null
+++ b/wai/research/catalogue-prior/d_exact.json
@@ -0,0 +1,122 @@
+{
+ "ED": {
+  "frames": 15693,
+  "av1_bytes_crf36": 21882751,
+  "kbps": 267.7300829669279,
+  "flat": {
+   "frames": 228,
+   "seconds": 9.5,
+   "runtime_pct": 1.4528770789523993,
+   "bytes": 46263,
+   "bytes_pct": 0.21141308969790865,
+   "thumb_std_median": 0.03606089172949539
+  },
+  "cross35": {
+   "frames": 13,
+   "seconds": 0.5416666666666666,
+   "runtime_pct": 0.08283948257184733,
+   "bytes": 398,
+   "bytes_pct": 0.0018187841190534042,
+   "thumb_std_median": 3.000226016223685
+  },
+  "self_far35": {
+   "frames": 11,
+   "seconds": 0.4583333333333333,
+   "runtime_pct": 0.07009494679156311,
+   "bytes": 341,
+   "bytes_pct": 0.0015583049864251529,
+   "thumb_std_median": 2.2512774405956963
+  },
+  "nonflat_thumb_std_median": 30.291161437876323
+ },
+ "BBB": {
+  "frames": 14315,
+  "av1_bytes_crf36": 20683682,
+  "kbps": 277.41997513098147,
+  "flat": {
+   "frames": 94,
+   "seconds": 3.9166666666666665,
+   "runtime_pct": 0.656653859587845,
+   "bytes": 134242,
+   "bytes_pct": 0.6490237086414304,
+   "thumb_std_median": 0.0
+  },
+  "cross35": {
+   "frames": 5,
+   "seconds": 0.20833333333333334,
+   "runtime_pct": 0.0349283967865875,
+   "bytes": 284,
+   "bytes_pct": 0.0013730630745531671,
+   "thumb_std_median": 2.6479624238745973
+  },
+  "self_far35": {
+   "frames": 7,
+   "seconds": 0.2916666666666667,
+   "runtime_pct": 0.0488997555012225,
+   "bytes": 350,
+   "bytes_pct": 0.001692155197512706,
+   "thumb_std_median": 2.6479624238745973
+  },
+  "nonflat_thumb_std_median": 35.89060543273839
+ },
+ "SIN": {
+  "frames": 21312,
+  "av1_bytes_crf36": 30750110,
+  "kbps": 277.02801801801803,
+  "flat": {
+   "frames": 482,
+   "seconds": 20.083333333333332,
+   "runtime_pct": 2.2616366366366365,
+   "bytes": 53243,
+   "bytes_pct": 0.17314734809078733,
+   "thumb_std_median": 0.0
+  },
+  "cross35": {
+   "frames": 36,
+   "seconds": 1.5,
+   "runtime_pct": 0.16891891891891891,
+   "bytes": 3187,
+   "bytes_pct": 0.010364190567123175,
+   "thumb_std_median": 2.980607967828133
+  },
+  "self_far35": {
+   "frames": 267,
+   "seconds": 11.125,
+   "runtime_pct": 1.2528153153153154,
+   "bytes": 302619,
+   "bytes_pct": 0.9841233088271879,
+   "thumb_std_median": 58.21047824127981
+  },
+  "nonflat_thumb_std_median": 30.714039515592862
+ },
+ "TOS": {
+  "frames": 17620,
+  "av1_bytes_crf36": 31963904,
+  "kbps": 348.3013375709421,
+  "flat": {
+   "frames": 270,
+   "seconds": 11.25,
+   "runtime_pct": 1.532349602724177,
+   "bytes": 88896,
+   "bytes_pct": 0.2781137122674377,
+   "thumb_std_median": 0.0
+  },
+  "cross35": {
+   "frames": 0,
+   "seconds": 0.0,
+   "runtime_pct": 0.0,
+   "bytes": 0,
+   "bytes_pct": 0.0,
+   "thumb_std_median": null
+  },
+  "self_far35": {
+   "frames": 0,
+   "seconds": 0.0,
+   "runtime_pct": 0.0,
+   "bytes": 0,
+   "bytes_pct": 0.0,
+   "thumb_std_median": null
+  },
+  "nonflat_thumb_std_median": 37.15009941663702
+ }
+}
\ No newline at end of file
diff --git a/wai/research/catalogue-prior/run.json b/wai/research/catalogue-prior/run.json
new file mode 100644
index 000000000..6c1985814
--- /dev/null
+++ b/wai/research/catalogue-prior/run.json
@@ -0,0 +1,91 @@
+{
+ "format": "wai-catalogue-prior/1",
+ "about": "The run behind research/catalogue-prior.md: its corpus, tools, timeline, decision rule and verdicts, and the SHA-256 of each reading file beside this one, as the run wrote it.",
+ "readings": {
+  "a_basis.json": "5186b5b77adfc9aabd685b3e4a3d1fa34fa82c345093b6b28c9a0c1c2467ca68",
+  "b_recur.json": "10247e1f4bfdd8b847e362daa870d80b7f3b83e12cd896f924c8215e267bc9e2",
+  "c_rd.json": "c81b95872c384a1f8c70acc1b95fd11dae6488388d0d605b9ebb74a5af2a83d0",
+  "d_exact.json": "74444442b5150dff0412834861170cfeec7a71bc8ac728d14671fb640df3520d"
+ },
+ "corpus": [
+  {
+   "code": "ED",
+   "title": "Elephants Dream",
+   "year": 2006,
+   "file": "elephantsdream-720-h264-st-aac.mov",
+   "bytes": 152496338,
+   "sha256": "6fc295e0e92835316ec95929ccfc664b05ed2434f673e6f5e684f98f7558a547",
+   "raster": "720x405",
+   "frames": 15693,
+   "licence": "Creative Commons; the version is not stated on the page fetched",
+   "licence_page_sha256": "96a8277544d978deb79503c2b82667501429d6c1cbeeabb9adba5eaa5cb01e36",
+   "licence_fetched": "2026-11-04T01:38:17Z"
+  },
+  {
+   "code": "BBB",
+   "title": "Big Buck Bunny",
+   "year": 2008,
+   "file": "big_buck_bunny_720p_h264.mov, read from the archive whose SHA-256 is given",
+   "bytes": 416751396,
+   "sha256": "b0c9ade80b086179feee41514929de583966eec87c703575d97f051f88b33b67",
+   "raster": "1280x720",
+   "frames": 14315,
+   "licence": "Creative Commons Attribution 3.0",
+   "licence_page_sha256": "bd9eb73ddec647bbb486c47316163042a9655a8483358c093121341d0dc080cf",
+   "licence_fetched": "2026-11-04T01:38:17Z"
+  },
+  {
+   "code": "SIN",
+   "title": "Sintel",
+   "year": 2010,
+   "file": "the 1080p release",
+   "bytes": 1172428172,
+   "sha256": "fae11ed41b2c99dfbeabce7ea5458446e21e7b840a982f510f2341919c3d9920",
+   "raster": "1920x818",
+   "frames": 21312,
+   "licence": "Creative Commons Attribution 3.0",
+   "licence_page_sha256": "67f30f30c55022c6f7de66beb997f04117b24f7aca928a912286a53edaa4b627",
+   "licence_fetched": "2026-11-03T12:26:59Z"
+  },
+  {
+   "code": "TOS",
+   "title": "Tears of Steel",
+   "year": 2012,
+   "file": "the 1080p release",
+   "bytes": 583774083,
+   "sha256": "99486359be7e3681168a0fe94e1cbb0284c48b57b2a3ce7df4fa75e185987a15",
+   "raster": "1920x800",
+   "frames": 17620,
+   "licence": "Creative Commons Attribution 3.0",
+   "licence_page_sha256": "c3defc0521e16af0a3afb6a96d084517620adf6434038e6a3a2ba1cf50f86947",
+   "licence_fetched": "2026-11-03T12:26:44Z"
+  }
+ ],
+ "analysis_raster": "luma at 640x360, area scaling, 8-bit limited range as decoded; 128x72 by 5x5 area mean (float) and 64x36 by 10x10 area mean (rounded to 8 bits) derived from it",
+ "tools": {
+  "ffmpeg": "9.0.2",
+  "svt_av1": "4.2.0",
+  "dav1d": "1.5.4",
+  "libvmaf": "3.2.1, model vmaf_v0.6.1",
+  "python": "3.14.4",
+  "numpy": "2.3.5"
+ },
+ "host": "a shared Linux x86_64 build host, 32 threads; each step in a user scope with MemoryMax=64G, CPUWeight=50 and nice 10; BLAS capped at 8 threads",
+ "timeline": [
+  {"utc": "2026-11-04T01:38:17Z", "event": "licence pages fetched"},
+  {"utc": "2026-11-04T01:43:38Z", "event": "the rate proxy's target PSNRs (30, 35, 40 and 45 dB) recorded with the basis script in a snapshot of the measurement scripts"},
+  {"utc": "2026-11-04T01:49:04Z", "event": "moment accumulation finished: the last moments file written; no figure printed"},
+  {"utc": "2026-11-04T01:58Z", "event": "the decision rule written, before any result of measurements (a) or (c) was seen; attested, not committed before the results"},
+  {"utc": "2026-11-04T01:58:51Z", "event": "the residual-coding script last changed"},
+  {"utc": "2026-11-04T02:00:43Z", "event": "the first result file written: a_basis.json"},
+  {"utc": "2026-11-04T02:01:24Z", "event": "the basis script's only later change: a print format"},
+  {"utc": "2026-11-04T02:38:59Z", "event": "c_rd.json written"}
+ ],
+ "rule": {
+  "go": {"bd_rate_max": -10, "metrics": ["psnr", "vmaf"], "titles_min": 3, "breakeven_titles_per_sink_max": 5},
+  "k1": {"below_dct_pct": 5, "targets": [35, 40], "of": "the catalogue basis's rate proxy on the median title, at every granularity"},
+  "k2": {"bd_rate_above": 0, "metrics": ["psnr", "vmaf"], "of": "the catalogue-basis residual and its coefficients against AV1 through the same pipeline, on the median title"},
+  "k3": {"runtime_pct_below": 1, "thumb_psnr_min": 40, "full_psnr_min": 35, "of": "non-flat material recurring across titles, per title"}
+ },
+ "verdicts": {"go": false, "k1": true, "k2": true, "k3": true, "no_psnr_saving_any_title": true}
+}
diff --git a/wai/research/syfai-input.md b/wai/research/syfai-input.md
index 7edbfbe89..f0d83952b 100644
--- a/wai/research/syfai-input.md
+++ b/wai/research/syfai-input.md
@@ -517,6 +517,15 @@ to say what each one reaches. Neither is a defect.
   that. **No WAI capability is registered as using it** (determinism-tiers §5
   lists none). A contribution should say that plainly rather than imply
   coverage.
+  *Corrected 2026-11-03, body read:* the body states what happens to the
+  reconstruction: "floating-point differences still cause gradual divergence in
+  the features and reconstructed frames", and "without a synchronization
+  mechanism, the divergence can grow arbitrarily large" (§3.2). It bounds the
+  divergence with periodic I-frames that "fully resynchronize encoder and
+  decoder feature states" (§3.2), and those I-frames are decoded by the same
+  float model (Appendix B.8). So the route guarantees the entropy stage and not
+  the reconstruction, which the paper states and does not leave to inference; a
+  resynchronisation re-anchors the picture to one float decode.
 - **Reproducibility by cross-check.** The IEEE 1857.11 call for proposals came
   out of the working subgroup's meeting on 9 Dec 2021, and its timeline lists
   the announcement on 21 Dec 2021. It states that "every proponent who submits
@@ -926,7 +935,8 @@ Read on 2026-09-27 unless noted.
    2510.13867. The PDF was read for the conformance-point, integer-pipeline and
    excluded-features passages. <https://arxiv.org/abs/2510.13867>
 7. *MLVC: Multi-platform Learned Video Codec for Real-World Deployment*, arXiv
-   2606.28027, ECCV 2026. Abstract page only.
+   2606.28027, ECCV 2026. Abstract page only. *Corrected 2026-11-03:* the body's
+   §3.1, §3.2, Appendix B.8 and Supplement D.1 (version 2) were read that day.
    <https://arxiv.org/abs/2606.28027>
 8. IEEE 1857.11 call for proposals, the outcome of the 9 Dec 2021 working
    subgroup meeting; its timeline lists the announcement on 21 Dec 2021 (the
diff --git a/wai/tools/wai_catalogue_prior_verify.py b/wai/tools/wai_catalogue_prior_verify.py
new file mode 100644
index 000000000..bd6ebb520
--- /dev/null
+++ b/wai/tools/wai_catalogue_prior_verify.py
@@ -0,0 +1,286 @@
+#!/usr/bin/env python3
+"""Check the catalogue-prior record against its own readings, and every figure the text
+quotes from it.
+
+    python3 -I wai_catalogue_prior_verify.py <record-dir> [<text.md> ...]
+    python3 -I wai_catalogue_prior_verify.py --summary <record-dir>
+
+Written from the procedure in research/catalogue-prior.md, sharing no code with the
+measurement scripts that wrote the readings. It checks each reading file against the
+SHA-256 the record's run.json states. From c_rd.json's rate and quality points alone it
+recomputes every BD-rate of the residual arms and of the two AV1 bars against each other:
+by the run's own method (the natural log of bytes fitted as a least-squares cubic in
+quality, each fit integrated over the quality range the two curves share), here solved in
+exact rational arithmetic, and by piecewise-linear interpolation of the same points. The
+cubic figures must equal the ones the run stored. It recomputes the rate proxy of
+measurement (a) from a_basis.json and the recurrence of measurement (b) from d_exact.json
+and b_recur.json, applies the decision rule run.json records (GO, K1, K2, K3) and requires
+the verdicts it states. Then each `<!--cat:PATH-->` marker in the text files must follow
+a number equal to the computed value at PATH, rounded as printed. Standard library only.
+Exit 0 when everything holds, 1 with one line per finding. With `--summary` it prints
+the computed values as JSON and checks nothing.
+"""
+import hashlib
+import json
+import math
+import os
+import re
+import sys
+from fractions import Fraction
+
+FINDINGS = []
+
+TITLES = ("ED", "BBB", "SIN", "TOS")
+ARMS = ("cat0", "cat64", "cat256", "self64")
+CATALOGUE_ARMS = ("cat0", "cat64", "cat256")
+K_OF = {"cat0": 0, "cat64": 64, "cat256": 256, "self64": 64}
+METRICS = ("psnr", "vmaf")
+GRANULARITIES = ("frame128x72", "block8", "block16", "block8_diff")
+READINGS = ("a_basis.json", "b_recur.json", "c_rd.json", "d_exact.json")
+BASIS_DIMS = 128 * 72
+
+
+def bad(msg):
+    FINDINGS.append(msg)
+
+
+def median(xs):
+    xs = sorted(xs)
+    n = len(xs)
+    return xs[(n - 1) // 2] if n % 2 == 1 else 0.5 * (xs[n // 2 - 1] + xs[n // 2])
+
+
+def cubic_fit(xs, ys):
+    """The least-squares cubic through the points, c0 + c1 x + c2 x^2 + c3 x^3, solved from
+    the normal equations in exact rational arithmetic over the floats given."""
+    X = [Fraction(x) for x in xs]
+    Y = [Fraction(y) for y in ys]
+    m = [[sum(x ** (i + j) for x in X) for j in range(4)] + [sum(y * x ** i for x, y in zip(X, Y))] for i in range(4)]
+    for col in range(4):
+        piv = next(r for r in range(col, 4) if m[r][col] != 0)
+        m[col], m[piv] = m[piv], m[col]
+        for r in range(4):
+            if r != col and m[r][col] != 0:
+                f = m[r][col] / m[col][col]
+                m[r] = [a - f * b for a, b in zip(m[r], m[col])]
+    return [m[i][4] / m[i][i] for i in range(4)]
+
+
+def cubic_integral(c, lo, hi):
+    def prim(x):
+        return sum(ci * x ** (i + 1) / (i + 1) for i, ci in enumerate(c))
+    return prim(hi) - prim(lo)
+
+
+def linear_integral(qs, ys, lo, hi):
+    """The integral over [lo, hi] of the piecewise-linear interpolant of (q, y), points
+    taken in order of q."""
+    pts = sorted(zip(qs, ys))
+    total = []
+    for (q0, y0), (q1, y1) in zip(pts, pts[1:]):
+        a, b = max(q0, lo), min(q1, hi)
+        if b <= a or q1 == q0:
+            continue
+        ya = y0 + (y1 - y0) * (a - q0) / (q1 - q0)
+        yb = y0 + (y1 - y0) * (b - q0) / (q1 - q0)
+        total.append(0.5 * (ya + yb) * (b - a))
+    return math.fsum(total)
+
+
+def bd_rate(r_ref, q_ref, r_test, q_test, how):
+    """The percentage of bytes the test curve spends more than the reference at the same
+    quality, over the quality range both reach; None when they share none."""
+    lo, hi = max(min(q_ref), min(q_test)), min(max(q_ref), max(q_test))
+    if hi <= lo:
+        return None
+    lr, lt = [math.log(r) for r in r_ref], [math.log(r) for r in r_test]
+    if how == "cubic":
+        flo, fhi = Fraction(lo), Fraction(hi)
+        avg = float((cubic_integral(cubic_fit(q_test, lt), flo, fhi) - cubic_integral(cubic_fit(q_ref, lr), flo, fhi)) / (fhi - flo))
+    else:
+        avg = (linear_integral(q_test, lt, lo, hi) - linear_integral(q_ref, lr, lo, hi)) / (hi - lo)
+    return (math.exp(avg) - 1) * 100
+
+
+def with_median(per_title):
+    out = dict(per_title)
+    vals = [v for v in per_title.values() if v is not None]
+    out["median"] = median(vals) if vals else None
+    return out
+
+
+def summarise(rec):
+    c, a, d, b = rec["c_rd.json"], rec["a_basis.json"], rec["d_exact.json"], rec["b_recur.json"]
+    s = {}
+
+    def pts(t, arm, key):
+        return [p[key] for p in c[t]["arms"][arm]["points"]]
+
+    def bd(t, ref, arm, metric, how, bytes_key="total_bytes"):
+        return bd_rate(pts(t, ref, "total_bytes"), pts(t, ref, metric), pts(t, arm, bytes_key), pts(t, arm, metric), how)
+
+    for how, key in (("cubic", "bd"), ("linear", "bd_linear")):
+        s[key] = {}
+        for arm in ARMS + ("base8", "base10"):
+            for ref in ("base10", "base8"):
+                if arm == ref:
+                    continue
+                for m in METRICS:
+                    s[key].setdefault(arm, {}).setdefault(ref, {})[m] = with_median({t: bd(t, ref, arm, m, how) for t in TITLES})
+    s["bd_av1only"] = {arm: {ref: {m: with_median({t: bd(t, ref, arm, m, "cubic", "av1_bytes") for t in TITLES}) for m in METRICS}
+                             for ref in ("base10", "base8")} for arm in ARMS}
+    s["linear_shift_psnr_max"] = max(abs(s["bd_linear"][arm][ref]["psnr"][t] - s["bd"][arm][ref]["psnr"][t])
+                                     for arm in ARMS for ref in ("base10", "base8") for t in TITLES)
+    s["basis_bytes"] = {arm: (K_OF[arm] + 1) * BASIS_DIMS * 2 for arm in ARMS}
+    s["basis_only_psnr"] = {arm: {"min": min(c[t]["arms"][arm]["basis_only_psnr"] for t in TITLES),
+                                  "max": max(c[t]["arms"][arm]["basis_only_psnr"] for t in TITLES)} for arm in ARMS}
+    s["side_pct"] = {arm: {"min": min(p["side_bytes"] / p["total_bytes"] * 100 for t in TITLES for p in c[t]["arms"][arm]["points"]),
+                           "max": max(p["side_bytes"] / p["total_bytes"] * 100 for t in TITLES for p in c[t]["arms"][arm]["points"])}
+                     for arm in ARMS}
+    s["clipped_max"] = max(c[t]["arms"][arm]["clipped_frac"] for t in TITLES for arm in ARMS)
+    base8 = [p["total_bytes"] for t in TITLES for p in c[t]["arms"]["base8"]["points"]]
+    s["base8_mb"] = {"min": min(base8) / 1e6, "max": max(base8) / 1e6}
+    s["clip"] = {t: {"start_s": c[t]["clip_start_s"], "frames": c[t]["frames"]} for t in TITLES}
+
+    # Measurement (a): the rate proxy of the catalogue basis and of the per-title basis,
+    # against the DCT's, at each target PSNR; and what the catalogue mean and the DC capture.
+    s["proxy"] = {}
+    for g in GRANULARITIES:
+        for basis, key in (("catalogue", "catalogue"), ("per_title", "per_title")):
+            for target in ("35", "40"):
+                per = {t: (a[g][t][basis]["bpp_at_psnr"][target] / a[g][t]["dct"]["bpp_at_psnr"][target] - 1) * 100 for t in TITLES}
+                s["proxy"].setdefault(g, {}).setdefault(key, {})[target] = with_median(per)
+    fr = a["frame128x72"]
+    s["mean_frame"] = {t: {"captured_ac_pct": fr[t]["catalogue"]["captured_ac"]["0"] * 100,
+                           "captured_incl_dc_pct": fr[t]["catalogue"]["captured_incl_dc"]["0"] * 100,
+                           "k16_incl_dc_pct": fr[t]["catalogue"]["captured_incl_dc"]["16"] * 100,
+                           "dct_k16_incl_dc_pct": fr[t]["dct"]["captured_incl_dc"]["16"] * 100} for t in TITLES}
+    for k in ("captured_incl_dc_pct", "k16_incl_dc_pct"):
+        s["mean_frame"][k] = {"min": min(s["mean_frame"][t][k] for t in TITLES), "max": max(s["mean_frame"][t][k] for t in TITLES)}
+
+    # Measurement (b): what recurs, from d_exact.json's runtime and byte shares and
+    # b_recur.json's exact and 40 dB counts.
+    s["recur"] = {}
+    for t in TITLES:
+        r = {}
+        for part in ("flat", "cross35", "self_far35"):
+            r[part] = {"runtime_pct": d[t][part]["runtime_pct"], "bytes_pct": d[t][part]["bytes_pct"], "frames": d[t][part]["frames"]}
+        r["exact_nonflat"] = b[t]["exact_thumb_cross_nonflat"]
+        r["full_ge40"] = b[t]["cross_ge35"]["full_psnr_ge40"]
+        r["av1_mb"] = d[t]["av1_bytes_crf36"] / 1e6
+        s["recur"][t] = r
+    s["recur"]["cross35_runtime_pct_max"] = max(s["recur"][t]["cross35"]["runtime_pct"] for t in TITLES)
+    s["recur"]["cross35_bytes_pct_max"] = max(s["recur"][t]["cross35"]["bytes_pct"] for t in TITLES)
+    near = [d[t]["cross35"]["thumb_std_median"] for t in TITLES if d[t]["cross35"]["frames"] > 0]
+    ordinary = [d[t]["nonflat_thumb_std_median"] for t in TITLES]
+    s["recur"]["cross35_spread"] = {"min": min(near), "max": max(near)}
+    s["recur"]["ordinary_spread"] = {"min": min(ordinary), "max": max(ordinary)}
+
+    # What the rule's figures come to: how far the GO bar is from cat64, what the GO bar
+    # against base10 is against plain AV1, and how close the block-scale bases are.
+    s["go_gap_points"] = {arm: min(s["bd"][arm]["base10"][m]["median"] for m in METRICS) - run_rule_go(rec) for arm in CATALOGUE_ARMS}
+    s["go_vs_base8_pct"] = ((1 + run_rule_go(rec) / 100) * (1 + s["bd"]["base10"]["base8"]["psnr"]["median"] / 100) - 1) * 100
+    blocks = ("block8", "block16", "block8_diff")
+    s["proxy_block_median_abs_max"] = max(abs(s["proxy"][g]["catalogue"][x]["median"]) for g in blocks for x in ("35", "40"))
+    s["proxy_block_title_abs_max"] = max(abs(s["proxy"][g]["catalogue"][x][t]) for g in ("block8", "block16") for x in ("35", "40") for t in TITLES)
+    s["mean_frame"]["k16_vs_dct_points_max"] = max(abs(s["mean_frame"][t]["k16_incl_dc_pct"] - s["mean_frame"][t]["dct_k16_incl_dc_pct"]) for t in TITLES)
+    return s
+
+
+def run_rule_go(rec):
+    return rec["run.json"]["rule"]["go"]["bd_rate_max"]
+
+
+def verdicts(s, rule):
+    """The decision rule as run.json records it, applied to the computed values."""
+    go = rule["go"]
+    k1, k2, k3 = rule["k1"], rule["k2"], rule["k3"]
+    go_met = any(sum(1 for t in TITLES if all(s["bd"][arm]["base10"][m][t] <= go["bd_rate_max"] for m in METRICS)) >= go["titles_min"]
+                 for arm in CATALOGUE_ARMS)
+    k1_holds = all(not all(s["proxy"][g]["catalogue"][str(target)]["median"] <= -k1["below_dct_pct"] for target in k1["targets"])
+                   for g in GRANULARITIES)
+    k2_holds = all(s["bd"][arm]["base10"][m]["median"] > k2["bd_rate_above"] for arm in CATALOGUE_ARMS for m in METRICS)
+    k3_holds = all(s["recur"][t]["cross35"]["runtime_pct"] < k3["runtime_pct_below"] for t in TITLES)
+    no_saving = all(s["bd"][arm][ref]["psnr"][t] > 0 for arm in ARMS for ref in ("base10", "base8") for t in TITLES)
+    return {"go": go_met, "k1": k1_holds, "k2": k2_holds, "k3": k3_holds, "no_psnr_saving_any_title": no_saving}
+
+
+def lookup(s, path):
+    v = s
+    for part in path.split("."):
+        if isinstance(v, dict) and part in v:
+            v = v[part]
+        else:
+            return KeyError
+    return v
+
+
+def check_text(s, files):
+    marker = re.compile(r"([+\-−]?\d[\d,]*(?:\.\d+)?)\s*(%|dB|MB|B)?\**\s*<!--cat:([\w.\-/]+)-->")
+    n = 0
+    for f in files:
+        text = open(f, encoding="utf-8").read()
+        for m in marker.finditer(text):
+            n += 1
+            printed, path = m.group(1).replace("−", "-").replace(",", ""), m.group(3)
+            v = lookup(s, path)
+            if v is KeyError or not isinstance(v, (int, float)) or isinstance(v, bool):
+                bad(f"{f}: <!--cat:{path}--> names no number")
+                continue
+            dec = len(printed.split(".")[1]) if "." in printed else 0
+            want = round(v, dec)
+            if abs(float(printed) - want) > 0.5 * 10 ** -dec + 1e-12 or (dec == 0 and float(printed) != round(v)):
+                bad(f"{f}: prints {m.group(1)} for {path}, the readings give {v} ({want} at {dec} decimals)")
+        every = len(re.findall(r"<!--cat:", text))
+        found = len(marker.findall(text))
+        if every != found:
+            bad(f"{f}: {every - found} cat marker(s) do not directly follow a number")
+    return n
+
+
+def main(argv):
+    summary_only = len(argv) > 1 and argv[1] == "--summary"
+    args = argv[2:] if summary_only else argv[1:]
+    if not args:
+        sys.exit(__doc__)
+    d = args[0]
+    run = json.load(open(os.path.join(d, "run.json"), encoding="utf-8"))
+    if run.get("format") != "wai-catalogue-prior/1":
+        sys.exit(f"not a catalogue-prior record: {run.get('format')!r}")
+    rec = {}
+    for name in READINGS:
+        raw = open(os.path.join(d, name), "rb").read()
+        if hashlib.sha256(raw).hexdigest() != run["readings"].get(name):
+            bad(f"{name}: SHA-256 {hashlib.sha256(raw).hexdigest()} is not the one run.json states")
+        rec[name] = json.loads(raw)
+    if sorted(rec["c_rd.json"]) != sorted(TITLES):
+        bad(f"c_rd.json: titles {sorted(rec['c_rd.json'])}, not {sorted(TITLES)}")
+    rec["run.json"] = run
+    s = summarise(rec)
+    if summary_only:
+        print(json.dumps(s, indent=1, sort_keys=True))
+        return 0
+    # The cubic BD-rates the run stored, recomputed from the points.
+    for t in TITLES:
+        for arm, got in rec["c_rd.json"][t]["bd_vs_base10"].items():
+            for m in METRICS:
+                mine = s["bd"][arm]["base10"][m][t]
+                if got[m] is None or mine is None or abs(got[m] - mine) > 1e-6 * max(1.0, abs(mine)):
+                    bad(f"c_rd.json: {t} {arm} {m} BD-rate against base10 stored {got[m]}, recomputed {mine}")
+    v = verdicts(s, run["rule"])
+    for k, want in run["verdicts"].items():
+        if v.get(k) != want:
+            bad(f"run.json states verdict {k} = {want}; the readings give {v.get(k)}")
+    quoted = check_text(s, args[1:])
+    for f in FINDINGS:
+        print(f)
+    if FINDINGS:
+        print(f"{len(FINDINGS)} finding(s)")
+        return 1
+    print(f"ok: 4 readings match run.json; {sum(1 for t in TITLES for _ in rec['c_rd.json'][t]['bd_vs_base10']) * 2} stored "
+          f"BD-rates recomputed; verdicts {json.dumps(v, sort_keys=True)}; {quoted} quoted figure(s) match")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv))
-- 
2.54.0 (Apple Git-157)


From 3d1582327a5e850a9c049b71bf16acc353411a6a Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:42:06 -0500
Subject: [PATCH 13/29] =?UTF-8?q?determinism-tiers=20=C2=A73:=20no=20tier?=
 =?UTF-8?q?=20lies=20between=20the=20two,=20and=20a=20session=20claim's=20?=
 =?UTF-8?q?decode=20path=20has=20three=20refusals?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

A reconstruction kept within a bound of the exact one between points at which
it is made exact again is measured on named devices and a named corpus, and a
sink cannot check it from what it received. So §3 says, normatively, that no
tier, class or claim states such a bound, and three rules follow: a decode of a
NeuralIntegerExact capability in reduced precision or float is not a decode of
it; a capability whose payload holds several frame types takes the class of its
weakest; and a clip digest, a stage hash or a binding over a sequence covers
only frames each decoded at decode-equivalence. SPEC §7 points to it.

A reconstruct-session decoder entry's `variant` now names the path that ran:
`exact`, `float`, or null for a capability with one path. Rule 13 refuses a
NeuralIntegerExact capability named with another path
(session_variant_not_exact); rule 14 refuses a float path under a pin of
decode-equivalence (session_tier_exceeds_variant); and against the stage
claims, a stage hash over a base an entry decoded at a weaker step, or over a
base or layer decoded by a path that is not exact, is refused
(session_stage_hash_not_exact).

claim-conformance gains a vector for each refusal and accepted ones for both
of wai.image.jpegai's paths, appended after the earlier vectors, which a test
pins by digest. claim_verify.py reads the same rules, and differential.py
generates sessions around them.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md                              |  12 ++
 wai/SPEC.md                                   |   6 +
 wai/claim-conformance/README.md               |  26 +++
 wai/claim-conformance/claim_verify.py         |  28 ++-
 wai/claim-conformance/differential.py         |  33 +++-
 wai/claim-conformance/vectors.json            | 137 ++++++++++++++
 wai/extensions/determinism-tiers.md           |  32 ++++
 wai/extensions/jwp-receipts.md                |  30 ++-
 wai/wai-rs/src/claim_receipt.rs               |  30 ++-
 .../src/claim_receipt/corpus/session.rs       | 172 ++++++++++++++++++
 wai/wai-rs/src/claim_receipt/session.rs       |  51 +++++-
 11 files changed, 543 insertions(+), 14 deletions(-)

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index fcf8d63cc..68877e9e2 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -8,6 +8,18 @@ wire format may change and such changes are recorded here.
 
 ### A decode reports the path that ran, and no tier lies between the two
 
+- **No tier between the two** (determinism-tiers §3, normative). The tier vocabulary stays closed. A decode of a
+  `NeuralIntegerExact` capability in reduced precision or floating point is not a decode of it; a capability whose
+  payload holds several frame types takes the class of its weakest; and a clip digest, a stage hash or a reconstruction
+  binding over a sequence covers only frames each decoded at `decode-equivalence`.
+- **A session claim states its decode path** ([extensions/jwp-receipts.md](extensions/jwp-receipts.md) §2.4.12). A
+  decoder entry's `variant` is `exact`, `float`, or `null` for a capability with one path. Rule 13 refuses a
+  `NeuralIntegerExact` capability named with another path (`session_variant_not_exact`); rule 14 refuses a `float` path
+  under a pin of `decode-equivalence` (`session_tier_exceeds_variant`); and against the stage claims, a stage hash over
+  a base an entry decoded at a weaker step, or over a base or layer decoded by a path that is not exact, is refused
+  (`session_stage_hash_not_exact`). `claim-conformance/` gains a vector for each, beside accepted ones for both
+  `wai.image.jpegai` paths, appended after the earlier vectors, which a test pins by digest; `claim_verify.py` and
+  `differential.py` read the same rules.
 - **determinism-tiers §1 and §7.1** record the five published routes to a reproducible stage, from the paper sections
   read, and a line that reaches no tier: float synchronised on a tested hardware class (arXiv 2608.00483). The research
   notes `syfai-input.md` §5 and `2026-09-global-sweep.md` (its row of arXiv 2608.00483, G7 and F5) are corrected against
diff --git a/wai/SPEC.md b/wai/SPEC.md
index 7e5a9fb69..882d7e08c 100644
--- a/wai/SPEC.md
+++ b/wai/SPEC.md
@@ -2685,6 +2685,12 @@ registered at `decode-equivalence`. Two weaker properties MUST NOT back
   `replicate` resting on them promises something no sink can check from what it
   received.
 
+No tier lies between these: a reconstruction kept within a bound of the exact one
+between points at which it is made exact again is no tier and backs nothing, a
+reduced-precision decode of a `NeuralIntegerExact` capability is not a decode of it,
+and a clip digest covers only frames each decoded at `decode-equivalence`
+([determinism-tiers](extensions/determinism-tiers.md) §3, *No tier between the two*).
+
 The reference impl factors the halves of the contract into ORT-free, pure-i64
 modules a codec composes: `int_entropy` (the deterministic scale→bucket step
 that removes the float-σ desync), `int_synth` (integer synthesis), and the
diff --git a/wai/claim-conformance/README.md b/wai/claim-conformance/README.md
index 86b47d305..99c6e5eff 100644
--- a/wai/claim-conformance/README.md
+++ b/wai/claim-conformance/README.md
@@ -349,6 +349,32 @@ with neither is a hole; there is none.
 | SPEC §7.1, *Refinement components*, *Selection* (every `WaiMulti::check_component_roles` code: the refinement codes, `refines-on-non-refinement`, `refinement-only-as-component`, the companion codes, `component-id-malformed`, `duplicate-component-id`) | a sink offers a `WAI2`'s refinements only when every component conforms | rule 5, the whole envelope | `session_wai2_refinement_identity_missing`, `session_wai2_refines_base_missing`, `session_wai2_refines_base_companion`, `session_wai2_layer_base_not_admitted`, `session_wai2_refinement_prev_missing`, `session_wai2_refinement_before_selectable`, `session_roles_refinement_before_selectable_exploded`, `session_wai2_refines_on_a_rendition`, `session_roles_base_without_sha256`, `session_roles_layer_before_its_prev`, `session_roles_unbound_companion`, `session_roles_record_companion`, `session_roles_derivation_companion`, `session_roles_record_rendition`; accepted: `session_roles_companion_and_refinement_accept`, `session_roles_companion_by_id_and_refinement_accept` |
 | SPEC §7.1, *Companion components* | a companion that breaks rules 1–3 or its `companion_of_id`; an enhancement layer whose base is not unique | for the rendition picked, no bearing: a sink does not apply the companion and may still present the rendition, the output's digest holding whatever was applied; for the envelope's refinements, the row above | `session_roles_unbound_companion` |
 
+### The decode path that ran
+
+A decoder entry's `variant` states which path of its decoder ran: `exact`, `float`, or
+`null` for a capability with one path (jwp-receipts §2.4.12). Three rules close the door
+determinism-tiers §3 closes (*No tier between the two*), each with a refusal here. The
+vectors are appended after every earlier vector, which a test pins by digest:
+
+1. Rule 13, `session_variant_not_exact`: a `NeuralIntegerExact` capability named with a
+   path other than its integer one. `session_variant_not_exact` names `wai.neural.int_hyper`
+   with `float` under a pin of `decode-equivalence`, and `session_variant_not_exact_unpinned`
+   with an unregistered term and no pin. Accepted: `session_variant_exact_integer_accept`,
+   the integer path stated as `exact`.
+2. Rule 14, `session_tier_exceeds_variant`: a `float` path stated under a pin of
+   `decode-equivalence`, over a capability the registry does not classify, decoded by a
+   module, which rule 6 does not reach. `session_tier_exceeds_class_before_variant` gives
+   `wai.image.jpegai`'s float path the same pin, which rule 6 refuses first. Accepted:
+   `session_variant_float_accept` and `session_variant_exact_accept`, `wai.image.jpegai`'s
+   two paths stated as they ran.
+3. Against the stage claims, `session_stage_hash_not_exact`: the model session's stage
+   claim stated at `decode-equivalence`, with its stage hash, a clip digest, while its base
+   was also decoded through its fallback, a capability held to a tolerance
+   (`session_stage_hash_not_exact`), while one of its layers is stated as decoded on a
+   `float` path (`session_stage_hash_over_a_float_layer`), or while its base is stated as
+   decoded with no pin (`session_stage_hash_over_an_unpinned_base`). Accepted:
+   `session_stage_hash_exact_accept`.
+
 ## Steps 1 and 2: types, and the order a body is read in
 
 §3.1 step 2 checks exactly the type a member table gives, member by member in the
diff --git a/wai/claim-conformance/claim_verify.py b/wai/claim-conformance/claim_verify.py
index bdae6336f..8ea18b832 100644
--- a/wai/claim-conformance/claim_verify.py
+++ b/wai/claim-conformance/claim_verify.py
@@ -731,6 +731,14 @@ def session_rules(b):
     if e is not None:
         need(e["joules_micro"] != 0, 4, "energy_zero_figure")
         acquisition(e)
+    # Rule 13: a NeuralIntegerExact capability is decoded by its integer path only, which
+    # an entry states as variant null or exact (determinism-tiers §3, rule 1).
+    need(not any(d["variant"] is not None and d["variant"] != "exact" and replicate(d["capability"]) == "neural-int-exact"
+                 for d in ds), 4, "session_variant_not_exact")
+    # Rule 14: frames a float path decoded are not decode-equivalent, whatever the pin
+    # declares (rule 2 there).
+    need(not any(d["variant"] == "float" and d["tier"] == "decode-equivalence" for d in ds), 4,
+         "session_tier_exceeds_variant")
 
 
 # ------------------------------------------- §2.4.12 step 6: what a session is checked against
@@ -1276,7 +1284,8 @@ def check_session_envelopes(c, envs, objs, ps):
 def check_session_stage_claims(c, wires, ps, objs):
     """§2.4.12 against the stage claims. A sink names one for each epoch of a stage set,
     and bases over the same units of the title each have their own: each claim's
-    identity and objects first, then with the outputs each claim's units and its count
+    identity and objects first, then a stage hash against the decoder entries that
+    decoded its base and layers, then with the outputs each claim's units and its count
     from below, and each unit of a set from above, by the objects listed as applied or
     partial by the claims whose sets hold it."""
     if not isinstance(wires, list) or not all(isinstance(w, str) for w in wires):
@@ -1304,6 +1313,23 @@ def check_session_stage_claims(c, wires, ps, objs):
         for h in [s["subject"]] + [o["object"] for o in s["body"]["objects"]]:
             at.setdefault(h, set()).update(named)
     need(all(d["first_unit"] in at[d["object"]] for d in b["decoders"] if d["object"] in at), 6, "session_first_unit_mismatch")
+    # A stage hash is a clip digest over the set's units (staged-delivery §5): every entry
+    # naming the set's base decodes it at decode-equivalence by that section's base step,
+    # and no entry naming the base or a layer the claim lists ran a path that is not exact
+    # (determinism-tiers §3, rule 3).
+    def base_exact(d):
+        r = replicate(d["capability"])
+        return r == "standard-defined" or (r == "neural-int-exact" and d["tier"] == "decode-equivalence")
+
+    for s, _ in claims:
+        if s["body"]["stage_hash"] is None:
+            continue
+        layers = {o["object"] for o in s["body"]["objects"]}
+        for d in b["decoders"]:
+            base = d["object"] == s["subject"]
+            exact_path = d["variant"] is None or d["variant"] == "exact"
+            need(not (base and not base_exact(d)) and not ((base or d["object"] in layers) and not exact_path),
+                 6, "session_stage_hash_not_exact")
     if ps is None:
         return
     fin = finals(ps)
diff --git a/wai/claim-conformance/differential.py b/wai/claim-conformance/differential.py
index fcb915247..aa37866ac 100644
--- a/wai/claim-conformance/differential.py
+++ b/wai/claim-conformance/differential.py
@@ -591,10 +591,11 @@ SESSION_ENERGY = [
 ]
 
 
-def session_stage_claim(rng, n_set, at_stage_1, signer, subject, late=0):
+def session_stage_claim(rng, n_set, at_stage_1, signer, subject, late=0, tier="none"):
     """A stage claim (§2.4.10) that passes its own rules: `n_set` units, `at_stage_1` of
     them at stage 1 through one applied first layer, the rest at stage 0; and, with
-    `late`, a first layer over the next `late` units that arrived late."""
+    `late`, a first layer over the next `late` units that arrived late. At `tier`
+    `decode-equivalence` it carries a stage hash, a clip digest of its units."""
     ubs = [n_set - at_stage_1, at_stage_1] if at_stage_1 else [n_set]
     objects = [] if not at_stage_1 else [{
         "layer": 1, "object": CV.blake3(b"differential: a stage layer").hex(), "units": {"from": 0, "count": at_stage_1},
@@ -604,8 +605,9 @@ def session_stage_claim(rng, n_set, at_stage_1, signer, subject, late=0):
                         "units": {"from": at_stage_1, "count": late}, "bytes": 650, "status": "late", "evidence": "none",
                         "reason": None})
     body = {"vocabulary": CV.STAGE_VOCABULARY, "op_requested": None, "op_committed": None, "op_delivered": None,
-            "presented": {"wai-stage:layers": 0 if ubs[0] else 1}, "units_by_stage": ubs, "tier": "none",
-            "stage_hash": None, "objects": objects, "conflicts": 0, "deadline_misses": 0, "bytes_envelopes": 5700}
+            "presented": {"wai-stage:layers": 0 if ubs[0] else 1}, "units_by_stage": ubs, "tier": tier,
+            "stage_hash": CV.blake3(b"differential: a clip digest").hex() if tier == "decode-equivalence" else None,
+            "objects": objects, "conflicts": 0, "deadline_misses": 0, "bytes_envelopes": 5700}
     c = {"kind": "claim", "class": "stage", "subject_type": "object", "subject": subject, "seq": 0,
          "issued_at": "2026-09-27T12:00:00Z", "body": body, "signer_id": "did:example:sink", "parent_receipt_hash": None}
     return CV.jcs(sign(c, signer))
@@ -721,8 +723,12 @@ def sessions(ins, rng, n):
             reached = sum(1 for u, p in fin.items() if f <= u < f + n_set and p["layers"] > 1)
             at1 = min(n_set, max(0, reached + rng.choice([0, 0, 1, -1])))
             signer = VIEWER if rng.random() < 0.9 else RELAY
-            base_hash = CV.blake3(b"differential: a base %d" % i).hex()
-            w = session_stage_claim(rng, n_set, at1, signer, base_hash)
+            # Its base: one no entry names, or the entry object, which entries name and decode
+            # at whatever step their capability, pin and path give; at decode-equivalence the
+            # claim carries a stage hash (§2.4.12 step 6, against the stage claims, item 5).
+            base_hash = CV.blake3(b"differential: a base %d" % i).hex() if rng.random() < 0.7 else entry
+            stage_tier = rng.choice(["none", "none", "decode-equivalence"])
+            w = session_stage_claim(rng, n_set, at1, signer, base_hash, tier=stage_tier)
             if rng.random() < 0.05:
                 w = w.replace('"tier":"none"', '"tier":"nonе"')
             stage_wires.append(w)
@@ -752,7 +758,7 @@ def sessions(ins, rng, n):
         # One or two defects, in the body or in a context.
         inert = None
         for _ in range(rng.choice([0, 1, 1, 2])):
-            k = rng.randrange(31)
+            k = rng.randrange(33)
             ds = body["decoders"]
             if k == 0:
                 body["primary_units"] += rng.choice([1, -1]) if body["primary_units"] else 1
@@ -833,6 +839,19 @@ def sessions(ins, rng, n):
                 o = rng.choice(objects)
                 o["counted"] = rng.choice([None, 0, (o["counted"] or 0) + 1, max(0, (o["counted"] or 0) - 1)])
                 restate(body, objects)
+            elif k == 31 and ds:
+                # The decode path an entry states (§2.4.12 rules 13 and 14): exact, float, a
+                # term this revision does not register, or none.
+                rng.choice(ds)["variant"] = rng.choice([None, "exact", "float", "fp16"])
+            elif k == 32 and ds:
+                # A float path under a pin of either tier, or none, through a capability the
+                # registry classifies or one it does not.
+                d = rng.choice(ds)
+                d["variant"] = rng.choice(["float", "exact"])
+                d["pin"], d["tier"] = rng.choice([(CV.blake3(b"differential: a pin").hex(), "decode-equivalence"),
+                                                  (CV.blake3(b"differential: a pin").hex(), "entropy-consistency"), (None, None)])
+                if rng.random() < 0.3:
+                    d["capability"] = rng.choice(["org.example.picture", "wai.image.jpegai"])
         if rng.random() < 0.5:
             ctx["redecoded"] = [{"index": j, "out_digest": rng.choice([p["out_digest"], CV.blake3(b"differential: redecoded").hex()])}
                                 for j, p in enumerate(ps) if rng.random() < 0.3]
diff --git a/wai/claim-conformance/vectors.json b/wai/claim-conformance/vectors.json
index 07ea68684..0b1740c1b 100644
--- a/wai/claim-conformance/vectors.json
+++ b/wai/claim-conformance/vectors.json
@@ -24651,6 +24651,143 @@
         "step": 2,
         "reason": "unknown_member"
       }
+    },
+    {
+      "name": "session_variant_float_accept",
+      "about": "a picture decoded by wai.image.jpegai's float path, stated as variant float: the path that ran, beside the capability, which is registered at entropy-consistency",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.image.jpegai\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":\"float\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"343061bf444e80e3420dde80299ee8f4da440f269933abf85506e024eb5b8414839841cf942306399bacecc1d0fe5376df1519098378d0a400f6a667e3accc06\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "accept",
+        "link_hash": "3bac449e6ddf59c72fb2c6f7922fe12952c597db69c2c36c18a0ce2f8dbca86a"
+      }
+    },
+    {
+      "name": "session_variant_exact_accept",
+      "about": "the same picture decoded by wai.image.jpegai's exact decoder, stated as variant exact, under a pin of entropy-consistency",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.image.jpegai\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"entropy-consistency\",\"variant\":\"exact\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"f128bb5e54c0122b393661b58ffccd2cceec289e2cbc090a55e122988669bfe46839ec5a4199f3a30e44a2aed125a6cccceb966786cc33da9d25ce54c1b31d0f\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "accept",
+        "link_hash": "5e5d01814308da6c65efe318051861c3c6abd36662afb51ef09cdee7c59715c4"
+      }
+    },
+    {
+      "name": "session_variant_exact_integer_accept",
+      "about": "a NeuralIntegerExact capability whose entry states its integer path as variant exact, under a pin of decode-equivalence",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":\"exact\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"40639afc5ae903add05e95b3c30a0bb5f55affb86849d3f10c86b532f59f195f70464a73c747679f2ebf4456e341a5741b7a218d3160272bd7877ba937ac8f0e\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "accept",
+        "link_hash": "068b95433825974508a3cf0adfece27e78314207865a14130504fe02e4023d4a"
+      }
+    },
+    {
+      "name": "session_variant_not_exact",
+      "about": "a NeuralIntegerExact capability named with variant float: a decode of it in floating point is not a decode of it (determinism-tiers 3, rule 1)",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":\"float\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"675f06e109118c31bf9987cc257c4c244de7bf1318ade4284ddf108b77294503b9c1b4568dffd0ee448dd8a27b1b9304735d4592edfc09a68dd1fba40e8a4d0f\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "reject",
+        "step": 4,
+        "reason": "session_variant_not_exact"
+      }
+    },
+    {
+      "name": "session_variant_not_exact_unpinned",
+      "about": "the same with no pin, and a path named by a term this revision does not register: only null or exact names the integer path",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":\"fp16\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"95f4b49237eed8b2c76bfb0c9a00f9f8ffc5a8e2e6f47454e4a1c021c373a33fb16ddccfb3f69e7b98fe991028fff9e5d37d832b6ac50e35e28e3b28da468d06\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "reject",
+        "step": 4,
+        "reason": "session_variant_not_exact"
+      }
+    },
+    {
+      "name": "session_tier_exceeds_variant",
+      "about": "a capability the registry does not classify, decoded by a module on its float path and stated under a pin of decode-equivalence: a float path's frames are not decode-equivalent, whatever the pin declares (determinism-tiers 3, rule 2)",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"org.example.picture\",\"decoder\":\"module:sha256:5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e\",\"first_unit\":0,\"leaves\":1,\"module_status\":\"pinned\",\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":\"float\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"85e1d82d42125df5862e49a1ef44da98097a07a723c3fa63b39f1a03081fd9b700d52c2a02d583716ebc3b71d72a4825224eec0187bb324d23ef89504c5e9504\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "reject",
+        "step": 4,
+        "reason": "session_tier_exceeds_variant"
+      }
+    },
+    {
+      "name": "session_tier_exceeds_class_before_variant",
+      "about": "wai.image.jpegai's float path under a pin of decode-equivalence: the class rule, rule 6, refuses it first",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.image.jpegai\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":\"float\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"e4dc0e43885391ef620f38606f58847b2392c385652a269ab2bdefa04684851665958c49826cff1accfd57c052d005aaf0a72c8b42826c664e49b5e9c5a8fc01\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "reject",
+        "step": 4,
+        "reason": "session_tier_exceeds_class"
+      }
+    },
+    {
+      "name": "session_stage_hash_exact_accept",
+      "about": "the model session's stage set stated at decode-equivalence, with its stage hash: its base decoded only through a pinned integer capability, its layers through a residual refinement, every path exact",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":4,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":5,\"leaves\":1,\"module_status\":null,\"object\":\"13b04249eca2b1a9061dba8ca2846169d3f5fa3548d9e138a0cf2cb553798cf4\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"b62a8425b7fe3a085a14f2f2156ea22ae3b4349884d799b5afe66dccc7d4dd7f\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":1,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0c89804c9e944f3d69dbeadebc29214349ce7ddd74f09bf8b1a579217e3ec4bc0b91d8e5a78b8281d6907cbfb1e3443d4a8de050b9a3869797873b1ad6ed4b04\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
+      "context": {
+        "stage_claims": [
+          "{\"body\":{\"bytes_envelopes\":10607,\"conflicts\":0,\"deadline_misses\":2,\"objects\":[{\"bytes\":900,\"evidence\":\"none\",\"layer\":1,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":901,\"evidence\":\"none\",\"layer\":1,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":1}},{\"bytes\":902,\"evidence\":\"none\",\"layer\":1,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":2}},{\"bytes\":903,\"evidence\":\"none\",\"layer\":1,\"object\":\"792a12990c5207ca83a86b7420815f1f2d34918553a0130a4f9570a11354f2ca\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":3}},{\"bytes\":1000,\"evidence\":\"none\",\"layer\":2,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":1001,\"evidence\":\"none\",\"layer\":2,\"object\":\"1c6eae0ed41ed7fc1b2f00b46ea1de98f21a6d6cbf62bf6b22cf76a664a49376\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":1}}],\"op_committed\":null,\"op_delivered\":null,\"op_requested\":null,\"presented\":{\"wai-stage:layers\":0},\"stage_hash\":\"082bc00210e6d32f18f39ee723d863fd7b99f206a415ac2f44a8687793b2e068\",\"tier\":\"decode-equivalence\",\"units_by_stage\":[1,2,1],\"vocabulary\":\"urn:wai:staged-delivery:1\"},\"class\":\"stage\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0c29c7a2e9cfdb393273506b5acc9672b225192b28b9114f3b814d19c91bf4f994a74f2eecc077abbe6840c84f98f09ebbd2bd4f08d84592cca9a944918f590d\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}"
+        ]
+      },
+      "expected": {
+        "outcome": "accept",
+        "link_hash": "08f4298f5975d5594ed62f58bac25726e7ca0bccdf9d4e7b83d25656e13746a7"
+      }
+    },
+    {
+      "name": "session_stage_hash_not_exact",
+      "about": "the same stage claim, while the base's own payload was also decoded through its fallback, a capability held to a tolerance: a clip digest over a sequence some of whose frames were decoded at a weaker tier (determinism-tiers 3, rule 3)",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":4,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"b62a8425b7fe3a085a14f2f2156ea22ae3b4349884d799b5afe66dccc7d4dd7f\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":1,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"607fcbc40cc5a84709125d99f6b8fe2e3e840c218d6f8b70319c09cb92e40e4a1626d0c3d93a1e0d6efbfed6594ef0fbcdc1a0f762cebd6e98d1e4966b77d80b\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
+      "context": {
+        "stage_claims": [
+          "{\"body\":{\"bytes_envelopes\":10607,\"conflicts\":0,\"deadline_misses\":2,\"objects\":[{\"bytes\":900,\"evidence\":\"none\",\"layer\":1,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":901,\"evidence\":\"none\",\"layer\":1,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":1}},{\"bytes\":902,\"evidence\":\"none\",\"layer\":1,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":2}},{\"bytes\":903,\"evidence\":\"none\",\"layer\":1,\"object\":\"792a12990c5207ca83a86b7420815f1f2d34918553a0130a4f9570a11354f2ca\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":3}},{\"bytes\":1000,\"evidence\":\"none\",\"layer\":2,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":1001,\"evidence\":\"none\",\"layer\":2,\"object\":\"1c6eae0ed41ed7fc1b2f00b46ea1de98f21a6d6cbf62bf6b22cf76a664a49376\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":1}}],\"op_committed\":null,\"op_delivered\":null,\"op_requested\":null,\"presented\":{\"wai-stage:layers\":0},\"stage_hash\":\"082bc00210e6d32f18f39ee723d863fd7b99f206a415ac2f44a8687793b2e068\",\"tier\":\"decode-equivalence\",\"units_by_stage\":[1,2,1],\"vocabulary\":\"urn:wai:staged-delivery:1\"},\"class\":\"stage\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0c29c7a2e9cfdb393273506b5acc9672b225192b28b9114f3b814d19c91bf4f994a74f2eecc077abbe6840c84f98f09ebbd2bd4f08d84592cca9a944918f590d\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}"
+        ]
+      },
+      "expected": {
+        "outcome": "reject",
+        "step": 6,
+        "reason": "session_stage_hash_not_exact"
+      }
+    },
+    {
+      "name": "session_stage_hash_over_a_float_layer",
+      "about": "the same stage claim, while one of the set's first layers is stated as decoded on a float path",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":4,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":5,\"leaves\":1,\"module_status\":null,\"object\":\"13b04249eca2b1a9061dba8ca2846169d3f5fa3548d9e138a0cf2cb553798cf4\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":\"float\"},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"b62a8425b7fe3a085a14f2f2156ea22ae3b4349884d799b5afe66dccc7d4dd7f\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":1,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"162060aa0f64815777fe03a59eb676528c1cfb6390b530d885bb65d1d97d29f7dc0729ba5534b265e0c5cc0b5e45fcb2ff53548d106c153078ad93ba9260740f\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
+      "context": {
+        "stage_claims": [
+          "{\"body\":{\"bytes_envelopes\":10607,\"conflicts\":0,\"deadline_misses\":2,\"objects\":[{\"bytes\":900,\"evidence\":\"none\",\"layer\":1,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":901,\"evidence\":\"none\",\"layer\":1,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":1}},{\"bytes\":902,\"evidence\":\"none\",\"layer\":1,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":2}},{\"bytes\":903,\"evidence\":\"none\",\"layer\":1,\"object\":\"792a12990c5207ca83a86b7420815f1f2d34918553a0130a4f9570a11354f2ca\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":3}},{\"bytes\":1000,\"evidence\":\"none\",\"layer\":2,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":1001,\"evidence\":\"none\",\"layer\":2,\"object\":\"1c6eae0ed41ed7fc1b2f00b46ea1de98f21a6d6cbf62bf6b22cf76a664a49376\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":1}}],\"op_committed\":null,\"op_delivered\":null,\"op_requested\":null,\"presented\":{\"wai-stage:layers\":0},\"stage_hash\":\"082bc00210e6d32f18f39ee723d863fd7b99f206a415ac2f44a8687793b2e068\",\"tier\":\"decode-equivalence\",\"units_by_stage\":[1,2,1],\"vocabulary\":\"urn:wai:staged-delivery:1\"},\"class\":\"stage\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0c29c7a2e9cfdb393273506b5acc9672b225192b28b9114f3b814d19c91bf4f994a74f2eecc077abbe6840c84f98f09ebbd2bd4f08d84592cca9a944918f590d\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}"
+        ]
+      },
+      "expected": {
+        "outcome": "reject",
+        "step": 6,
+        "reason": "session_stage_hash_not_exact"
+      }
+    },
+    {
+      "name": "session_stage_hash_over_an_unpinned_base",
+      "about": "the same stage claim, while the base is stated as decoded with no pin: its NeuralIntegerExact capability holds decode-equivalence only under a pin that declares it (staged-delivery 5)",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":4,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":5,\"leaves\":1,\"module_status\":null,\"object\":\"13b04249eca2b1a9061dba8ca2846169d3f5fa3548d9e138a0cf2cb553798cf4\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"b62a8425b7fe3a085a14f2f2156ea22ae3b4349884d799b5afe66dccc7d4dd7f\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":1,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"d8f9aa9a0222d8a800e880d7fe6f6295ca39d842f727244b6d6458cbaafe4e3e7acb0d26015d64c6cd843cb2ca27aa0eaa775d172487a68ed32def05fa0edc0a\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
+      "context": {
+        "stage_claims": [
+          "{\"body\":{\"bytes_envelopes\":10607,\"conflicts\":0,\"deadline_misses\":2,\"objects\":[{\"bytes\":900,\"evidence\":\"none\",\"layer\":1,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":901,\"evidence\":\"none\",\"layer\":1,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":1}},{\"bytes\":902,\"evidence\":\"none\",\"layer\":1,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":2}},{\"bytes\":903,\"evidence\":\"none\",\"layer\":1,\"object\":\"792a12990c5207ca83a86b7420815f1f2d34918553a0130a4f9570a11354f2ca\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":3}},{\"bytes\":1000,\"evidence\":\"none\",\"layer\":2,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":1001,\"evidence\":\"none\",\"layer\":2,\"object\":\"1c6eae0ed41ed7fc1b2f00b46ea1de98f21a6d6cbf62bf6b22cf76a664a49376\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":1}}],\"op_committed\":null,\"op_delivered\":null,\"op_requested\":null,\"presented\":{\"wai-stage:layers\":0},\"stage_hash\":\"082bc00210e6d32f18f39ee723d863fd7b99f206a415ac2f44a8687793b2e068\",\"tier\":\"decode-equivalence\",\"units_by_stage\":[1,2,1],\"vocabulary\":\"urn:wai:staged-delivery:1\"},\"class\":\"stage\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0c29c7a2e9cfdb393273506b5acc9672b225192b28b9114f3b814d19c91bf4f994a74f2eecc077abbe6840c84f98f09ebbd2bd4f08d84592cca9a944918f590d\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}"
+        ]
+      },
+      "expected": {
+        "outcome": "reject",
+        "step": 6,
+        "reason": "session_stage_hash_not_exact"
+      }
     }
   ]
 }
diff --git a/wai/extensions/determinism-tiers.md b/wai/extensions/determinism-tiers.md
index a6bf9d18f..a1fb0280b 100644
--- a/wai/extensions/determinism-tiers.md
+++ b/wai/extensions/determinism-tiers.md
@@ -373,6 +373,38 @@ headroom and the rendition's digest in its rendition record (exact-tonemap §7).
   any conforming sink MUST come from a `decode-equivalence` capability. A hash
   over a reconstruction that one implementation pins (§5) is that
   implementation's, and a receipt MUST NOT present it as the capability's.
+- **No tier between the two (normative).** The tier vocabulary is closed:
+  `decode-equivalence`, `entropy-consistency` and `none` (SPEC §3.1). No tier,
+  registry class or claim states that a reconstruction stays within a bound of the
+  exact one between points at which it is made exact again. Such a bound is
+  measured on named devices and a named corpus, and a sink cannot check it from the
+  bytes it received (§4); §1 gives the published line that reaches it. Three rules
+  follow:
+  1. A decode of a `NeuralIntegerExact` capability that runs any stage in reduced
+     precision or in floating point is not a decode of that capability. A sink
+     MUST NOT report it under that capability's identifier, in a receipt or in a
+     claim. A faster decode of the same payload is registered as a capability of
+     its own, with its own class, as `wai.audio.opus.dred` and
+     `wai.audio.opus.dred.int` are two capabilities over one payload (SPEC §5,
+     *Broadcast and streaming codecs*).
+  2. A capability whose payload holds frames of more than one type, such as intra
+     frames and predicted frames, takes the class, and with it the tier, of its
+     weakest frame type. A pin's `tier` MUST NOT state more than that class gives
+     (SPEC §3.1, `tier_exceeds_class`). A frame of such a payload decoded exactly
+     does not raise the tier of the capability.
+  3. A clip digest (SPEC §7, *Digests*), a stage hash
+     ([staged-delivery](staged-delivery.md) §5), or a reconstruction binding over
+     a sequence of frames, covers only frames that were each decoded at
+     `decode-equivalence`. A sequence in which any frame was decoded at a weaker
+     tier, or by a path that does not compute the registered reconstruction, has
+     no such digest. Each frame of it that was decoded exactly keeps its own.
+
+  A receipt of a decode states the path that ran ([jwp-receipts](jwp-receipts.md)
+  §2.4.12, a decoder entry's `variant`), and the claim corpus refuses a receipt
+  that breaks each rule: a `NeuralIntegerExact` capability named with a path that
+  is not exact (rule 1); a pin's `tier` above what the path that ran holds
+  (rule 2); and a stage hash over a base decoded at a weaker tier by any of its
+  decoder entries (rule 3).
 - **Post-decode processing.** A tier claim is void if the sink alters the medium
   after decode (§2, *What a tier covers*). This applies to the emitted result,
   not to the decoder's internal conformance.
diff --git a/wai/extensions/jwp-receipts.md b/wai/extensions/jwp-receipts.md
index d966c8542..d1f5a1222 100644
--- a/wai/extensions/jwp-receipts.md
+++ b/wai/extensions/jwp-receipts.md
@@ -1035,7 +1035,7 @@ A **decoder entry**:
 | `first_unit` | integer | The title's unit that is unit 0 of the object's base: of the object itself for a base or a rendition with no `refines`, and of the base it refines for a layer. A presentation through the entry of the title's unit `u` is of the object's unit `u − first_unit`. Where a verifier holds a layer's envelope and an entry names its base, the layer's entry has that entry's `first_unit`; where it holds a `stage` claim of a set, an entry naming the set's base or a layer the claim lists has the unit `stage_sets` names that claim at (step 6). |
 | `path` | string | `primary` or `fallback`. |
 | `decoder` | string | `builtin:<crate>@<version>` — `<crate>` 1 to 64 of `a`–`z`, `0`–`9`, `_` and `-`, `<version>` 1 to 64 of ASCII letters, digits, `.`, `+` and `-`, beginning with a digit — or `module:sha256:` followed by a module's digest (§7.9), 64 lowercase hexadecimal digits. |
-| `variant` | term or `null` | A decode path within the decoder, where it has more than one, such as an exact and a floating-point path. |
+| `variant` | term or `null` | The decode path within the decoder that ran, where the capability has more than one: `exact` for the path that computes the reconstruction the capability registers (for `wai.image.jpegai`, its exact decoder), or `float` for a path that runs a stage in floating point the registration does not fix (for `wai.image.jpegai`, its float decode) ([determinism-tiers](determinism-tiers.md) §5). `null` for a capability with one path. A sink states the path that ran, as the reference dispatch reports it, not the tier its capability is registered at. |
 | `pin` | 32 B hex or `null` | The `sha256` of the parameter-set pin the sink verified (SPEC §3.1). |
 | `tier` | string or `null` | That pin's declared tier as SPEC §3.1 reads it: `decode-equivalence`, `entropy-consistency` or `none`. `null` exactly when `pin` is. |
 | `module_status` | string or `null` | For a `module:` decoder, the anchor its grant was confirmed under (§7.9): `pinned` or `fetched`. `null` for a `builtin:` decoder. |
@@ -1088,6 +1088,13 @@ Rules, checked at §3.1 step 4 in this order:
     (`energy_zero_figure`), and its class is admitted as §2.4.5's rule 2 admits one
     (`energy_unknown_acquisition`, `energy_legacy_acquisition`,
     `energy_acquisition_evidence`).
+13. A decoder entry whose `capability` has the replicate class `neural-int-exact`
+    (SPEC §7) **MUST** have `variant` `null` or `exact` (`session_variant_not_exact`).
+    A decode of such a capability by any other path is not a decode of it
+    ([determinism-tiers](determinism-tiers.md) §3, *No tier between the two*, rule 1).
+14. A decoder entry whose `variant` is `float` **MUST NOT** have `tier`
+    `decode-equivalence` (`session_tier_exceeds_variant`). Frames a float path decoded
+    are not decode-equivalent, whatever the pin declares (determinism-tiers §3, rule 2).
 
 **Against the outputs** (§3.1 step 6). The context is the list of presentations, each
 `{ "unit", "layers", "decoder", "out_digest" }`: a unit, the layers it was presented
@@ -1258,7 +1265,16 @@ epochs, and bases that cover the same units of the title each have their own. In
    one lists, has as its `first_unit` the `first_unit` of an entry of `stage_sets` naming
    a claim of that object (`session_first_unit_mismatch`): a set's base and its layers
    are at the units the session names the set at;
-5. where the verifier holds the outputs as well, for each claim and each entry of
+5. each of them that states a `stage_hash`, a clip digest over its set's units
+   ([staged-delivery](staged-delivery.md) §5), has every decoder entry whose `object` is
+   its `subject` at `decode-equivalence` by that section's base step (a capability of
+   the replicate class `standard-defined`, or of `neural-int-exact` with `tier`
+   `decode-equivalence`), and no decoder entry whose `object` is its `subject` or a layer
+   object it lists has a `variant` other than `null` or `exact`
+   (`session_stage_hash_not_exact`). A sequence any frame of which was decoded at a
+   weaker tier, or by a path that is not exact, has no clip digest
+   ([determinism-tiers](determinism-tiers.md) §3, *No tier between the two*, rule 3);
+6. where the verifier holds the outputs as well, for each claim and each entry of
    `stage_sets` naming it, with `f` that entry's `first_unit` and `n` the units the claim
    counts (`session_stage_disagrees`):
    - every unit from `f` to `f + n − 1` is presented;
@@ -1267,7 +1283,7 @@ epochs, and bases that cover the same units of the title each have their own. In
      unit presented on time at a stage was presented at every lower stage before it, and
      late presentations are leaves of the tree as well, so the tree can hold more than a
      `stage` claim counts and never less;
-6. where it holds the outputs, from above: every unit of an entry's units presented at
+7. where it holds the outputs, from above: every unit of an entry's units presented at
    more than `s` layers lies in the span of a layer-`s` object that a claim named by an
    entry holding the unit lists as `applied` or `partial` (`session_stage_disagrees`). A
    unit reaches a stage only through a layer the sink applied: a `late` layer missed the
@@ -2285,6 +2301,8 @@ and never folds it in: confirmation changes the outcome of no step here.
 | 4 | `session_stage_sets_invalid` | `stage_sets` out of order, or an entry twice |
 | 4 | `session_counted_invalid` | `counted_discarded` above `counted_primary` and `counted_fallback` together |
 | 4 | `session_classical_invalid` | `classical_source` present without `classical_declared` or absent with it, or not `wai2-table` |
+| 4 | `session_variant_not_exact` | a decoder entry names a capability of the replicate class `neural-int-exact` with a `variant` other than `null` or `exact` |
+| 4 | `session_tier_exceeds_variant` | a decoder entry's `variant` is `float` and its `tier` is `decode-equivalence` |
 | 4 | `consumption_objects_unordered` | a `consumption` claim's `objects` are not in strictly ascending order ([ai-usage](ai-usage.md) §6.2) |
 | 4 | `consumption_subject_mismatch` | its `subject` is not the first of its `objects` |
 | 4 | `consumption_usage_unordered` | its `usage` entries are not in strictly ascending order of `vocabulary` |
@@ -2352,6 +2370,7 @@ and never folds it in: confirmation changes the outcome of no step here.
 | 6 | `schedule_content_mismatch` | an entry's `content_hash` is not that of the object receipt the verifier holds at its position |
 | 6 | `schedule_length_mismatch` | an entry's transfer length is not the length of the object's envelope the verifier holds |
 | 6 | `session_stage_disagrees` | a `stage` claim the session names is about a base, or lists an arrived layer object, the session did not receive; counts a unit the session never presented, or more units at a stage than the session presented at it or above; or no claim whose set holds a unit lists an applied layer that reaches a stage the session presented it at |
+| 6 | `session_stage_hash_not_exact` | a `stage` claim the session names states a `stage_hash`, and a decoder entry names its base at a step weaker than `decode-equivalence`, or names its base or a layer it lists with a `variant` other than `null` or `exact` |
 
 ### 3.2. Recovery through a soft binding, and the ancestor rule
 
@@ -2958,7 +2977,10 @@ fewer than it counts; a base forgotten and decoded again (`staged-conformance/`'
 every reason of §2.4.12's rules and
 checks, the caps of the switches and missed presentations written at and past them; a
 metered session's figure of zero (`energy_zero_figure`) and every acquisition-class
-reason; and step 2's enumerations and order. `claim_verify.py` gives every outcome.
+reason; and step 2's enumerations and order. The decode path an entry states has a
+refusal for each of rules 13 and 14 and for a stage hash over a base or layer decoded at
+a weaker tier or by a path that is not exact, beside accepted entries for each of
+`wai.image.jpegai`'s two paths. `claim_verify.py` gives every outcome.
 `differential.py` generates sessions around every rule — units, leaves and paths off by
 one, switches and missed presentations at and past their caps, a `null` deadline with
 misses, presences that disagree, decoder indices out of range, leaves moved between
diff --git a/wai/wai-rs/src/claim_receipt.rs b/wai/wai-rs/src/claim_receipt.rs
index fc4ac29b4..528108552 100644
--- a/wai/wai-rs/src/claim_receipt.rs
+++ b/wai/wai-rs/src/claim_receipt.rs
@@ -363,6 +363,12 @@ pub enum Reason {
     SessionDecoderMalformed,
     /// A decoder entry whose tier exceeds its capability's class tier.
     SessionTierExceedsClass,
+    /// A decoder entry naming a `NeuralIntegerExact` capability with a decode path other
+    /// than its integer one: a `variant` other than `null` or `exact`.
+    SessionVariantNotExact,
+    /// A decoder entry whose `tier` exceeds what the path that ran holds: a `float` path
+    /// stating `decode-equivalence`.
+    SessionTierExceedsVariant,
     /// A switch between equal paths or for an unlisted reason, switches out of order,
     /// more than 1024 written or more than their total.
     SessionSwitchInvalid,
@@ -423,6 +429,10 @@ pub enum Reason {
     /// A `stage` claim the session names counts units the session's outputs do not hold
     /// at its stages.
     SessionStageDisagrees,
+    /// A `stage` claim the session names states a stage hash, a clip digest, over a base
+    /// a decoder entry decoded at a weaker step than `decode-equivalence`, or over a base
+    /// or layer an entry decoded by a path that is not exact.
+    SessionStageHashNotExact,
     /// A `consumption` claim's `objects` are not in strictly ascending order.
     ConsumptionObjectsUnordered,
     /// A `consumption` claim's `subject` is not the first of its `objects`.
@@ -578,6 +588,8 @@ impl Reason {
             Reason::SessionHostMalformed => "session_host_malformed",
             Reason::SessionDecoderMalformed => "session_decoder_malformed",
             Reason::SessionTierExceedsClass => "session_tier_exceeds_class",
+            Reason::SessionVariantNotExact => "session_variant_not_exact",
+            Reason::SessionTierExceedsVariant => "session_tier_exceeds_variant",
             Reason::SessionSwitchInvalid => "session_switch_invalid",
             Reason::SessionMissedInvalid => "session_missed_invalid",
             Reason::SessionStageSetsInvalid => "session_stage_sets_invalid",
@@ -599,6 +611,7 @@ impl Reason {
             Reason::SessionUnitOutsideObject => "session_unit_outside_object",
             Reason::SessionClassicalMismatch => "session_classical_mismatch",
             Reason::SessionStageDisagrees => "session_stage_disagrees",
+            Reason::SessionStageHashNotExact => "session_stage_hash_not_exact",
             Reason::ConsumptionObjectsUnordered => "consumption_objects_unordered",
             Reason::ConsumptionSubjectMismatch => "consumption_subject_mismatch",
             Reason::ConsumptionUsageUnordered => "consumption_usage_unordered",
@@ -4247,7 +4260,7 @@ pub(crate) fn rfc3339(s: &str) -> bool {
 /// fails instead of passing on nothing. (A function rather than a public constant so
 /// cbindgen does not export it into the C header.)
 pub fn corpus_vector_count() -> usize {
-    636
+    647
 }
 
 /// The conformance corpus (`claim-conformance/vectors.json`), regenerated. Ed25519
@@ -5489,6 +5502,8 @@ mod corpus {
         v.extend(span_vectors());
         // §2.4.8's fabric record, appended the same way.
         v.extend(fabric_vectors());
+        // §2.4.12's decode path (rules 13 and 14, and a stage hash), appended the same way.
+        v.extend(session::decode_path_vectors());
         v
     }
 
@@ -7500,6 +7515,19 @@ mod tests {
         assert_eq!(hex(digest.as_bytes()), "74f802d701daa0620b1577cb100a80951fedb797f9d47057efca61f21d40b8a6");
     }
 
+    /// The vectors the corpus held before a session's decode path was given its rules
+    /// (all 636), pinned as the tests above pin theirs. The digest was computed from the
+    /// committed corpus before those rules were added and must never be edited to make the
+    /// test pass.
+    #[test]
+    fn the_vectors_before_the_decode_path_rules_are_unchanged() {
+        let doc: Value = crate::oer::load(conformance_vectors().as_bytes()).unwrap();
+        let vs = doc["vectors"].as_array().unwrap();
+        let first = Value::Array(vs[..636].to_vec());
+        let digest = blake3::hash(serde_json::to_string(&first).unwrap().as_bytes());
+        assert_eq!(hex(digest.as_bytes()), "885e3ab74f416f9f8a0ef560dca930b31e430aa2b06a0d3bb126ab95fed841f0");
+    }
+
     /// A session claim's grammars (§2.4.12): `media`, a decoder identifier, and the
     /// rounding of its worker count.
     #[test]
diff --git a/wai/wai-rs/src/claim_receipt/corpus/session.rs b/wai/wai-rs/src/claim_receipt/corpus/session.rs
index 1026e6edd..83edf4bac 100644
--- a/wai/wai-rs/src/claim_receipt/corpus/session.rs
+++ b/wai/wai-rs/src/claim_receipt/corpus/session.rs
@@ -2720,3 +2720,175 @@ fn sealed_or_bad(s: SessionClaim) -> String {
         Err(e) => panic!("a step-6 vector's session must pass steps 1-4: {e}"),
     }
 }
+
+// ───────── the decode path that ran (§2.4.12 rules 13 and 14, and a stage hash) ─────────
+
+/// The object a one-picture session decoded from: not held as an envelope by any vector.
+fn picture() -> [u8; 32] {
+    object_content_hash(b"WAI1 claim-conformance: a picture")
+}
+
+/// A session of one picture through one decoder entry: `capability` decoded by `decoder`
+/// on the path `variant` names, under a pin of `tier` or none.
+fn one_picture(capability: &str, decoder: &str, module_status: Option<&str>, variant: Option<&str>, tier: Option<&str>) -> SessionClaim {
+    let ps = [pres(0, 1, 0)];
+    SessionClaim {
+        media: "image".into(),
+        decoders: vec![DecoderEntry {
+            capability: capability.into(),
+            rendition: None,
+            object: picture(),
+            first_unit: 0,
+            path: SessionPath::Primary,
+            decoder: decoder.into(),
+            variant: variant.map(Into::into),
+            pin: tier.map(|_| pin()),
+            tier: tier.map(Into::into),
+            module_status: module_status.map(Into::into),
+            leaves: 1,
+        }],
+        units: 1,
+        primary_units: 1,
+        fallback_units: 0,
+        switches: vec![],
+        switches_total: 0,
+        deadline_us: None,
+        missed: 0,
+        missed_units: vec![],
+        stage_sets: vec![],
+        counted_primary: 0,
+        counted_fallback: 0,
+        counted_discarded: 0,
+        uncounted: 1,
+        transferred: 367,
+        classical_declared: None,
+        classical_source: None,
+        ..over(model(), &ps)
+    }
+}
+
+/// The model session's stage claim, with its tier stated as `tier`: at
+/// `decode-equivalence` it carries a stage hash, the clip digest of its units.
+fn model_stage_at(tier: &str) -> ClaimReceipt {
+    let ClaimBody::Stage(mut st) = model_stage_ok().body else { unreachable!("the model stage claim is a stage claim") };
+    st.stage_hash = (tier == "decode-equivalence").then(|| object_content_hash(b"claim-conformance: a clip digest"));
+    st.tier = tier.into();
+    seal(&viewer(), SINK_ID, SubjectType::Object, entry(), 0, ClaimBody::Stage(st), None)
+}
+
+/// The vectors of the decode path a decoder entry states (`variant`): rules 13 and 14 at
+/// step 4, and a stage hash against the entries that decoded its set at step 6. Appended
+/// after every earlier vector, which a test pins by digest.
+pub(super) fn decode_path_vectors() -> Vec<V> {
+    const BUILTIN: &str = "builtin:wai@0.4.0";
+    let module = format!("module:sha256:{}", "5e".repeat(32));
+    let picture_ok = |s: SessionClaim| sealed_about(picture(), s).to_json();
+    let picture_bad = |s: SessionClaim| seal_bad(&viewer(), SINK_ID, SubjectType::Object, picture(), 0, ClaimBody::ReconstructSession(s));
+    let mut v = vec![
+        claim(
+            "session_variant_float_accept",
+            "a picture decoded by wai.image.jpegai's float path, stated as variant float: the path that ran, beside the capability, which is registered at entropy-consistency",
+            picture_ok(one_picture("wai.image.jpegai", BUILTIN, None, Some("float"), None)),
+            Ok(()),
+        ),
+        claim(
+            "session_variant_exact_accept",
+            "the same picture decoded by wai.image.jpegai's exact decoder, stated as variant exact, under a pin of entropy-consistency",
+            picture_ok(one_picture("wai.image.jpegai", BUILTIN, None, Some("exact"), Some("entropy-consistency"))),
+            Ok(()),
+        ),
+        claim(
+            "session_variant_exact_integer_accept",
+            "a NeuralIntegerExact capability whose entry states its integer path as variant exact, under a pin of decode-equivalence",
+            picture_ok(one_picture("wai.neural.int_hyper", BUILTIN, None, Some("exact"), Some("decode-equivalence"))),
+            Ok(()),
+        ),
+        reject(
+            "session_variant_not_exact",
+            "a NeuralIntegerExact capability named with variant float: a decode of it in floating point is not a decode of it (determinism-tiers 3, rule 1)",
+            picture_bad(one_picture("wai.neural.int_hyper", BUILTIN, None, Some("float"), Some("decode-equivalence"))),
+            4,
+            Reason::SessionVariantNotExact,
+        ),
+        reject(
+            "session_variant_not_exact_unpinned",
+            "the same with no pin, and a path named by a term this revision does not register: only null or exact names the integer path",
+            picture_bad(one_picture("wai.neural.int_hyper", BUILTIN, None, Some("fp16"), None)),
+            4,
+            Reason::SessionVariantNotExact,
+        ),
+        reject(
+            "session_tier_exceeds_variant",
+            "a capability the registry does not classify, decoded by a module on its float path and stated under a pin of decode-equivalence: a float path's frames are not decode-equivalent, whatever the pin declares (determinism-tiers 3, rule 2)",
+            picture_bad(one_picture("org.example.picture", &module, Some("pinned"), Some("float"), Some("decode-equivalence"))),
+            4,
+            Reason::SessionTierExceedsVariant,
+        ),
+        reject(
+            "session_tier_exceeds_class_before_variant",
+            "wai.image.jpegai's float path under a pin of decode-equivalence: the class rule, rule 6, refuses it first",
+            picture_bad(one_picture("wai.image.jpegai", BUILTIN, None, Some("float"), Some("decode-equivalence"))),
+            4,
+            Reason::SessionTierExceedsClass,
+        ),
+    ];
+
+    // ── step 6: a stage hash, the clip digest of a set's units ──
+    let de_stage = model_stage_at("decode-equivalence");
+    let named = |stage: &ClaimReceipt| SessionClaim { stage_sets: vec![StageSetRef { first_unit: 0, claim: stage.link_hash() }], ..model() };
+    let held = |v: V, stage: &ClaimReceipt| with(v, "stage_claims", json!([stage.to_json()]));
+    v.push(held(
+        claim(
+            "session_stage_hash_exact_accept",
+            "the model session's stage set stated at decode-equivalence, with its stage hash: its base decoded only through a pinned integer capability, its layers through a residual refinement, every path exact",
+            sealed(named(&de_stage)).to_json(),
+            Ok(()),
+        ),
+        &de_stage,
+    ));
+    v.push(held(
+        reject(
+            "session_stage_hash_not_exact",
+            "the same stage claim, while the base's own payload was also decoded through its fallback, a capability held to a tolerance: a clip digest over a sequence some of whose frames were decoded at a weaker tier (determinism-tiers 3, rule 3)",
+            sealed_or_bad({
+                let mut s = named(&de_stage);
+                s.decoders[2].object = entry();
+                s.decoders[2].first_unit = 0;
+                s
+            }),
+            6,
+            Reason::SessionStageHashNotExact,
+        ),
+        &de_stage,
+    ));
+    v.push(held(
+        reject(
+            "session_stage_hash_over_a_float_layer",
+            "the same stage claim, while one of the set's first layers is stated as decoded on a float path",
+            sealed_or_bad({
+                let mut s = named(&de_stage);
+                s.decoders[3].variant = Some("float".into());
+                s
+            }),
+            6,
+            Reason::SessionStageHashNotExact,
+        ),
+        &de_stage,
+    ));
+    v.push(held(
+        reject(
+            "session_stage_hash_over_an_unpinned_base",
+            "the same stage claim, while the base is stated as decoded with no pin: its NeuralIntegerExact capability holds decode-equivalence only under a pin that declares it (staged-delivery 5)",
+            sealed_or_bad({
+                let mut s = named(&de_stage);
+                s.decoders[0].pin = None;
+                s.decoders[0].tier = None;
+                s
+            }),
+            6,
+            Reason::SessionStageHashNotExact,
+        ),
+        &de_stage,
+    ));
+    v
+}
diff --git a/wai/wai-rs/src/claim_receipt/session.rs b/wai/wai-rs/src/claim_receipt/session.rs
index ff021e3bc..1e8ba7f65 100644
--- a/wai/wai-rs/src/claim_receipt/session.rs
+++ b/wai/wai-rs/src/claim_receipt/session.rs
@@ -34,6 +34,13 @@ pub const SESSION_MODULE_STATUSES: &[&str] = &["pinned", "fetched"];
 pub const BUILTIN_DECODER: &str = "builtin:";
 /// A decoder entry's `module:` prefix: a module's SHA-256 follows.
 pub const MODULE_DECODER: &str = "module:sha256:";
+/// The `variant` of a decoder entry whose path computes the reconstruction its capability
+/// registers (determinism-tiers §5): `wai.image.jpegai`'s exact decoder, or a
+/// `NeuralIntegerExact` capability's integer path, which may also be stated as `null`.
+pub const SESSION_VARIANT_EXACT: &str = "exact";
+/// The `variant` of a decoder entry whose path runs a stage in floating point that the
+/// capability's registration does not fix: `wai.image.jpegai`'s float decode.
+pub const SESSION_VARIANT_FLOAT: &str = "float";
 
 /// The most switches a claim writes. Private, so the C header does not gain it.
 const MAX_SWITCHES: usize = 1024;
@@ -134,6 +141,8 @@ pub struct DecoderEntry {
     pub path: SessionPath,
     /// `builtin:<crate>@<version>` or `module:sha256:<hex>`.
     pub decoder: String,
+    /// The decode path that ran: [`SESSION_VARIANT_EXACT`] or [`SESSION_VARIANT_FLOAT`]
+    /// (`neural::DecodePath::variant`), or `None` for a capability with one path.
     pub variant: Option<String>,
     /// The SHA-256 of the pin the sink verified.
     pub pin: Option<[u8; 32]>,
@@ -719,6 +728,19 @@ impl SessionClaim {
             }
             e.acquisition_class()?;
         }
+        // Rule 13: a NeuralIntegerExact capability is decoded by its integer path only; a
+        // decode by any other path is not a decode of it (determinism-tiers §3, *No tier
+        // between the two*, rule 1).
+        let integer = |cap: &str| crate::codecs::capability_replicate_class(cap) == Some(crate::codecs::ReplicateClass::NeuralIntegerExact);
+        if self.decoders.iter().any(|d| d.variant.as_deref().is_some_and(|v| v != SESSION_VARIANT_EXACT) && integer(&d.capability)) {
+            return Err(Reason::SessionVariantNotExact);
+        }
+        // Rule 14: no pin's tier above what the path that ran holds; frames a float path
+        // decoded are not decode-equivalent, whatever the pin declares (rule 2 there).
+        let de = Some(crate::codecs::DeclaredTier::DecodeEquivalence);
+        if self.decoders.iter().any(|d| d.variant.as_deref() == Some(SESSION_VARIANT_FLOAT) && d.tier.as_deref().and_then(crate::codecs::DeclaredTier::from_token) == de) {
+            return Err(Reason::SessionTierExceedsVariant);
+        }
         Ok(())
     }
 
@@ -1330,7 +1352,10 @@ impl ClaimReceipt {
     /// layer object that arrived are among them (`session_stage_disagrees`). Each decoder
     /// entry whose object is a claim's base, or a layer object a claim lists, has a
     /// `first_unit` at which `stage_sets` names a claim of that object
-    /// (`session_first_unit_mismatch`). Then, with the
+    /// (`session_first_unit_mismatch`). A claim that states a stage hash has every entry
+    /// naming its base at `decode-equivalence` by staged-delivery §5's base step, and no
+    /// entry naming its base or a layer it lists with a path that is not exact
+    /// (`session_stage_hash_not_exact`). Then, with the
     /// presentations (`session_stage_disagrees`): for each claim and each entry naming it,
     /// every unit of its set is presented, and for each stage `s` from 1 up the units of
     /// the set presented at more than `s` layers are no fewer than it counts at `s` or
@@ -1376,6 +1401,30 @@ impl ClaimReceipt {
         if s.decoders.iter().any(|d| at.get(&d.object).is_some_and(|fs| !fs.contains(&d.first_unit))) {
             return Err(Rejection::at(6, Reason::SessionFirstUnitMismatch));
         }
+        // A stage hash is a clip digest over the set's units (staged-delivery §5): every
+        // entry naming the set's base decodes it at decode-equivalence by the base step of
+        // staged-delivery §5, and no entry naming the base or a layer the claim lists ran a
+        // path that is not exact (determinism-tiers §3, *No tier between the two*, rule 3).
+        let base_exact = |d: &DecoderEntry| match crate::codecs::capability_replicate_class(&d.capability) {
+            Some(crate::codecs::ReplicateClass::StandardDefined) => true,
+            Some(crate::codecs::ReplicateClass::NeuralIntegerExact) => d.tier.as_deref() == Some("decode-equivalence"),
+            _ => false,
+        };
+        let path_exact = |d: &DecoderEntry| d.variant.as_deref().is_none_or(|v| v == SESSION_VARIANT_EXACT);
+        for (c, _) in &claims {
+            let ClaimBody::Stage(st) = &c.body else { continue };
+            if st.stage_hash.is_none() {
+                continue;
+            }
+            let mixed = s.decoders.iter().any(|d| {
+                let base = d.object == c.subject;
+                let layer = st.objects.iter().any(|o| o.object == d.object);
+                (base && !base_exact(d)) || ((base || layer) && !path_exact(d))
+            });
+            if mixed {
+                return Err(Rejection::at(6, Reason::SessionStageHashNotExact));
+            }
+        }
         let Some(ps) = outputs else { return Ok(()) };
         let finals = SessionClaim::finals(ps);
         // (set-local unit, largest layers) of every unit presented in [f, f + n).
-- 
2.54.0 (Apple Git-157)


From 15fecef33f2e1e9ac863cdc164207562e17d43bb Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:42:25 -0500
Subject: [PATCH 14/29] neural: a decode reports the capability it decoded
 through and the path that ran
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

wai.image.jpegai decodes through its exact decoder or through a float decode,
and both returned the same result, so neither a receipt nor a binding check
could tell which had run. decode_envelope, decode_envelope_strict,
decode_multi, decode_multi_strict and decode_payload now return a Dispatched:
the output, the capability the payload decoded through (its own or the
fallback) and the path, DecodePath::Sole, Exact or Float. DecodePath::variant
gives the term a session claim's decoder entry states.

check_reconstruction_binding reads the path: over a decode-equivalence
capability or the exact decoder it reports whether the hash holds, and over
the float decode it reports the binding as not checked, never as failed, as
reconstruction-binding §2 requires. The notes in determinism-tiers §5 and
reconstruction-binding §2 that the dispatch did not report the path are gone.

tests/jpegai_decode_path.rs runs the float path and requires the result, the
session claim and the binding check to say so, and the exact decoder likewise.
Its JPEG-AI cases need inputs outside the repository and skip without them;
the case with one path runs in the neural gates.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-neural-gates.yml   |  17 ++
 wai/CHANGELOG.md                         |  10 ++
 wai/extensions/determinism-tiers.md      |  15 +-
 wai/extensions/reconstruction-binding.md |   8 +-
 wai/wai-rs/Cargo.toml                    |   5 +
 wai/wai-rs/src/ffi_neural.rs             |   7 +-
 wai/wai-rs/src/neural/mod.rs             | 209 +++++++++++++++++------
 wai/wai-rs/tests/jpegai_decode_path.rs   | 173 +++++++++++++++++++
 wai/wai-rs/tests/neural_cost_limit.rs    |   2 +-
 wai/wai-rs/tests/neural_native_decode.rs |  38 +++--
 10 files changed, 411 insertions(+), 73 deletions(-)
 create mode 100644 wai/wai-rs/tests/jpegai_decode_path.rs

diff --git a/.github/workflows/wai-neural-gates.yml b/.github/workflows/wai-neural-gates.yml
index 9a3761ff9..56742d8d7 100644
--- a/.github/workflows/wai-neural-gates.yml
+++ b/.github/workflows/wai-neural-gates.yml
@@ -66,6 +66,10 @@ on:
       - 'wai/wai-rs/src/ffi_neural.rs'
       - 'wai/wai-rs/tests/ffi_neural_pin.rs'
       - 'wai/wai-rs/tests/neural_cost_limit.rs'
+      - 'wai/wai-rs/tests/jpegai_decode_path.rs'
+      - 'wai/wai-rs/src/session_recorder.rs'
+      - 'wai/wai-rs/src/reconstruction_binding.rs'
+      - 'wai/wai-rs/src/claim_receipt/**'
       - 'wai/wai-rs/src/int_synth.rs'
       # The integer operator set the integer decoders load their sets through.
       - 'wai/wai-rs/src/int_ops.rs'
@@ -187,6 +191,19 @@ jobs:
           cargo test --no-default-features --features neural,ffi,video_int,audio_int,splat_int --test neural_cost_limit \
             2>&1 | tee "$RUNNER_TEMP/neural-cost.log"
           grep -q "test result: ok. 3 passed; 0 failed" "$RUNNER_TEMP/neural-cost.log"
+      # The dispatch reports the capability it decoded through and the path that ran
+      # (extensions/determinism-tiers.md §5): a capability with one path says so, through
+      # its own capability and through its fallback. The JPEG-AI float and exact paths need
+      # inputs no job here provides (the local asset directory and the jpegai/v1 bundle),
+      # so in this step they skip loudly; the count still requires all three to be present.
+      - name: The neural dispatch reports the decode path that ran
+        working-directory: wai/wai-rs
+        shell: bash
+        run: |
+          set -euo pipefail
+          cargo test --no-default-features --features neural,receipt_claims,video_int --test jpegai_decode_path \
+            2>&1 | tee "$RUNNER_TEMP/neural-path.log"
+          grep -q "test result: ok. 3 passed; 0 failed" "$RUNNER_TEMP/neural-path.log"
       - name: A key log refuses a revoked pinned set at the neural sink (SPEC §3.1 check 5)
         working-directory: wai/wai-rs
         shell: bash
diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index 68877e9e2..3bffbc49a 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -8,6 +8,16 @@ wire format may change and such changes are recorded here.
 
 ### A decode reports the path that ran, and no tier lies between the two
 
+- **The reference dispatch reports the decode path** ([extensions/determinism-tiers.md](extensions/determinism-tiers.md)
+  §5). `neural::decode_envelope`, `decode_envelope_strict`, `decode_multi`, `decode_multi_strict` and `decode_payload`
+  return a `Dispatched`: the output, the capability the payload decoded through (its own or the fallback) and the path
+  of it that ran, `DecodePath::Sole`, `Exact` or `Float`. `wai.image.jpegai`'s float decode through ONNX Runtime and its
+  exact decoder now say which ran. `neural::check_reconstruction_binding` reads the path
+  ([extensions/reconstruction-binding.md](extensions/reconstruction-binding.md) §2): over the exact decoder it reports
+  whether the hash holds, and over the float decode it reports the binding as not checked, never as failed. The notes
+  that the dispatch did not report the path are gone from both texts. `tests/jpegai_decode_path.rs` runs the float path
+  and requires the result, the session claim and the binding check to say so; its JPEG-AI cases need inputs outside the
+  repository and skip without them.
 - **No tier between the two** (determinism-tiers §3, normative). The tier vocabulary stays closed. A decode of a
   `NeuralIntegerExact` capability in reduced precision or floating point is not a decode of it; a capability whose
   payload holds several frame types takes the class of its weakest; and a clip digest, a stage hash or a reconstruction
diff --git a/wai/extensions/determinism-tiers.md b/wai/extensions/determinism-tiers.md
index a1fb0280b..bdf42d607 100644
--- a/wai/extensions/determinism-tiers.md
+++ b/wai/extensions/determinism-tiers.md
@@ -368,7 +368,8 @@ headroom and the rendition's digest in its rendition record (exact-tonemap §7).
   declaration, not the arithmetic. Neither step inspects the decode path a sink
   actually runs; for `wai.neural.int_synth` and `wai.neural.int_mlicpp`, the
   `NeuralIntegerExact` capabilities the reference neural dispatch decodes, there is
-  only the integer path.
+  only the integer path. The decode then reports the path that ran
+  (`neural::Dispatched::path`), which a sink states in a receipt (§5).
 - **Receipts.** A receipt that carries a decoded-medium hash as reproducible by
   any conforming sink MUST come from a `decode-equivalence` capability. A hash
   over a reconstruction that one implementation pins (§5) is that
@@ -520,8 +521,16 @@ this configuration rather than decoding it to the wrong image. Within it, the ex
 decoder runs when the sink supplies its integer weight bundle and the thread's
 floating-point environment is the IEEE-754 default. Otherwise the reference sink
 decodes the same capability ID through the float ONNX Runtime path, whose
-reconstruction is not pinned. Both paths return the same result type, and the
-reference dispatch does not report which one ran.
+reconstruction is not pinned. The reference dispatch reports which path ran:
+`neural::decode_envelope` returns, with the image, the capability it decoded
+through and the path, `DecodePath::Exact` or `DecodePath::Float`. A sink states that
+path in a `reconstruct-session` claim as its decoder entry's `variant`, `exact` or
+`float` ([jwp-receipts](jwp-receipts.md) §2.4.12), and the reference check of a
+reconstruction binding reads it (`neural::check_reconstruction_binding`,
+[reconstruction-binding](reconstruction-binding.md) §2).
+`wai-rs/tests/jpegai_decode_path.rs` runs the float path and requires the result,
+the claim and the binding check to say so; it runs only where the float path's
+model bundle is supplied.
 
 On both paths the stages the capability's claim covers, the scale path and the two
 entropy decodes, run through the integer operator set, by the lowering `jpegai-int/1`
diff --git a/wai/extensions/reconstruction-binding.md b/wai/extensions/reconstruction-binding.md
index c48de0947..8757b27a8 100644
--- a/wai/extensions/reconstruction-binding.md
+++ b/wai/extensions/reconstruction-binding.md
@@ -103,9 +103,11 @@ Conformance criterion:
   and then MUST obtain the recorded `hash`. A verifier that decodes any other way
   has not checked the binding: a mismatch it finds is not evidence that the asset
   changed, and it MUST NOT report the binding as failed on that basis. The
-  reference dispatch (`decode_envelope`) does not report which path ran, so a
-  verifier calls the exact decoder directly (`jpegai_decode::decode`), which
-  refuses rather than falls back.
+  reference dispatch (`neural::decode_envelope`) reports the path that ran
+  ([determinism-tiers](determinism-tiers.md) §5), and the reference check
+  (`neural::check_reconstruction_binding`) reads it: over the exact decoder it
+  reports whether the hash holds, and over the float fallback it reports the
+  binding as not checked, with the path, never as failed.
 
 ## 3. C2PA / JPEG Trust carriage
 
diff --git a/wai/wai-rs/Cargo.toml b/wai/wai-rs/Cargo.toml
index e89b45c75..5f750822b 100644
--- a/wai/wai-rs/Cargo.toml
+++ b/wai/wai-rs/Cargo.toml
@@ -2160,6 +2160,11 @@ name = "neural_cost_limit"
 path = "tests/neural_cost_limit.rs"
 required-features = ["ffi", "neural", "video_int"]
 
+[[test]]
+name = "jpegai_decode_path"
+path = "tests/jpegai_decode_path.rs"
+required-features = ["neural", "receipt_claims", "video_int"]
+
 [[test]]
 name = "semantic_path"
 path = "tests/semantic_path.rs"
diff --git a/wai/wai-rs/src/ffi_neural.rs b/wai/wai-rs/src/ffi_neural.rs
index 2a181733b..47ba5e828 100644
--- a/wai/wai-rs/src/ffi_neural.rs
+++ b/wai/wai-rs/src/ffi_neural.rs
@@ -93,7 +93,7 @@ fn decode_one(envelope: *const u8, envelope_len: usize, model_path: *const c_cha
         return Err(no_kind(&[env.capability()]));
     }
     let reg = ModelRegistry::new().register(env.capability().to_string(), Path::new(path));
-    decode_envelope(&env, &reg).map_err(decode_status)
+    decode_envelope(&env, &reg).map(|d| d.output).map_err(decode_status)
 }
 
 fn wrong_media(want: &str) -> (c_int, String) {
@@ -289,12 +289,13 @@ pub unsafe extern "C" fn wai_neural_decode_ex(
             let env = Wai::from_bytes(bytes).map_err(|e| (WAI_ERR_CONTAINER, format!("envelope parse: {e}")))?;
             let d = if strict { decode_envelope_strict(&env, &reg) } else { decode_envelope(&env, &reg) };
             let named = [Some(env.capability()), env.fallback()].into_iter().flatten().collect();
-            (0, d.map_err(|e| explain(named, e))?)
+            (0, d.map_err(|e| explain(named, e))?.output)
         } else {
             let m = WaiMulti::from_bytes(bytes).map_err(|e| (WAI_ERR_CONTAINER, format!("envelope parse: {e}")))?;
             let r = if strict { decode_multi_strict(&m, &reg) } else { decode_multi(&m, &reg) };
             let named = m.manifest.renditions.iter().map(|r| r.capability.as_str()).collect();
-            r.map_err(|e| explain(named, e))?
+            let (i, d) = r.map_err(|e| explain(named, e))?;
+            (i, d.output)
         };
         let mut o = WaiDecoded { index: index as u32, ..WaiDecoded::empty() };
         match decoded {
diff --git a/wai/wai-rs/src/neural/mod.rs b/wai/wai-rs/src/neural/mod.rs
index 0e4cdaf9d..1ae7c97b5 100644
--- a/wai/wai-rs/src/neural/mod.rs
+++ b/wai/wai-rs/src/neural/mod.rs
@@ -523,6 +523,54 @@ pub enum Decoded {
     Int(crate::int_payload::IntDecoded),
 }
 
+/// Which of a capability's decode paths gave a decode's output
+/// (extensions/determinism-tiers.md §5). A sink states it in a `reconstruct-session`
+/// claim as a decoder entry's `variant` ([`DecodePath::variant`],
+/// extensions/jwp-receipts.md §2.4.12): the path that ran, which a receipt reports
+/// beside the tier the capability is registered at.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
+#[non_exhaustive]
+pub enum DecodePath {
+    /// The capability's one decode path: every capability this dispatch decodes but
+    /// `wai.image.jpegai`.
+    Sole,
+    /// `wai.image.jpegai`'s exact decoder (`jpegai_decode`): the reconstruction a
+    /// decoded-pixel hash or a reconstruction binding over the capability is over.
+    Exact,
+    /// `wai.image.jpegai`'s float decode through ONNX Runtime, taken by a sink without
+    /// the integer weight bundle, or on a thread or target that cannot run the exact
+    /// decoder's float stages. Its symbols are the exact decoder's; its pixels are not
+    /// pinned, so no decoded-pixel hash or binding is checked over them.
+    Float,
+}
+
+impl DecodePath {
+    /// The `variant` a `reconstruct-session` decoder entry states for this path: `None`
+    /// for a capability with one path, `exact` for the exact decoder, `float` for the
+    /// float decode (extensions/jwp-receipts.md §2.4.12).
+    pub fn variant(self) -> Option<&'static str> {
+        match self {
+            Self::Sole => None,
+            Self::Exact => Some("exact"),
+            Self::Float => Some("float"),
+        }
+    }
+}
+
+/// A decode's output, with what produced it: the capability the payload decoded through
+/// and the path of that capability that ran (extensions/determinism-tiers.md §5).
+#[derive(Debug, Clone)]
+#[non_exhaustive]
+pub struct Dispatched {
+    /// The capability the payload decoded through: a `WAI1` manifest's `capability` or
+    /// its `fallback` (SPEC §4 steps 3 and 5), or the picked `WAI2` rendition's.
+    pub capability: String,
+    /// Which of that capability's decode paths ran.
+    pub path: DecodePath,
+    /// The decoded output.
+    pub output: Decoded,
+}
+
 #[derive(Debug)]
 #[non_exhaustive]
 pub enum DecodeError {
@@ -605,9 +653,10 @@ fn is_unsupported(e: &DecodeError, cap: &str) -> bool {
 }
 
 /// Decode `payload` through `capability` against `params` — the per-capability
-/// dispatch, shared by the `WAI1` and `WAI2` paths.
-pub fn decode_payload(capability: &str, payload: &[u8], params: &ParamSource) -> Result<Decoded, DecodeError> {
-    decode_payload_within(capability, payload, params, None)
+/// dispatch, shared by the `WAI1` and `WAI2` paths. The result names the capability and
+/// the path of it that ran ([`Dispatched`]).
+pub fn decode_payload(capability: &str, payload: &[u8], params: &ParamSource) -> Result<Dispatched, DecodeError> {
+    decode_payload_within(capability, payload, params, None).map(|(output, path)| Dispatched { capability: capability.to_string(), path, output })
 }
 
 /// What decoding `payload` through the integer capability `capability` against
@@ -624,11 +673,14 @@ pub fn declared_cost(capability: &str, payload: &[u8], params: &ParamSource) ->
 }
 
 /// [`decode_payload`] with a cost limit applied to the integer capabilities first
-/// ([`ModelRegistry::with_cost_limit`]).
+/// ([`ModelRegistry::with_cost_limit`]): the output, and the path that gave it.
 fn decode_payload_within(
     capability: &str, payload: &[u8], params: &ParamSource, limit: Option<&crate::int_payload::DeclaredCost>,
-) -> Result<Decoded, DecodeError> {
-    match capability {
+) -> Result<(Decoded, DecodePath), DecodeError> {
+    if capability == crate::codecs::CAP_IMAGE_JPEGAI {
+        return decode_jpegai(payload, params);
+    }
+    let sole = match capability {
         "wai.neural.encodec32"    => audio::decode(payload, params, 32_000).map(Decoded::Audio),
         "wai.neural.dac"          => audio::decode(payload, params, 44_100).map(Decoded::Audio),
         "wai.neural.mimi"         => audio::decode(payload, params, 24_000).map(Decoded::Audio),
@@ -662,41 +714,45 @@ fn decode_payload_within(
         // orchestration; their temporal nets are sink-supplied.
         "wai.neural.dcvc_rt" | "wai.neural.dcvc_fm"
                                       => video_dcvc::decode(payload, params).map(Decoded::Video),
-        // ISO/IEC 6048 JPEG AI over a sink-supplied model bundle (several files).
-        "wai.image.jpegai" => {
-            // Both decode paths implement one configuration (tools_0, β = 0, the
-            // high-operation-point synthesis, one eICCI model pair, 256×256 4:4:4);
-            // a codestream that asks for anything else is refused, not mis-decoded.
-            crate::jpegai_decode::check_codestream(payload).map_err(|why| {
-                DecodeError::InvalidPayload(format!("wai.image.jpegai: unsupported codestream: {why}"))
-            })?;
-            // Prefer the exact decoder when the parameters include the integer weight
-            // bundle (`norm.json` + `weights/*.json`) — that path pins one
-            // reconstruction, reproducible across machines under the conditions in
-            // `extensions/determinism-tiers.md` §5. Without the bundle, or in a
-            // non-default floating-point environment, the ORT float decode runs; it
-            // needs files a set holding only the exact bundle does not contain, so a
-            // pinned decode is then refused (`set_incomplete`) rather than run on
-            // unverified files. A refusal of the exact decode itself (`model_mismatch`
-            // for its bundle, `stream_exhausted` for a substream, integer-operators
-            // §13.6.1) is the decode's result: it does not fall back to the float
-            // decode. Both paths decode the scale path and the entropy through the
-            // integer operator set (`jpegai_ops`), which is all the capability
-            // guarantees (entropy-consistency, `NeuralEntropyExact`).
-            match jpegai_exact(payload, params) {
-                Exact::Decoded(img) => Ok(Decoded::Image(img)),
-                Exact::Refused(e) => Err(e),
-                Exact::Unavailable => {
-                    let to_err = |e: jpegai::JpegAiError| match e {
-                        jpegai::JpegAiError::Pin(r) => DecodeError::PriorMismatch(r),
-                        other => DecodeError::InvalidPayload(other.to_string()),
-                    };
-                    let model = jpegai::JpegAiModel::load_from(params.clone()).map_err(to_err)?;
-                    jpegai::decode(payload, &model).map(Decoded::Image).map_err(to_err)
-                }
-            }
-        }
         other => Err(DecodeError::UnknownCapability(other.to_string())),
+    };
+    sole.map(|d| (d, DecodePath::Sole))
+}
+
+/// ISO/IEC 6048 JPEG AI over a sink-supplied model bundle (several files), through the
+/// exact decoder or the float decode, with the path that ran.
+fn decode_jpegai(payload: &[u8], params: &ParamSource) -> Result<(Decoded, DecodePath), DecodeError> {
+    // Both decode paths implement one configuration (tools_0, β = 0, the
+    // high-operation-point synthesis, one eICCI model pair, 256×256 4:4:4);
+    // a codestream that asks for anything else is refused, not mis-decoded.
+    crate::jpegai_decode::check_codestream(payload).map_err(|why| {
+        DecodeError::InvalidPayload(format!("wai.image.jpegai: unsupported codestream: {why}"))
+    })?;
+    // Prefer the exact decoder when the parameters include the integer weight
+    // bundle (`norm.json` + `weights/*.json`) — that path pins one
+    // reconstruction, reproducible across machines under the conditions in
+    // `extensions/determinism-tiers.md` §5. Without the bundle, or in a
+    // non-default floating-point environment, the ORT float decode runs; it
+    // needs files a set holding only the exact bundle does not contain, so a
+    // pinned decode is then refused (`set_incomplete`) rather than run on
+    // unverified files. A refusal of the exact decode itself (`model_mismatch`
+    // for its bundle, `stream_exhausted` for a substream, integer-operators
+    // §13.6.1) is the decode's result: it does not fall back to the float
+    // decode. Both paths decode the scale path and the entropy through the
+    // integer operator set (`jpegai_ops`), which is all the capability
+    // guarantees (entropy-consistency, `NeuralEntropyExact`). The path that ran is
+    // reported with the image, so a receipt or a binding check can tell them apart.
+    match jpegai_exact(payload, params) {
+        Exact::Decoded(img) => Ok((Decoded::Image(img), DecodePath::Exact)),
+        Exact::Refused(e) => Err(e),
+        Exact::Unavailable => {
+            let to_err = |e: jpegai::JpegAiError| match e {
+                jpegai::JpegAiError::Pin(r) => DecodeError::PriorMismatch(r),
+                other => DecodeError::InvalidPayload(other.to_string()),
+            };
+            let model = jpegai::JpegAiModel::load_from(params.clone()).map_err(to_err)?;
+            jpegai::decode(payload, &model).map(|img| (Decoded::Image(img), DecodePath::Float)).map_err(to_err)
+        }
     }
 }
 
@@ -774,7 +830,7 @@ fn decode_envelope_with(
     env: &Wai,
     registry: &ModelRegistry,
     gate: impl Fn(&str) -> Result<(), DecodeError>,
-) -> Result<Decoded, DecodeError> {
+) -> Result<Dispatched, DecodeError> {
     let mut excluded: Vec<String> = Vec::new();
     // The first refusal raised after dispatch picked a capability — `set_incomplete`
     // from a pinned decode, or the key log's refusal of an unpinned one — with that
@@ -822,7 +878,8 @@ fn decode_envelope_with(
                     incomplete = Some((d.capability.to_string(), e));
                 }
             }
-            r => return r,
+            Err(e) => return Err(e),
+            Ok((output, path)) => return Ok(Dispatched { capability: d.capability.to_string(), path, output }),
         }
     }
 }
@@ -833,8 +890,10 @@ fn decode_envelope_with(
 /// verified. An envelope whose manifest declares an encryption is inert here
 /// ([`DecodeError::Encrypted`], SPEC §6.4 step 2): this sink opens no scheme. Returns [`DecodeError::PriorMismatch`] with the first refusal when no
 /// capability's pin resolves, and [`DecodeError::ModelNotRegistered`] when the
-/// registry holds neither capability.
-pub fn decode_envelope(env: &Wai, registry: &ModelRegistry) -> Result<Decoded, DecodeError> {
+/// registry holds neither capability. The output comes with the capability it decoded
+/// through and the path of it that ran ([`Dispatched`]), which a receipt of the decode
+/// states (extensions/jwp-receipts.md §2.4.12).
+pub fn decode_envelope(env: &Wai, registry: &ModelRegistry) -> Result<Dispatched, DecodeError> {
     decode_envelope_with(env, registry, |_| Ok(()))
 }
 
@@ -883,7 +942,7 @@ fn replicate_gate(cap: &str, refusal: Option<crate::codecs::ReplicateRefusal>) -
 /// inference through a runtime is not byte-identical across hardware, and even a
 /// fixed-order float stage depends on the build and on the process's floating-point
 /// environment, which neither the envelope nor the registry can state.
-pub fn decode_envelope_strict(env: &Wai, registry: &ModelRegistry) -> Result<Decoded, DecodeError> {
+pub fn decode_envelope_strict(env: &Wai, registry: &ModelRegistry) -> Result<Dispatched, DecodeError> {
     let cap = env.capability();
     let mut candidates: Vec<&str> = [Some(cap), env.fallback()].into_iter().flatten()
         .filter(|c| crate::codecs::is_dispatchable(c) && registry.advertises(c))
@@ -904,10 +963,11 @@ pub fn decode_envelope_strict(env: &Wai, registry: &ModelRegistry) -> Result<Dec
 /// pinned decode refused with `set_incomplete`, an unpinned entry whose set the
 /// registry's key log refuses, or an integer payload over the registry's cost limit
 /// (`cost_over_limit`), excludes that entry and picks again.
-/// Returns the picked index with the output, or [`DecodeError::NoRendition`] with a
+/// Returns the picked index with the output, the rendition's capability and the path
+/// of it that ran ([`Dispatched`]), or [`DecodeError::NoRendition`] with a
 /// reason per entry; an envelope whose manifest declares an encryption gives every
 /// entry `unsupported_encryption` (SPEC §6.4 step 2).
-pub fn decode_multi(m: &WaiMulti, registry: &ModelRegistry) -> Result<(usize, Decoded), DecodeError> {
+pub fn decode_multi(m: &WaiMulti, registry: &ModelRegistry) -> Result<(usize, Dispatched), DecodeError> {
     decode_multi_where(m, registry, |_| Ok(()))
 }
 
@@ -915,7 +975,7 @@ fn decode_multi_where(
     m: &WaiMulti,
     registry: &ModelRegistry,
     admit: impl Fn(usize) -> Result<(), Skip>,
-) -> Result<(usize, Decoded), DecodeError> {
+) -> Result<(usize, Dispatched), DecodeError> {
     let mut incomplete: Vec<(usize, String)> = Vec::new();
     loop {
         let picked = m.pick_where(
@@ -953,7 +1013,7 @@ fn decode_multi_where(
                         incomplete.push((p.index, format!("unsupported_capability: {e}")));
                     }
                     Err(e) => return Err(e),
-                    Ok(d) => return Ok((p.index, d)),
+                    Ok((output, path)) => return Ok((p.index, Dispatched { capability: p.meta.capability.clone(), path, output })),
                 }
             }
             Err(skips) => {
@@ -974,7 +1034,7 @@ fn decode_multi_where(
 /// one it leaves out is reported with its `replicate_*` reason.
 /// When none is admitted the result is [`DecodeError::UnsoundReplicate`] with the
 /// first selectable entry's reason.
-pub fn decode_multi_strict(m: &WaiMulti, registry: &ModelRegistry) -> Result<(usize, Decoded), DecodeError> {
+pub fn decode_multi_strict(m: &WaiMulti, registry: &ModelRegistry) -> Result<(usize, Dispatched), DecodeError> {
     if m.intent() == "replicate" {
         let selectable: Vec<usize> = m.manifest.renditions.iter().enumerate()
             .filter(|(_, r)| r.is_selectable()).map(|(i, _)| i).collect();
@@ -992,6 +1052,51 @@ pub fn decode_multi_strict(m: &WaiMulti, registry: &ModelRegistry) -> Result<(us
     decode_multi(m, registry)
 }
 
+/// What checking a reconstruction binding against a decode found
+/// ([`check_reconstruction_binding`], extensions/reconstruction-binding.md §2).
+#[cfg(feature = "provenance")]
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+#[non_exhaustive]
+pub enum BindingCheck {
+    /// The decode computed the reconstruction the binding is over, and its canonical
+    /// buffer hashes to the binding's hash.
+    Holds,
+    /// The decode computed that reconstruction, and its buffer hashes to another value.
+    Differs,
+    /// The decode ran a path that does not compute the reconstruction the binding is
+    /// over: `wai.image.jpegai`'s float decode, or a capability at neither
+    /// `decode-equivalence` nor that capability's exact decoder. The binding was not
+    /// checked, and a mismatch would not be evidence that the asset changed.
+    NotChecked(DecodePath),
+}
+
+/// Check a reconstruction binding over the decode of a parsed `WAI1` envelope
+/// (extensions/reconstruction-binding.md §2): decode it as [`decode_envelope`] does, then
+/// compare its canonical `RGB8` buffer with the binding only where the path that ran
+/// computes the reconstruction the binding is over: a capability whose registry class
+/// gives `decode-equivalence`, or `wai.image.jpegai`'s exact decoder. Over any other
+/// path the binding is reported as not checked, with that path, never as failed. A
+/// decode that is not a picture is [`DecodeError::InvalidPayload`]: the binding hashes a
+/// picture.
+#[cfg(feature = "provenance")]
+pub fn check_reconstruction_binding(
+    env: &Wai, registry: &ModelRegistry, binding: &crate::reconstruction_binding::ReconstructionBinding,
+) -> Result<BindingCheck, DecodeError> {
+    let d = decode_envelope(env, registry)?;
+    let exact = match d.path {
+        DecodePath::Exact => true,
+        DecodePath::Float => false,
+        DecodePath::Sole => crate::codecs::class_tier(&d.capability) == Some(crate::codecs::DeclaredTier::DecodeEquivalence),
+    };
+    let Decoded::Image(img) = &d.output else {
+        return Err(DecodeError::InvalidPayload(format!("{}: a reconstruction binding is over a picture", d.capability)));
+    };
+    if !exact {
+        return Ok(BindingCheck::NotChecked(d.path));
+    }
+    Ok(if binding.verify(&img.rgb) { BindingCheck::Holds } else { BindingCheck::Differs })
+}
+
 /// What the exact JPEG AI decoder made of a decode ([`jpegai_exact`]).
 enum Exact {
     /// The parameters hold no integer weight bundle (no `norm.json`), or this target or
@@ -1161,8 +1266,10 @@ mod jpegai_exact_dispatch_tests {
             let e = e.unwrap();
             files.insert(format!("weights/{}", e.file_name().to_str().unwrap()), std::fs::read(e.path()).unwrap());
         }
-        let decode = |files: &std::collections::BTreeMap<String, Vec<u8>>| super::decode_payload("wai.image.jpegai", &payload, &super::ParamSource::UnpinnedRead(files.clone()));
+        let dispatched = |files: &std::collections::BTreeMap<String, Vec<u8>>| super::decode_payload("wai.image.jpegai", &payload, &super::ParamSource::UnpinnedRead(files.clone()));
+        let decode = |files: &std::collections::BTreeMap<String, Vec<u8>>| dispatched(files).map(|d| d.output);
         assert!(matches!(decode(&files), Ok(super::Decoded::Image(_))), "the intact bundle decodes");
+        assert_eq!(dispatched(&files).unwrap().path, super::DecodePath::Exact, "through the exact decoder, and says so");
         let mut bad = files.clone();
         let mut nv: serde_json::Value = serde_json::from_slice(&files["norm.json"]).unwrap();
         nv["bounds"][0] = serde_json::Value::from(256);
diff --git a/wai/wai-rs/tests/jpegai_decode_path.rs b/wai/wai-rs/tests/jpegai_decode_path.rs
new file mode 100644
index 000000000..ae23d7642
--- /dev/null
+++ b/wai/wai-rs/tests/jpegai_decode_path.rs
@@ -0,0 +1,173 @@
+//! The reference dispatch reports the decode path that ran (extensions/determinism-tiers.md
+//! §5), a receipt of the decode states it (extensions/jwp-receipts.md §2.4.12, a decoder
+//! entry's `variant`), and the reference check of a reconstruction binding reads it
+//! (extensions/reconstruction-binding.md §2).
+//!
+//! - A capability with one decode path reports `DecodePath::Sole`, and the capability it
+//!   decoded through: the manifest's `capability`, or its `fallback`. Committed inputs
+//!   only; it never skips.
+//! - `wai.image.jpegai` without the integer weight bundle runs its float decode through
+//!   ONNX Runtime. The result says `DecodePath::Float`; a session claim recorded from it
+//!   says `"variant":"float"` and verifies, and the same decode named under a
+//!   `NeuralIntegerExact` capability is refused (`session_variant_not_exact`); a
+//!   reconstruction binding over its own pixels is reported as not checked, never as
+//!   holding. Needs the float path's model bundle (`WAI_JPEGAI_ASSETS`, the local
+//!   JPEG-AI asset directory exported from the ISO/IEC 6048-1 reference software; not in
+//!   the repository).
+//! - With the integer weight bundle the exact decoder runs. The result says
+//!   `DecodePath::Exact`, the claim `"variant":"exact"`, and the binding check reports
+//!   whether the hash holds. Needs the bundle (`WAI_JPEGAI_WEIGHTS_DIR`, `WAI_JPEGAI_NORM`,
+//!   as `tests/jpegai_e2e_full.rs` reads them).
+//!
+//! Without their inputs the two JPEG-AI tests skip loudly, and with
+//! `WAI_REQUIRE_SOTA_VECTORS=1` they fail (support/require_vectors.rs).
+#![cfg(all(feature = "neural", feature = "receipt_claims", feature = "video_int"))]
+
+#[path = "support/require_vectors.rs"]
+mod require_vectors;
+
+use std::path::{Path, PathBuf};
+
+use ed25519_dalek::SigningKey;
+use wai::claim_receipt::{verify_claim_json, ClaimBody, Rejection, SessionHost, SessionPath};
+use wai::codecs::{CAP_IMAGE_JPEGAI, CAP_NEURAL_INT_HYPER, CAP_VIDEO_INT_HYPER, CAP_VIDEO_INT_MOTION};
+use wai::container::{Conditioning, Manifest, ModelRequirement, Wai};
+use wai::energy_report::{BuildProfile, BuildRecord};
+use wai::neural::{check_reconstruction_binding, decode_envelope, BindingCheck, DecodePath, Decoded, Dispatched, ModelRegistry};
+use wai::reconstruction_binding::{ReconstructionBinding, ReconstructionDescriptor};
+use wai::session_recorder::{DecoderKey, SessionRecorder};
+
+const TEST256: &[u8] = include_bytes!("vectors/jpegai/test256.bin");
+const T: &str = "2026-11-04T12:00:00Z";
+
+fn envelope(media: &str, req: ModelRequirement, payload: Vec<u8>) -> Wai {
+    Wai::new(Manifest::new(media, "improve", req, Conditioning::new("none")), payload)
+}
+
+fn picture() -> Wai {
+    envelope("image", ModelRequirement::new(CAP_IMAGE_JPEGAI), TEST256.to_vec())
+}
+
+/// The session claim a sink records of one decode of `env`, its decoder entry naming
+/// `capability` on the path the dispatch reported, sealed and checked as
+/// `SessionRecorder::finish` checks it.
+fn record(env: &Wai, capability: &str, d: &Dispatched) -> Result<String, Rejection> {
+    let Decoded::Image(img) = &d.output else { panic!("{}: not a picture", d.capability) };
+    let bytes = env.to_bytes().expect("an envelope");
+    let host = SessionHost { profile: "native-ffi".into(), target: None, build: BuildRecord { profile: BuildProfile::Release, toolchain: None, target: None }, threads: 1 };
+    let mut rec = SessionRecorder::new(wai::pin::sha256(b"jpegai_decode_path: a session nonce"), "image", host, None);
+    let entry = rec.decoder(DecoderKey {
+        capability: capability.to_string(),
+        rendition: None,
+        object: wai::jwp_hash::content_hash(&bytes),
+        first_unit: 0,
+        path: SessionPath::Primary,
+        decoder: format!("builtin:wai@{}", env!("CARGO_PKG_VERSION")),
+        variant: d.path.variant().map(str::to_owned),
+        pin: None,
+        tier: None,
+        module_status: None,
+    });
+    rec.received_envelope(&bytes, SessionPath::Primary);
+    rec.presented(0, 1, entry, wai::pin::sha256(&img.rgb), None).expect("one presentation");
+    let key = SigningKey::from_bytes(&[0x44; 32]);
+    rec.finish(wai::jwp_hash::content_hash(&bytes), &key, "did:example:sink", 0, None, T).map(|s| s.claim.to_json())
+}
+
+/// The variant the one decoder entry of a verified session claim states.
+fn variant_of(wire: &str) -> Option<String> {
+    let c = verify_claim_json(wire).expect("the claim verifies");
+    let ClaimBody::ReconstructSession(s) = &c.body else { panic!("a session claim") };
+    s.decoders[0].variant.clone()
+}
+
+fn descriptor() -> ReconstructionDescriptor {
+    ReconstructionDescriptor {
+        decoder_id: 2,
+        profile: "tools_0".into(),
+        level: "L1".into(),
+        width: 256,
+        height: 256,
+        format: "RGB8".into(),
+        color_space: "bt709".into(),
+    }
+}
+
+/// A capability with one decode path reports it, with the capability the payload decoded
+/// through: the manifest's own, and the fallback when the sink does not hold the first.
+#[test]
+fn a_capability_with_one_path_reports_it_and_the_capability_it_decoded_through() {
+    let clip = std::fs::read(Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/vectors/int_video/int_video_bitstream.wiv1")).unwrap();
+    let reg = ModelRegistry::new().builtin(CAP_VIDEO_INT_MOTION);
+    let own = decode_envelope(&envelope("video", ModelRequirement::new(CAP_VIDEO_INT_MOTION), clip.clone()), &reg).expect("decodes");
+    assert_eq!((own.capability.as_str(), own.path, own.path.variant()), (CAP_VIDEO_INT_MOTION, DecodePath::Sole, None));
+    let fallback = envelope("video", ModelRequirement::new(CAP_VIDEO_INT_HYPER).with_fallback(CAP_VIDEO_INT_MOTION), clip);
+    let via = decode_envelope(&fallback, &reg).expect("decodes through the fallback");
+    assert_eq!((via.capability.as_str(), via.path), (CAP_VIDEO_INT_MOTION, DecodePath::Sole));
+    assert!(matches!(via.output, Decoded::Int(_)));
+}
+
+/// The float decode of `wai.image.jpegai` says so in the result, in the claim and in the
+/// binding check.
+#[test]
+fn the_jpegai_float_fallback_reports_itself() {
+    const TEST: &str = "jpegai_decode_path::the_jpegai_float_fallback_reports_itself";
+    const PRODUCE: &str = "the local JPEG-AI asset directory wai/jpegai-vm-assets (exported from the \
+ISO/IEC 6048-1 reference software; not in the repository), or WAI_JPEGAI_ASSETS=<dir>";
+    let assets = require_vectors::env_dir(TEST, "WAI_JPEGAI_ASSETS", "../jpegai-vm-assets", PRODUCE);
+    if !assets.join("model16/CCS_SGMM").exists() || !assets.join("jpegai_scaler_vec.json").exists() {
+        assets.absent(TEST, &format!("the float path's model bundle is not in {}", assets.display()), PRODUCE);
+        return;
+    }
+    assert!(!assets.join("norm.json").exists(), "the float path's bundle must not hold the exact decoder's");
+    let env = picture();
+    let reg = ModelRegistry::new().register(CAP_IMAGE_JPEGAI, assets.path());
+    let d = decode_envelope(&env, &reg).expect("the float decode");
+    assert_eq!((d.capability.as_str(), d.path, d.path.variant()), (CAP_IMAGE_JPEGAI, DecodePath::Float, Some("float")));
+    let Decoded::Image(img) = &d.output else { panic!("not a picture") };
+    assert_eq!((img.width, img.height, img.rgb.len()), (256, 256, 256 * 256 * 3));
+
+    // The receipt states the path that ran, and verifies.
+    let wire = record(&env, CAP_IMAGE_JPEGAI, &d).expect("a float decode of wai.image.jpegai is recorded under it");
+    assert!(wire.contains("\"variant\":\"float\""), "{wire}");
+    assert_eq!(variant_of(&wire).as_deref(), Some("float"));
+    // Named under a NeuralIntegerExact capability, the same decode is no decode of it.
+    let refused = record(&env, CAP_NEURAL_INT_HYPER, &d).expect_err("a float path under a NeuralIntegerExact capability");
+    assert_eq!((refused.step, refused.reason.label()), (4, "session_variant_not_exact"));
+
+    // A binding over this very decode's pixels is not checked by it, and not failed.
+    let binding = ReconstructionBinding::bind(&img.rgb, descriptor());
+    assert_eq!(check_reconstruction_binding(&env, &reg, &binding).unwrap(), BindingCheck::NotChecked(DecodePath::Float));
+}
+
+/// The exact decoder of `wai.image.jpegai` says so in the result and the claim, and the
+/// binding check reads its pixels.
+#[test]
+fn the_jpegai_exact_decoder_reports_itself() {
+    const TEST: &str = "jpegai_decode_path::the_jpegai_exact_decoder_reports_itself";
+    const PRODUCE: &str = "tools/wai_jpegai_fetch_e2e_inputs.sh (the models root's jpegai/v1/, pinned by tools/jpegai-e2e-v1.sha256)";
+    let Some([dir, norm]) = require_vectors::env_paths(TEST, ["WAI_JPEGAI_WEIGHTS_DIR", "WAI_JPEGAI_NORM"], PRODUCE) else { return };
+    // The bundle as the dispatch reads it: norm.json, and weights/<member>.json beside it.
+    let root: PathBuf = std::env::temp_dir().join(format!("wai-jpegai-decode-path-{}", std::process::id()));
+    std::fs::create_dir_all(root.join("weights")).unwrap();
+    std::fs::copy(&norm, root.join("norm.json")).unwrap();
+    for m in ["hs_y", "hs_uv", "hyp_y", "hyp_uv", "synth_y", "synth_uv", "mcm0_y", "mcm1_y", "mcm2_y", "mcm3_y", "icci"] {
+        std::fs::copy(Path::new(&dir).join(format!("{m}.json")), root.join(format!("weights/{m}.json"))).unwrap();
+    }
+    let env = picture();
+    let reg = ModelRegistry::new().register(CAP_IMAGE_JPEGAI, &root);
+    let d = decode_envelope(&env, &reg).expect("the exact decode");
+    assert_eq!((d.capability.as_str(), d.path, d.path.variant()), (CAP_IMAGE_JPEGAI, DecodePath::Exact, Some("exact")));
+    let Decoded::Image(img) = &d.output else { panic!("not a picture") };
+
+    let wire = record(&env, CAP_IMAGE_JPEGAI, &d).expect("an exact decode is recorded");
+    assert_eq!(variant_of(&wire).as_deref(), Some("exact"));
+
+    let binding = ReconstructionBinding::bind(&img.rgb, descriptor());
+    assert_eq!(check_reconstruction_binding(&env, &reg, &binding).unwrap(), BindingCheck::Holds);
+    let mut other = img.rgb.clone();
+    other[0] ^= 1;
+    let changed = ReconstructionBinding::bind(&other, descriptor());
+    assert_eq!(check_reconstruction_binding(&env, &reg, &changed).unwrap(), BindingCheck::Differs);
+    std::fs::remove_dir_all(&root).ok();
+}
diff --git a/wai/wai-rs/tests/neural_cost_limit.rs b/wai/wai-rs/tests/neural_cost_limit.rs
index 4450594bd..371d68354 100644
--- a/wai/wai-rs/tests/neural_cost_limit.rs
+++ b/wai/wai-rs/tests/neural_cost_limit.rs
@@ -144,7 +144,7 @@ fn a_payload_over_the_cost_limit_is_refused_before_any_symbol_is_decoded() {
     let n_frames = wire::parse_wiv1(&small, Some(0)).unwrap().n_frames as usize;
     let m = WaiMulti::new("video", "improve", serde_json::Value::Null,
                           vec![Rendition::media(CAP_VIDEO_INT_MOTION, "bytes", bomb.clone()), Rendition::media(CAP_VIDEO_INT_MOTION, "bytes", small)]);
-    let (r, peak) = peak_during(|| decode_multi(&m, &reg));
+    let (r, peak) = peak_during(|| decode_multi(&m, &reg).map(|(i, d)| (i, d.output)));
     match r {
         Ok((1, Decoded::Int(IntDecoded::Video { frames, .. }))) => assert_eq!(frames.len(), n_frames),
         Ok((i, _)) => panic!("decoded entry {i}"),
diff --git a/wai/wai-rs/tests/neural_native_decode.rs b/wai/wai-rs/tests/neural_native_decode.rs
index f4a2011fa..d98fb0d09 100644
--- a/wai/wai-rs/tests/neural_native_decode.rs
+++ b/wai/wai-rs/tests/neural_native_decode.rs
@@ -16,9 +16,26 @@
 use std::path::{Path, PathBuf};
 
 use wai::container::{Conditioning, Manifest, ModelRequirement, Prior, Wai};
-use wai::neural::{
-    decode_envelope, decode_envelope_strict, model_sha256, DecodeError, Decoded, ModelRegistry,
-};
+use wai::neural::{model_sha256, DecodeError, Decoded, ModelRegistry};
+
+// These tests are about what a decode outputs or refuses. The dispatch also reports the
+// capability it decoded through and the path that ran (`neural::Dispatched`), which
+// tests/jpegai_decode_path.rs tests; here each dispatch is read for its output.
+fn decode_envelope(env: &Wai, reg: &ModelRegistry) -> Result<Decoded, DecodeError> {
+    wai::neural::decode_envelope(env, reg).map(|d| d.output)
+}
+fn decode_envelope_strict(env: &Wai, reg: &ModelRegistry) -> Result<Decoded, DecodeError> {
+    wai::neural::decode_envelope_strict(env, reg).map(|d| d.output)
+}
+fn decode_multi(m: &wai::container::WaiMulti, reg: &ModelRegistry) -> Result<(usize, Decoded), DecodeError> {
+    wai::neural::decode_multi(m, reg).map(|(i, d)| (i, d.output))
+}
+fn decode_multi_strict(m: &wai::container::WaiMulti, reg: &ModelRegistry) -> Result<(usize, Decoded), DecodeError> {
+    wai::neural::decode_multi_strict(m, reg).map(|(i, d)| (i, d.output))
+}
+fn decode_payload(cap: &str, payload: &[u8], params: &wai::neural::ParamSource) -> Result<Decoded, DecodeError> {
+    wai::neural::decode_payload(cap, payload, params).map(|d| d.output)
+}
 
 /// The reason a pinned decode was refused with, or a description of what happened.
 fn refusal(r: Result<Decoded, DecodeError>) -> String {
@@ -611,7 +628,7 @@ fn int_synth_check_5_covers_the_spec_a_model_directory_holds() {
     let reg = ModelRegistry::new().register(cap, dir.clone()).with_key_log(Revokes(vec![]));
     let params = reg.resolve_verified(cap, None).expect("the unrevoked spec resolves");
     std::fs::copy(src.join("../bloom/int_synth.json"), dir.join("int_synth.json")).unwrap();
-    assert_eq!(rgb_hash(wai::neural::decode_payload(cap, &latent, &params)), want, "the decode read what check 5 ran over");
+    assert_eq!(rgb_hash(decode_payload(cap, &latent, &params)), want, "the decode read what check 5 ran over");
     std::fs::remove_dir_all(&dir).ok();
 }
 
@@ -663,15 +680,15 @@ fn pinned_decode_reads_only_verified_bytes() {
     let reg = ModelRegistry::new().register("wai.neural.int_synth", dir.join("model"));
     let pin = Prior::new("gradient", wai::pin::sha256_hex(&spec), "int-synth/i64");
     let params = reg.resolve_verified("wai.neural.int_synth", Some(&pin)).expect("verified");
-    let before = wai::neural::decode_payload("wai.neural.int_synth", &latent, &params);
+    let before = decode_payload("wai.neural.int_synth", &latent, &params);
     // Replace the spec on disk with another one (the bloom vector's).
     std::fs::copy(src.join("../bloom/int_synth.json"), dir.join("int_synth.json")).unwrap();
-    let after = wai::neural::decode_payload("wai.neural.int_synth", &latent, &params);
+    let after = decode_payload("wai.neural.int_synth", &latent, &params);
     assert_eq!(image_fnv(before), image_fnv(after), "the decode read the bytes it verified");
     // An unpinned decode reads the file as it is now, and so differs (or fails).
-    let unpinned = wai::neural::decode_payload("wai.neural.int_synth", &latent,
+    let unpinned = decode_payload("wai.neural.int_synth", &latent,
         &reg.resolve_verified("wai.neural.int_synth", None).unwrap());
-    assert_ne!(image_fnv(unpinned), image_fnv(wai::neural::decode_payload("wai.neural.int_synth", &latent, &params)));
+    assert_ne!(image_fnv(unpinned), image_fnv(decode_payload("wai.neural.int_synth", &latent, &params)));
     std::fs::remove_dir_all(&dir).ok();
 }
 
@@ -763,7 +780,6 @@ fn set_incomplete_at_step_3_outranks_the_fallback_refusal() {
 #[test]
 fn decode_multi_strict_reports_replicate_reasons() {
     use wai::container::{Rendition, WaiMulti};
-    use wai::neural::decode_multi_strict;
     let d = int_mlicpp_dir();
     let pl = Wai::from_bytes(&std::fs::read(d.join("stream.wai")).unwrap()).unwrap().payload;
     let m = WaiMulti::new("image", "replicate", serde_json::Value::Null, vec![
@@ -792,7 +808,6 @@ fn decode_multi_strict_reports_replicate_reasons() {
 fn wai2_decode_multi_picks_the_verified_rendition() {
     use wai::codecs::DeclaredTier;
     use wai::container::{Rendition, Supersedes, WaiMulti};
-    use wai::neural::{decode_multi, decode_multi_strict};
     let d = int_mlicpp_dir();
     let pl = Wai::from_bytes(&std::fs::read(d.join("stream.wai")).unwrap()).unwrap().payload;
     let committed = model_sha256(&d.join("model.wmb")).unwrap();
@@ -1508,7 +1523,7 @@ fn a_granted_listing_authorises_only_the_files_it_names() {
 /// are not the tables the decode reads. With no key log, the decode reads the path.
 #[test]
 fn a_revoke_only_key_log_checks_and_reads_the_set_once() {
-    use wai::neural::{decode_payload, ParamSource};
+    use wai::neural::ParamSource;
     use wai::pin::Verified;
     let bm = "wai.neural.bmshj2018";
     let root = std::env::temp_dir().join(format!("wai-revoke-only-{}", std::process::id()));
@@ -1576,7 +1591,6 @@ fn a_revoke_only_key_log_checks_and_reads_the_set_once() {
 fn an_encrypted_envelope_is_inert_at_the_reference_sink() {
     use wai::codecs::CAP_NEURAL_INT_HYPER as HYP;
     use wai::container::{Rendition, Security, WaiMulti};
-    use wai::neural::decode_multi;
     let v = Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/vectors");
     let dir = std::env::temp_dir().join(format!("wai-encrypted-{}", std::process::id()));
     std::fs::create_dir_all(&dir).unwrap();
-- 
2.54.0 (Apple Git-157)


From e72b52057129ac154e7b39f4e3a391ae6e1d4f13 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:53:15 -0500
Subject: [PATCH 15/29] =?UTF-8?q?jwp-receipts=20=C2=A72.4.12:=20one=20exac?=
 =?UTF-8?q?t=20path=20for=20rules=2013=20and=2014=20and=20the=20stage=20ha?=
 =?UTF-8?q?sh,=20checked=20per=20object?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Rule 14 refused only `float` under a pin of decode-equivalence, while rule 13
and the stage-hash check read any variant but null or exact as not exact. Rules
6 and 13 already refuse such a pin for every class the registry gives, so rule
14 reaches only capabilities with no class, and there `fp16` passed. The three
now share one predicate, DecoderEntry::exact_path: null, `exact`, or, for a
NeuralIntegerExact capability, the lowering its registration names. The last is
how opus-dred §7.2's present entry names wai.audio.opus.dred.int's integer
path, `dred-int/1`, which rule 13 would otherwise have refused; opus-dred says
a session claim of that decode may name it so.

The stage-hash refusal is per object: an entry naming the base at a weaker step
counts whichever units it presented, so the outcome does not depend on whether
the verifier holds the outputs. determinism-tiers §3 rule 3 now says so. The
vector session_stage_hash_not_exact moves its fallback decode inside the set,
at unit 2, so it is refused under either reading, and the earlier form stays as
session_stage_hash_not_exact_outside_the_set, the per-object case.

determinism-tiers §3 calls the closed vocabulary a pin's, says rule 3 governs
decoded media while generation-equivalence carries its own digest, and says
rule 2 is held over the registry class. The README says rule 6 enforces it, with
an inter-frame capability's vector, and rule 14 is its per-path complement.

Five vectors are appended after the earlier ones, which a test pins by
digest: a registered lowering accepted, another capability's lowering
refused, `fp16` under decode-equivalence refused, an inter-frame capability
over its class refused, and the per-object case.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/claim-conformance/README.md               | 51 ++++++++++-----
 wai/claim-conformance/claim_verify.py         | 28 ++++++---
 wai/claim-conformance/differential.py         |  9 +--
 wai/claim-conformance/vectors.json            | 63 ++++++++++++++++++-
 wai/extensions/determinism-tiers.md           | 23 +++++--
 wai/extensions/jwp-receipts.md                | 38 ++++++-----
 wai/extensions/opus-dred.md                   |  6 ++
 wai/wai-rs/src/claim_receipt.rs               |  9 +--
 .../src/claim_receipt/corpus/session.rs       | 57 ++++++++++++++++-
 wai/wai-rs/src/claim_receipt/session.rs       | 32 +++++++---
 10 files changed, 250 insertions(+), 66 deletions(-)

diff --git a/wai/claim-conformance/README.md b/wai/claim-conformance/README.md
index 99c6e5eff..f3ce95f79 100644
--- a/wai/claim-conformance/README.md
+++ b/wai/claim-conformance/README.md
@@ -352,27 +352,46 @@ with neither is a hole; there is none.
 ### The decode path that ran
 
 A decoder entry's `variant` states which path of its decoder ran: `exact`, `float`, or
-`null` for a capability with one path (jwp-receipts §2.4.12). Three rules close the door
-determinism-tiers §3 closes (*No tier between the two*), each with a refusal here. The
-vectors are appended after every earlier vector, which a test pins by digest:
+`null` for a capability with one path (jwp-receipts §2.4.12). A `NeuralIntegerExact`
+capability's integer path may also be named by the lowering its registration names, as
+opus-dred §7.2 names `dred-int/1`. `null`, `exact` and that lowering are the exact paths;
+any other term is not. Three rules close the door determinism-tiers §3 closes (*No tier
+between the two*), each with a refusal here. The vectors are appended after every earlier
+vector, which a test pins by digest:
 
 1. Rule 13, `session_variant_not_exact`: a `NeuralIntegerExact` capability named with a
-   path other than its integer one. `session_variant_not_exact` names `wai.neural.int_hyper`
-   with `float` under a pin of `decode-equivalence`, and `session_variant_not_exact_unpinned`
-   with an unregistered term and no pin. Accepted: `session_variant_exact_integer_accept`,
-   the integer path stated as `exact`.
-2. Rule 14, `session_tier_exceeds_variant`: a `float` path stated under a pin of
+   path that is not exact. `session_variant_not_exact` names `wai.neural.int_hyper`
+   with `float` under a pin of `decode-equivalence`, `session_variant_not_exact_unpinned`
+   with an unregistered term and no pin, and
+   `session_variant_lowering_of_another_capability` with `dred-int/1`, the lowering of
+   another capability. Accepted: `session_variant_exact_integer_accept`, the integer path
+   stated as `exact`, and `session_variant_lowering_accept`, `wai.audio.opus.dred.int`'s
+   stated as `dred-int/1`.
+2. Rule 14, `session_tier_exceeds_variant`: a path that is not exact stated under a pin of
    `decode-equivalence`, over a capability the registry does not classify, decoded by a
-   module, which rule 6 does not reach. `session_tier_exceeds_class_before_variant` gives
-   `wai.image.jpegai`'s float path the same pin, which rule 6 refuses first. Accepted:
-   `session_variant_float_accept` and `session_variant_exact_accept`, `wai.image.jpegai`'s
-   two paths stated as they ran.
+   module: `float` (`session_tier_exceeds_variant`) and `fp16`
+   (`session_tier_exceeds_variant_unregistered_term`). `session_tier_exceeds_class_before_variant`
+   gives `wai.image.jpegai`'s float path the same pin, which rule 6 refuses first.
+   Accepted: `session_variant_float_accept` and `session_variant_exact_accept`,
+   `wai.image.jpegai`'s two paths stated as they ran.
+
+   determinism-tiers §3 rule 2 (a capability whose payload holds several frame types
+   takes the class of its weakest) is enforced over the registry class, by rule 6
+   (`session_tier_exceeds_class`): the class a capability is registered at is that of
+   its weakest frame type, so a pin above it is refused whatever path an entry states.
+   `session_tier_exceeds_class_inter_frame` gives `wai.neural.dcvc_rt`, an inter-frame
+   capability of the class `neural-float`, a pin of `decode-equivalence`. Rule 14 is the
+   per-path complement: it refuses the same pin over the path that ran, where the registry
+   gives no class.
 3. Against the stage claims, `session_stage_hash_not_exact`: the model session's stage
    claim stated at `decode-equivalence`, with its stage hash, a clip digest, while its base
-   was also decoded through its fallback, a capability held to a tolerance
-   (`session_stage_hash_not_exact`), while one of its layers is stated as decoded on a
-   `float` path (`session_stage_hash_over_a_float_layer`), or while its base is stated as
-   decoded with no pin (`session_stage_hash_over_an_unpinned_base`). Accepted:
+   was also decoded through its fallback, a capability held to a tolerance, for unit 2,
+   inside the set (`session_stage_hash_not_exact`); while the fallback entry names the
+   base and presents only unit 5, outside the set, which is refused all the same, the
+   check being per object (`session_stage_hash_not_exact_outside_the_set`); while one of
+   its layers is stated as decoded on a `float` path
+   (`session_stage_hash_over_a_float_layer`); or while its base is stated as decoded with
+   no pin (`session_stage_hash_over_an_unpinned_base`). Accepted:
    `session_stage_hash_exact_accept`.
 
 ## Steps 1 and 2: types, and the order a body is read in
diff --git a/wai/claim-conformance/claim_verify.py b/wai/claim-conformance/claim_verify.py
index 8ea18b832..6af68cabf 100644
--- a/wai/claim-conformance/claim_verify.py
+++ b/wai/claim-conformance/claim_verify.py
@@ -671,6 +671,17 @@ def replicate(cap):
     return c["replicate"] if c else None
 
 
+def exact_path(d):
+    """§2.4.12, a decoder entry's `variant`: whether it states an exact path — null, `exact`,
+    or, for a capability of the replicate class neural-int-exact, the lowering its
+    registration's `ops` member names (such as `dred-int/1`). Any other term is not exact."""
+    v = d["variant"]
+    if v is None or v == "exact":
+        return True
+    c = REGISTRY.get(d["capability"])
+    return c is not None and c["replicate"] == "neural-int-exact" and (c.get("ops") or {}).get("lowering") == v
+
+
 def class_tier(cap):
     """SPEC §3.1 *Tier*: the tier a capability's class gives — decode-equivalence for an
     integer-exact neural decode, entropy-consistency for an entropy-exact one, none for
@@ -732,12 +743,12 @@ def session_rules(b):
         need(e["joules_micro"] != 0, 4, "energy_zero_figure")
         acquisition(e)
     # Rule 13: a NeuralIntegerExact capability is decoded by its integer path only, which
-    # an entry states as variant null or exact (determinism-tiers §3, rule 1).
-    need(not any(d["variant"] is not None and d["variant"] != "exact" and replicate(d["capability"]) == "neural-int-exact"
-                 for d in ds), 4, "session_variant_not_exact")
-    # Rule 14: frames a float path decoded are not decode-equivalent, whatever the pin
-    # declares (rule 2 there).
-    need(not any(d["variant"] == "float" and d["tier"] == "decode-equivalence" for d in ds), 4,
+    # an entry states as an exact path (determinism-tiers §3, rule 1).
+    need(not any(replicate(d["capability"]) == "neural-int-exact" and not exact_path(d) for d in ds), 4,
+         "session_variant_not_exact")
+    # Rule 14: frames a path that is not exact decoded are not decode-equivalent, whatever
+    # the pin declares (rule 2 there).
+    need(not any(not exact_path(d) and d["tier"] == "decode-equivalence" for d in ds), 4,
          "session_tier_exceeds_variant")
 
 
@@ -1316,7 +1327,7 @@ def check_session_stage_claims(c, wires, ps, objs):
     # A stage hash is a clip digest over the set's units (staged-delivery §5): every entry
     # naming the set's base decodes it at decode-equivalence by that section's base step,
     # and no entry naming the base or a layer the claim lists ran a path that is not exact
-    # (determinism-tiers §3, rule 3).
+    # (determinism-tiers §3, rule 3), whichever units the entry presented.
     def base_exact(d):
         r = replicate(d["capability"])
         return r == "standard-defined" or (r == "neural-int-exact" and d["tier"] == "decode-equivalence")
@@ -1327,8 +1338,7 @@ def check_session_stage_claims(c, wires, ps, objs):
         layers = {o["object"] for o in s["body"]["objects"]}
         for d in b["decoders"]:
             base = d["object"] == s["subject"]
-            exact_path = d["variant"] is None or d["variant"] == "exact"
-            need(not (base and not base_exact(d)) and not ((base or d["object"] in layers) and not exact_path),
+            need(not (base and not base_exact(d)) and not ((base or d["object"] in layers) and not exact_path(d)),
                  6, "session_stage_hash_not_exact")
     if ps is None:
         return
diff --git a/wai/claim-conformance/differential.py b/wai/claim-conformance/differential.py
index aa37866ac..1ed7a7b59 100644
--- a/wai/claim-conformance/differential.py
+++ b/wai/claim-conformance/differential.py
@@ -841,17 +841,18 @@ def sessions(ins, rng, n):
                 restate(body, objects)
             elif k == 31 and ds:
                 # The decode path an entry states (§2.4.12 rules 13 and 14): exact, float, a
-                # term this revision does not register, or none.
-                rng.choice(ds)["variant"] = rng.choice([None, "exact", "float", "fp16"])
+                # term this revision does not register, a lowering a registration names
+                # (whm1/1 is the model session's integer capabilities' own), or none.
+                rng.choice(ds)["variant"] = rng.choice([None, "exact", "float", "fp16", "whm1/1", "dred-int/1"])
             elif k == 32 and ds:
                 # A float path under a pin of either tier, or none, through a capability the
                 # registry classifies or one it does not.
                 d = rng.choice(ds)
-                d["variant"] = rng.choice(["float", "exact"])
+                d["variant"] = rng.choice(["float", "exact", "fp16", "dred-int/1"])
                 d["pin"], d["tier"] = rng.choice([(CV.blake3(b"differential: a pin").hex(), "decode-equivalence"),
                                                   (CV.blake3(b"differential: a pin").hex(), "entropy-consistency"), (None, None)])
                 if rng.random() < 0.3:
-                    d["capability"] = rng.choice(["org.example.picture", "wai.image.jpegai"])
+                    d["capability"] = rng.choice(["org.example.picture", "wai.image.jpegai", "wai.audio.opus.dred.int"])
         if rng.random() < 0.5:
             ctx["redecoded"] = [{"index": j, "out_digest": rng.choice([p["out_digest"], CV.blake3(b"differential: redecoded").hex()])}
                                 for j, p in enumerate(ps) if rng.random() < 0.3]
diff --git a/wai/claim-conformance/vectors.json b/wai/claim-conformance/vectors.json
index 0b1740c1b..7c15c5c4e 100644
--- a/wai/claim-conformance/vectors.json
+++ b/wai/claim-conformance/vectors.json
@@ -24695,7 +24695,7 @@
     },
     {
       "name": "session_variant_not_exact_unpinned",
-      "about": "the same with no pin, and a path named by a term this revision does not register: only null or exact names the integer path",
+      "about": "the same with no pin, and a path named by a term this revision does not register: only null, exact or the capability's registered lowering names the integer path",
       "procedure": "claim",
       "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":\"fp16\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"95f4b49237eed8b2c76bfb0c9a00f9f8ffc5a8e2e6f47454e4a1c021c373a33fb16ddccfb3f69e7b98fe991028fff9e5d37d832b6ac50e35e28e3b28da468d06\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
       "expected": {
@@ -24704,6 +24704,27 @@
         "reason": "session_variant_not_exact"
       }
     },
+    {
+      "name": "session_variant_lowering_accept",
+      "about": "a NeuralIntegerExact capability whose entry names its integer path by the lowering its registration names, as opus-dred 7.2's present entry names wai.audio.opus.dred.int's, dred-int/1, under a pin of decode-equivalence",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.audio.opus.dred.int\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":\"dred-int/1\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"audio\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"f82c5ef640689d9c5c7aae666a97fc474c9a38aebab818be479c2aa5c6b616fe7567868a831a7a63e25ad0ae93667543f8fce63d53dadd2d5602fbc09e266802\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "accept",
+        "link_hash": "60fb99e3ef042894f22fdc683aa5220c5d0b9d65fe6034b5cc7729c5d6494980"
+      }
+    },
+    {
+      "name": "session_variant_lowering_of_another_capability",
+      "about": "wai.neural.int_hyper named with dred-int/1, the lowering another capability registers: not its integer path",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":\"dred-int/1\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"da9ecd87b76cd33eb1d89d0e5bcf214feacdd8b1ac20d91966ff0569da0643a17fc0364ba32b5f19b296bb1a75e246792d1a27af2b82bd4adc8160d74e42710e\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "reject",
+        "step": 4,
+        "reason": "session_variant_not_exact"
+      }
+    },
     {
       "name": "session_tier_exceeds_variant",
       "about": "a capability the registry does not classify, decoded by a module on its float path and stated under a pin of decode-equivalence: a float path's frames are not decode-equivalent, whatever the pin declares (determinism-tiers 3, rule 2)",
@@ -24715,6 +24736,17 @@
         "reason": "session_tier_exceeds_variant"
       }
     },
+    {
+      "name": "session_tier_exceeds_variant_unregistered_term",
+      "about": "the same capability on a path named by a term this revision does not register, fp16, under a pin of decode-equivalence: for a capability the registry does not classify, only null or exact names an exact path",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"org.example.picture\",\"decoder\":\"module:sha256:5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e5e\",\"first_unit\":0,\"leaves\":1,\"module_status\":\"pinned\",\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":\"fp16\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"2a8f2a28351478bdba85f5407d17d57723f6b90446b0748cf857320d423b94440e647b168acc99179c01e57a45d61d066e36d40ebd12f09934fa4ca9d01f4b00\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "reject",
+        "step": 4,
+        "reason": "session_tier_exceeds_variant"
+      }
+    },
     {
       "name": "session_tier_exceeds_class_before_variant",
       "about": "wai.image.jpegai's float path under a pin of decode-equivalence: the class rule, rule 6, refuses it first",
@@ -24726,6 +24758,17 @@
         "reason": "session_tier_exceeds_class"
       }
     },
+    {
+      "name": "session_tier_exceeds_class_inter_frame",
+      "about": "an inter-frame capability, wai.neural.dcvc_rt, under a pin of decode-equivalence: its registry class, neural-float, is the class of its weakest frame type, and rule 6 refuses a tier above it whatever path the entry states (determinism-tiers 3, rule 2)",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.dcvc_rt\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":null}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"video\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"1e5a701567e8d1b628bacf8aebba898025ee7ef83ae427959836c8f7a81281d02899706a89f8f4a7d9ba17cdabd4f155ee46eca616c3400c85c9aacaf943c405\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "expected": {
+        "outcome": "reject",
+        "step": 4,
+        "reason": "session_tier_exceeds_class"
+      }
+    },
     {
       "name": "session_stage_hash_exact_accept",
       "about": "the model session's stage set stated at decode-equivalence, with its stage hash: its base decoded only through a pinned integer capability, its layers through a residual refinement, every path exact",
@@ -24743,7 +24786,23 @@
     },
     {
       "name": "session_stage_hash_not_exact",
-      "about": "the same stage claim, while the base's own payload was also decoded through its fallback, a capability held to a tolerance: a clip digest over a sequence some of whose frames were decoded at a weaker tier (determinism-tiers 3, rule 3)",
+      "about": "the same stage claim, while the base's unit 2, inside the set, was decoded through the base's fallback, wai.video.av1, a capability held to a tolerance, by an entry of its own naming the base: a clip digest over a sequence one of whose frames was decoded at a weaker tier (determinism-tiers 3, rule 3)",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":3,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":5,\"leaves\":1,\"module_status\":null,\"object\":\"13b04249eca2b1a9061dba8ca2846169d3f5fa3548d9e138a0cf2cb553798cf4\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"3063938fe3dec97a08fde52f174c51f16dfd9511dfe82534e325fdf1fd60431b\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"deadline-threshold\",\"to\":\"fallback\",\"unit\":2},{\"from\":\"fallback\",\"reason\":\"recovered\",\"to\":\"primary\",\"unit\":3},{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":3,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"b0d40437162946ed862996e4a3e8223bf42daddce5ba8d0e8c5517ce223d5feaeb4f55b87553126acba2f277e89cc8d102e1a795afd95c398eab40c5907ac200\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
+      "context": {
+        "stage_claims": [
+          "{\"body\":{\"bytes_envelopes\":10607,\"conflicts\":0,\"deadline_misses\":2,\"objects\":[{\"bytes\":900,\"evidence\":\"none\",\"layer\":1,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":901,\"evidence\":\"none\",\"layer\":1,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":1}},{\"bytes\":902,\"evidence\":\"none\",\"layer\":1,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":2}},{\"bytes\":903,\"evidence\":\"none\",\"layer\":1,\"object\":\"792a12990c5207ca83a86b7420815f1f2d34918553a0130a4f9570a11354f2ca\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":3}},{\"bytes\":1000,\"evidence\":\"none\",\"layer\":2,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":1001,\"evidence\":\"none\",\"layer\":2,\"object\":\"1c6eae0ed41ed7fc1b2f00b46ea1de98f21a6d6cbf62bf6b22cf76a664a49376\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":1}}],\"op_committed\":null,\"op_delivered\":null,\"op_requested\":null,\"presented\":{\"wai-stage:layers\":0},\"stage_hash\":\"082bc00210e6d32f18f39ee723d863fd7b99f206a415ac2f44a8687793b2e068\",\"tier\":\"decode-equivalence\",\"units_by_stage\":[1,2,1],\"vocabulary\":\"urn:wai:staged-delivery:1\"},\"class\":\"stage\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0c29c7a2e9cfdb393273506b5acc9672b225192b28b9114f3b814d19c91bf4f994a74f2eecc077abbe6840c84f98f09ebbd2bd4f08d84592cca9a944918f590d\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}"
+        ]
+      },
+      "expected": {
+        "outcome": "reject",
+        "step": 6,
+        "reason": "session_stage_hash_not_exact"
+      }
+    },
+    {
+      "name": "session_stage_hash_not_exact_outside_the_set",
+      "about": "the same stage claim, while the fallback entry, wai.video.av1, names the base and presents only unit 5, outside the set: refused all the same, since the check is per object, whichever units an entry presented (determinism-tiers 3, rule 3)",
       "procedure": "claim",
       "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":4,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"b62a8425b7fe3a085a14f2f2156ea22ae3b4349884d799b5afe66dccc7d4dd7f\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":1,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"607fcbc40cc5a84709125d99f6b8fe2e3e840c218d6f8b70319c09cb92e40e4a1626d0c3d93a1e0d6efbfed6594ef0fbcdc1a0f762cebd6e98d1e4966b77d80b\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
       "context": {
diff --git a/wai/extensions/determinism-tiers.md b/wai/extensions/determinism-tiers.md
index bdf42d607..6405d4ffb 100644
--- a/wai/extensions/determinism-tiers.md
+++ b/wai/extensions/determinism-tiers.md
@@ -374,7 +374,7 @@ headroom and the rendition's digest in its rendition record (exact-tonemap §7).
   any conforming sink MUST come from a `decode-equivalence` capability. A hash
   over a reconstruction that one implementation pins (§5) is that
   implementation's, and a receipt MUST NOT present it as the capability's.
-- **No tier between the two (normative).** The tier vocabulary is closed:
+- **No tier between the two (normative).** The tier vocabulary of a pin is closed:
   `decode-equivalence`, `entropy-consistency` and `none` (SPEC §3.1). No tier,
   registry class or claim states that a reconstruction stays within a bound of the
   exact one between points at which it is made exact again. Such a bound is
@@ -398,14 +398,27 @@ headroom and the rendition's digest in its rendition record (exact-tonemap §7).
      a sequence of frames, covers only frames that were each decoded at
      `decode-equivalence`. A sequence in which any frame was decoded at a weaker
      tier, or by a path that does not compute the registered reconstruction, has
-     no such digest. Each frame of it that was decoded exactly keeps its own.
+     no such digest. Each frame of it that was decoded exactly keeps its own. A
+     receipt is checked per object, not per frame: a stage hash is refused where any
+     decoder entry decoded the set's base at a weaker tier, or the base or a layer of
+     the set on a path that is not exact, whichever units that entry presented
+     ([jwp-receipts](jwp-receipts.md) §2.4.12, step 6 against the stage claims,
+     item 5). So the outcome does not depend on whether the verifier holds the
+     outputs, which alone say what each entry presented.
+
+  These rules govern decoded media. A generated medium's tier,
+  generation-equivalence (SPEC §7, `GenerativeDeterministic`), is not a pin's tier:
+  it carries its own digest of the frames generated
+  ([synthesis](synthesis.md) §9.2; [avatars](avatars.md) §7.7), and rule 3 does not
+  void it.
 
   A receipt of a decode states the path that ran ([jwp-receipts](jwp-receipts.md)
   §2.4.12, a decoder entry's `variant`), and the claim corpus refuses a receipt
   that breaks each rule: a `NeuralIntegerExact` capability named with a path that
-  is not exact (rule 1); a pin's `tier` above what the path that ran holds
-  (rule 2); and a stage hash over a base decoded at a weaker tier by any of its
-  decoder entries (rule 3).
+  is not exact (rule 1); a pin's `tier` above its capability's registry class,
+  which is that of its weakest frame type, and above what the path that ran holds
+  (rule 2); and a stage hash over a base or a layer one of whose decoder entries
+  decoded it at a weaker tier or on a path that is not exact (rule 3).
 - **Post-decode processing.** A tier claim is void if the sink alters the medium
   after decode (§2, *What a tier covers*). This applies to the emitted result,
   not to the decoder's internal conformance.
diff --git a/wai/extensions/jwp-receipts.md b/wai/extensions/jwp-receipts.md
index d1f5a1222..3a78a28bd 100644
--- a/wai/extensions/jwp-receipts.md
+++ b/wai/extensions/jwp-receipts.md
@@ -1035,7 +1035,7 @@ A **decoder entry**:
 | `first_unit` | integer | The title's unit that is unit 0 of the object's base: of the object itself for a base or a rendition with no `refines`, and of the base it refines for a layer. A presentation through the entry of the title's unit `u` is of the object's unit `u − first_unit`. Where a verifier holds a layer's envelope and an entry names its base, the layer's entry has that entry's `first_unit`; where it holds a `stage` claim of a set, an entry naming the set's base or a layer the claim lists has the unit `stage_sets` names that claim at (step 6). |
 | `path` | string | `primary` or `fallback`. |
 | `decoder` | string | `builtin:<crate>@<version>` — `<crate>` 1 to 64 of `a`–`z`, `0`–`9`, `_` and `-`, `<version>` 1 to 64 of ASCII letters, digits, `.`, `+` and `-`, beginning with a digit — or `module:sha256:` followed by a module's digest (§7.9), 64 lowercase hexadecimal digits. |
-| `variant` | term or `null` | The decode path within the decoder that ran, where the capability has more than one: `exact` for the path that computes the reconstruction the capability registers (for `wai.image.jpegai`, its exact decoder), or `float` for a path that runs a stage in floating point the registration does not fix (for `wai.image.jpegai`, its float decode) ([determinism-tiers](determinism-tiers.md) §5). `null` for a capability with one path. A sink states the path that ran, as the reference dispatch reports it, not the tier its capability is registered at. |
+| `variant` | term or `null` | The decode path within the decoder that ran, where the capability has more than one: `exact` for the path that computes the reconstruction the capability registers (for `wai.image.jpegai`, its exact decoder), or `float` for a path that runs a stage in floating point the registration does not fix (for `wai.image.jpegai`, its float decode) ([determinism-tiers](determinism-tiers.md) §5). `null` for a capability with one path. A `NeuralIntegerExact` capability's integer path may also be named by the lowering its registration names ([integer-operators](integer-operators.md) §11, its `ops` member's `lowering`), as a `present` entry of [opus-dred](opus-dred.md) §7.2 names `dred-int/1` ([synthesis](synthesis.md) §10.3). The entry states an **exact path** when its `variant` is `null`, `exact`, or, for a capability of the replicate class `neural-int-exact`, that lowering; any other term, `float` among them, names a path that is not exact. A sink states the path that ran, as the reference dispatch reports it, not the tier its capability is registered at. |
 | `pin` | 32 B hex or `null` | The `sha256` of the parameter-set pin the sink verified (SPEC §3.1). |
 | `tier` | string or `null` | That pin's declared tier as SPEC §3.1 reads it: `decode-equivalence`, `entropy-consistency` or `none`. `null` exactly when `pin` is. |
 | `module_status` | string or `null` | For a `module:` decoder, the anchor its grant was confirmed under (§7.9): `pinned` or `fetched`. `null` for a `builtin:` decoder. |
@@ -1089,12 +1089,16 @@ Rules, checked at §3.1 step 4 in this order:
     (`energy_unknown_acquisition`, `energy_legacy_acquisition`,
     `energy_acquisition_evidence`).
 13. A decoder entry whose `capability` has the replicate class `neural-int-exact`
-    (SPEC §7) **MUST** have `variant` `null` or `exact` (`session_variant_not_exact`).
-    A decode of such a capability by any other path is not a decode of it
-    ([determinism-tiers](determinism-tiers.md) §3, *No tier between the two*, rule 1).
-14. A decoder entry whose `variant` is `float` **MUST NOT** have `tier`
-    `decode-equivalence` (`session_tier_exceeds_variant`). Frames a float path decoded
-    are not decode-equivalent, whatever the pin declares (determinism-tiers §3, rule 2).
+    (SPEC §7) **MUST** state an exact path: `variant` `null`, `exact` or the lowering its
+    registration names (`session_variant_not_exact`). A decode of such a capability by
+    any other path is not a decode of it ([determinism-tiers](determinism-tiers.md) §3,
+    *No tier between the two*, rule 1).
+14. A decoder entry that does not state an exact path **MUST NOT** have `tier`
+    `decode-equivalence` (`session_tier_exceeds_variant`). Frames a path that is not
+    exact decoded are not decode-equivalent, whatever the pin declares
+    (determinism-tiers §3, rule 2). Rules 6 and 13 already refuse such an entry for every
+    capability the registry classifies; this rule reaches the others, such as a capability
+    a module decodes.
 
 **Against the outputs** (§3.1 step 6). The context is the list of presentations, each
 `{ "unit", "layers", "decoder", "out_digest" }`: a unit, the layers it was presented
@@ -1269,11 +1273,13 @@ epochs, and bases that cover the same units of the title each have their own. In
    ([staged-delivery](staged-delivery.md) §5), has every decoder entry whose `object` is
    its `subject` at `decode-equivalence` by that section's base step (a capability of
    the replicate class `standard-defined`, or of `neural-int-exact` with `tier`
-   `decode-equivalence`), and no decoder entry whose `object` is its `subject` or a layer
-   object it lists has a `variant` other than `null` or `exact`
-   (`session_stage_hash_not_exact`). A sequence any frame of which was decoded at a
-   weaker tier, or by a path that is not exact, has no clip digest
-   ([determinism-tiers](determinism-tiers.md) §3, *No tier between the two*, rule 3);
+   `decode-equivalence`), and every decoder entry whose `object` is its `subject` or a
+   layer object it lists states an exact path (`session_stage_hash_not_exact`). A
+   sequence any frame of which was decoded at a weaker tier, or by a path that is not
+   exact, has no clip digest ([determinism-tiers](determinism-tiers.md) §3, *No tier
+   between the two*, rule 3). The check is per object: an entry that names the base or
+   such a layer counts whichever units it presented, so the outcome is the same whether
+   or not the verifier holds the outputs;
 6. where the verifier holds the outputs as well, for each claim and each entry of
    `stage_sets` naming it, with `f` that entry's `first_unit` and `n` the units the claim
    counts (`session_stage_disagrees`):
@@ -2301,8 +2307,8 @@ and never folds it in: confirmation changes the outcome of no step here.
 | 4 | `session_stage_sets_invalid` | `stage_sets` out of order, or an entry twice |
 | 4 | `session_counted_invalid` | `counted_discarded` above `counted_primary` and `counted_fallback` together |
 | 4 | `session_classical_invalid` | `classical_source` present without `classical_declared` or absent with it, or not `wai2-table` |
-| 4 | `session_variant_not_exact` | a decoder entry names a capability of the replicate class `neural-int-exact` with a `variant` other than `null` or `exact` |
-| 4 | `session_tier_exceeds_variant` | a decoder entry's `variant` is `float` and its `tier` is `decode-equivalence` |
+| 4 | `session_variant_not_exact` | a decoder entry names a capability of the replicate class `neural-int-exact` and does not state an exact path: a `variant` other than `null`, `exact` or the lowering its registration names |
+| 4 | `session_tier_exceeds_variant` | a decoder entry does not state an exact path, and its `tier` is `decode-equivalence` |
 | 4 | `consumption_objects_unordered` | a `consumption` claim's `objects` are not in strictly ascending order ([ai-usage](ai-usage.md) §6.2) |
 | 4 | `consumption_subject_mismatch` | its `subject` is not the first of its `objects` |
 | 4 | `consumption_usage_unordered` | its `usage` entries are not in strictly ascending order of `vocabulary` |
@@ -2370,7 +2376,7 @@ and never folds it in: confirmation changes the outcome of no step here.
 | 6 | `schedule_content_mismatch` | an entry's `content_hash` is not that of the object receipt the verifier holds at its position |
 | 6 | `schedule_length_mismatch` | an entry's transfer length is not the length of the object's envelope the verifier holds |
 | 6 | `session_stage_disagrees` | a `stage` claim the session names is about a base, or lists an arrived layer object, the session did not receive; counts a unit the session never presented, or more units at a stage than the session presented at it or above; or no claim whose set holds a unit lists an applied layer that reaches a stage the session presented it at |
-| 6 | `session_stage_hash_not_exact` | a `stage` claim the session names states a `stage_hash`, and a decoder entry names its base at a step weaker than `decode-equivalence`, or names its base or a layer it lists with a `variant` other than `null` or `exact` |
+| 6 | `session_stage_hash_not_exact` | a `stage` claim the session names states a `stage_hash`, and a decoder entry names its base at a step weaker than `decode-equivalence`, or names its base or a layer it lists without stating an exact path, whichever units it presented |
 
 ### 3.2. Recovery through a soft binding, and the ancestor rule
 
@@ -2980,7 +2986,7 @@ metered session's figure of zero (`energy_zero_figure`) and every acquisition-cl
 reason; and step 2's enumerations and order. The decode path an entry states has a
 refusal for each of rules 13 and 14 and for a stage hash over a base or layer decoded at
 a weaker tier or by a path that is not exact, beside accepted entries for each of
-`wai.image.jpegai`'s two paths. `claim_verify.py` gives every outcome.
+`wai.image.jpegai`'s two paths and for an integer path named by its registered lowering. `claim_verify.py` gives every outcome.
 `differential.py` generates sessions around every rule — units, leaves and paths off by
 one, switches and missed presentations at and past their caps, a `null` deadline with
 misses, presences that disagree, decoder indices out of range, leaves moved between
diff --git a/wai/extensions/opus-dred.md b/wai/extensions/opus-dred.md
index 0c3527959..3f9276c26 100644
--- a/wai/extensions/opus-dred.md
+++ b/wai/extensions/opus-dred.md
@@ -505,6 +505,12 @@ for each run of lost packets it presents, a `present` entry whose
 - `output` is the digest of that run's reconstruction buffer (§5.3), `origin` and `class` as
   synthesis §10.3 has them (`null` where the player applies no claim).
 
+That `variant` is the lowering `wai.audio.opus.dred.int`'s registration names. A
+`reconstruct-session` claim of the same decode may name its decoder entry's path by it, by
+`exact` or by `null`: each is an exact path of a `NeuralIntegerExact` capability
+(jwp-receipts §2.4.12, `variant`). Any other term is not, and the entry is refused
+(`session_variant_not_exact`).
+
 A verifier checks the log as synthesis §10.3 does (`record_malformed`, `bad_key`, `bad_signature`,
 `player_chain_broken`), then each `present` entry: its decoder, variant and tier are those above,
 its subject is the content hash of the object that recovers a run of the stream, and its `output`
diff --git a/wai/wai-rs/src/claim_receipt.rs b/wai/wai-rs/src/claim_receipt.rs
index 528108552..736b534e4 100644
--- a/wai/wai-rs/src/claim_receipt.rs
+++ b/wai/wai-rs/src/claim_receipt.rs
@@ -364,10 +364,11 @@ pub enum Reason {
     /// A decoder entry whose tier exceeds its capability's class tier.
     SessionTierExceedsClass,
     /// A decoder entry naming a `NeuralIntegerExact` capability with a decode path other
-    /// than its integer one: a `variant` other than `null` or `exact`.
+    /// than its integer one: a `variant` other than `null`, `exact` or the lowering the
+    /// capability's registration names ([`session::DecoderEntry::exact_path`]).
     SessionVariantNotExact,
-    /// A decoder entry whose `tier` exceeds what the path that ran holds: a `float` path
-    /// stating `decode-equivalence`.
+    /// A decoder entry whose `tier` exceeds what the path that ran holds: a path that is
+    /// not exact, `float` among them, stating `decode-equivalence`.
     SessionTierExceedsVariant,
     /// A switch between equal paths or for an unlisted reason, switches out of order,
     /// more than 1024 written or more than their total.
@@ -4260,7 +4261,7 @@ pub(crate) fn rfc3339(s: &str) -> bool {
 /// fails instead of passing on nothing. (A function rather than a public constant so
 /// cbindgen does not export it into the C header.)
 pub fn corpus_vector_count() -> usize {
-    647
+    652
 }
 
 /// The conformance corpus (`claim-conformance/vectors.json`), regenerated. Ed25519
diff --git a/wai/wai-rs/src/claim_receipt/corpus/session.rs b/wai/wai-rs/src/claim_receipt/corpus/session.rs
index 83edf4bac..94fc8d9cb 100644
--- a/wai/wai-rs/src/claim_receipt/corpus/session.rs
+++ b/wai/wai-rs/src/claim_receipt/corpus/session.rs
@@ -2812,11 +2812,24 @@ pub(super) fn decode_path_vectors() -> Vec<V> {
         ),
         reject(
             "session_variant_not_exact_unpinned",
-            "the same with no pin, and a path named by a term this revision does not register: only null or exact names the integer path",
+            "the same with no pin, and a path named by a term this revision does not register: only null, exact or the capability's registered lowering names the integer path",
             picture_bad(one_picture("wai.neural.int_hyper", BUILTIN, None, Some("fp16"), None)),
             4,
             Reason::SessionVariantNotExact,
         ),
+        claim(
+            "session_variant_lowering_accept",
+            "a NeuralIntegerExact capability whose entry names its integer path by the lowering its registration names, as opus-dred 7.2's present entry names wai.audio.opus.dred.int's, dred-int/1, under a pin of decode-equivalence",
+            sealed_about(picture(), SessionClaim { media: "audio".into(), ..one_picture("wai.audio.opus.dred.int", BUILTIN, None, Some("dred-int/1"), Some("decode-equivalence")) }).to_json(),
+            Ok(()),
+        ),
+        reject(
+            "session_variant_lowering_of_another_capability",
+            "wai.neural.int_hyper named with dred-int/1, the lowering another capability registers: not its integer path",
+            picture_bad(one_picture("wai.neural.int_hyper", BUILTIN, None, Some("dred-int/1"), Some("decode-equivalence"))),
+            4,
+            Reason::SessionVariantNotExact,
+        ),
         reject(
             "session_tier_exceeds_variant",
             "a capability the registry does not classify, decoded by a module on its float path and stated under a pin of decode-equivalence: a float path's frames are not decode-equivalent, whatever the pin declares (determinism-tiers 3, rule 2)",
@@ -2824,6 +2837,13 @@ pub(super) fn decode_path_vectors() -> Vec<V> {
             4,
             Reason::SessionTierExceedsVariant,
         ),
+        reject(
+            "session_tier_exceeds_variant_unregistered_term",
+            "the same capability on a path named by a term this revision does not register, fp16, under a pin of decode-equivalence: for a capability the registry does not classify, only null or exact names an exact path",
+            picture_bad(one_picture("org.example.picture", &module, Some("pinned"), Some("fp16"), Some("decode-equivalence"))),
+            4,
+            Reason::SessionTierExceedsVariant,
+        ),
         reject(
             "session_tier_exceeds_class_before_variant",
             "wai.image.jpegai's float path under a pin of decode-equivalence: the class rule, rule 6, refuses it first",
@@ -2831,6 +2851,16 @@ pub(super) fn decode_path_vectors() -> Vec<V> {
             4,
             Reason::SessionTierExceedsClass,
         ),
+        reject(
+            "session_tier_exceeds_class_inter_frame",
+            "an inter-frame capability, wai.neural.dcvc_rt, under a pin of decode-equivalence: its registry class, neural-float, is the class of its weakest frame type, and rule 6 refuses a tier above it whatever path the entry states (determinism-tiers 3, rule 2)",
+            seal_bad(&viewer(), SINK_ID, SubjectType::Object, picture(), 0, ClaimBody::ReconstructSession(SessionClaim {
+                media: "video".into(),
+                ..one_picture("wai.neural.dcvc_rt", BUILTIN, None, None, Some("decode-equivalence"))
+            })),
+            4,
+            Reason::SessionTierExceedsClass,
+        ),
     ];
 
     // ── step 6: a stage hash, the clip digest of a set's units ──
@@ -2849,7 +2879,30 @@ pub(super) fn decode_path_vectors() -> Vec<V> {
     v.push(held(
         reject(
             "session_stage_hash_not_exact",
-            "the same stage claim, while the base's own payload was also decoded through its fallback, a capability held to a tolerance: a clip digest over a sequence some of whose frames were decoded at a weaker tier (determinism-tiers 3, rule 3)",
+            "the same stage claim, while the base's unit 2, inside the set, was decoded through the base's fallback, wai.video.av1, a capability held to a tolerance, by an entry of its own naming the base: a clip digest over a sequence one of whose frames was decoded at a weaker tier (determinism-tiers 3, rule 3)",
+            sealed_or_bad({
+                let mut ps = model_presentations();
+                ps[2] = pres(2, 1, 7);
+                let mut s = over(named(&de_stage), &ps);
+                s.decoders[0].leaves = 3;
+                s.decoders.push(entry_decoder("wai.video.av1", SessionPath::Fallback, false, None, entry(), 0, 1));
+                s.switches = vec![
+                    Switch { unit: 2, from: SessionPath::Primary, to: SessionPath::Fallback, reason: "deadline-threshold".into() },
+                    Switch { unit: 3, from: SessionPath::Fallback, to: SessionPath::Primary, reason: "recovered".into() },
+                    Switch { unit: 5, from: SessionPath::Primary, to: SessionPath::Fallback, reason: "set-incomplete".into() },
+                ];
+                s.switches_total = 3;
+                s
+            }),
+            6,
+            Reason::SessionStageHashNotExact,
+        ),
+        &de_stage,
+    ));
+    v.push(held(
+        reject(
+            "session_stage_hash_not_exact_outside_the_set",
+            "the same stage claim, while the fallback entry, wai.video.av1, names the base and presents only unit 5, outside the set: refused all the same, since the check is per object, whichever units an entry presented (determinism-tiers 3, rule 3)",
             sealed_or_bad({
                 let mut s = named(&de_stage);
                 s.decoders[2].object = entry();
diff --git a/wai/wai-rs/src/claim_receipt/session.rs b/wai/wai-rs/src/claim_receipt/session.rs
index 1e8ba7f65..9b6b2c3c1 100644
--- a/wai/wai-rs/src/claim_receipt/session.rs
+++ b/wai/wai-rs/src/claim_receipt/session.rs
@@ -36,7 +36,8 @@ pub const BUILTIN_DECODER: &str = "builtin:";
 pub const MODULE_DECODER: &str = "module:sha256:";
 /// The `variant` of a decoder entry whose path computes the reconstruction its capability
 /// registers (determinism-tiers §5): `wai.image.jpegai`'s exact decoder, or a
-/// `NeuralIntegerExact` capability's integer path, which may also be stated as `null`.
+/// `NeuralIntegerExact` capability's integer path, which may also be stated as `null` or
+/// by the lowering the capability's registration names ([`DecoderEntry::exact_path`]).
 pub const SESSION_VARIANT_EXACT: &str = "exact";
 /// The `variant` of a decoder entry whose path runs a stage in floating point that the
 /// capability's registration does not fix: `wai.image.jpegai`'s float decode.
@@ -142,7 +143,9 @@ pub struct DecoderEntry {
     /// `builtin:<crate>@<version>` or `module:sha256:<hex>`.
     pub decoder: String,
     /// The decode path that ran: [`SESSION_VARIANT_EXACT`] or [`SESSION_VARIANT_FLOAT`]
-    /// (`neural::DecodePath::variant`), or `None` for a capability with one path.
+    /// (`neural::DecodePath::variant`), or `None` for a capability with one path. A
+    /// `NeuralIntegerExact` capability's integer path may also be named by the lowering its
+    /// registration names, as opus-dred §7.2 names `dred-int/1`.
     pub variant: Option<String>,
     /// The SHA-256 of the pin the sink verified.
     pub pin: Option<[u8; 32]>,
@@ -157,6 +160,17 @@ pub struct DecoderEntry {
 type EntryKey<'a> = (&'a str, Option<u64>, [u8; 32], u64, SessionPath, &'a str, Option<&'a str>, Option<[u8; 32]>, Option<&'a str>, Option<&'a str>);
 
 impl DecoderEntry {
+    /// Whether the entry states an exact path (§2.4.12, `variant`): `null`, `exact`, or, for a
+    /// capability of the registry class `NeuralIntegerExact`, the lowering of the operator set
+    /// its registration names (its `ops` member's `lowering`, such as `dred-int/1`). Any other
+    /// term, `float` among them, names a path that is not exact.
+    pub fn exact_path(&self) -> bool {
+        let Some(v) = self.variant.as_deref() else { return true };
+        v == SESSION_VARIANT_EXACT
+            || (crate::codecs::capability_replicate_class(&self.capability) == Some(crate::codecs::ReplicateClass::NeuralIntegerExact)
+                && crate::codecs::capability_ops(&self.capability).and_then(|o| o.lowering) == Some(v))
+    }
+
     /// Everything but `leaves`: two entries with one key are one entry.
     fn key(&self) -> EntryKey<'_> {
         (
@@ -732,13 +746,13 @@ impl SessionClaim {
         // decode by any other path is not a decode of it (determinism-tiers §3, *No tier
         // between the two*, rule 1).
         let integer = |cap: &str| crate::codecs::capability_replicate_class(cap) == Some(crate::codecs::ReplicateClass::NeuralIntegerExact);
-        if self.decoders.iter().any(|d| d.variant.as_deref().is_some_and(|v| v != SESSION_VARIANT_EXACT) && integer(&d.capability)) {
+        if self.decoders.iter().any(|d| integer(&d.capability) && !d.exact_path()) {
             return Err(Reason::SessionVariantNotExact);
         }
-        // Rule 14: no pin's tier above what the path that ran holds; frames a float path
-        // decoded are not decode-equivalent, whatever the pin declares (rule 2 there).
+        // Rule 14: no pin's tier above what the path that ran holds; frames a path that is not
+        // exact decoded are not decode-equivalent, whatever the pin declares (rule 2 there).
         let de = Some(crate::codecs::DeclaredTier::DecodeEquivalence);
-        if self.decoders.iter().any(|d| d.variant.as_deref() == Some(SESSION_VARIANT_FLOAT) && d.tier.as_deref().and_then(crate::codecs::DeclaredTier::from_token) == de) {
+        if self.decoders.iter().any(|d| !d.exact_path() && d.tier.as_deref().and_then(crate::codecs::DeclaredTier::from_token) == de) {
             return Err(Reason::SessionTierExceedsVariant);
         }
         Ok(())
@@ -1355,6 +1369,7 @@ impl ClaimReceipt {
     /// (`session_first_unit_mismatch`). A claim that states a stage hash has every entry
     /// naming its base at `decode-equivalence` by staged-delivery §5's base step, and no
     /// entry naming its base or a layer it lists with a path that is not exact
+    /// ([`DecoderEntry::exact_path`]), whichever units the entry presented
     /// (`session_stage_hash_not_exact`). Then, with the
     /// presentations (`session_stage_disagrees`): for each claim and each entry naming it,
     /// every unit of its set is presented, and for each stage `s` from 1 up the units of
@@ -1405,12 +1420,13 @@ impl ClaimReceipt {
         // entry naming the set's base decodes it at decode-equivalence by the base step of
         // staged-delivery §5, and no entry naming the base or a layer the claim lists ran a
         // path that is not exact (determinism-tiers §3, *No tier between the two*, rule 3).
+        // The check is per object, whichever units an entry presented, so that its outcome
+        // does not depend on whether the verifier holds the outputs.
         let base_exact = |d: &DecoderEntry| match crate::codecs::capability_replicate_class(&d.capability) {
             Some(crate::codecs::ReplicateClass::StandardDefined) => true,
             Some(crate::codecs::ReplicateClass::NeuralIntegerExact) => d.tier.as_deref() == Some("decode-equivalence"),
             _ => false,
         };
-        let path_exact = |d: &DecoderEntry| d.variant.as_deref().is_none_or(|v| v == SESSION_VARIANT_EXACT);
         for (c, _) in &claims {
             let ClaimBody::Stage(st) = &c.body else { continue };
             if st.stage_hash.is_none() {
@@ -1419,7 +1435,7 @@ impl ClaimReceipt {
             let mixed = s.decoders.iter().any(|d| {
                 let base = d.object == c.subject;
                 let layer = st.objects.iter().any(|o| o.object == d.object);
-                (base && !base_exact(d)) || ((base || layer) && !path_exact(d))
+                (base && !base_exact(d)) || ((base || layer) && !d.exact_path())
             });
             if mixed {
                 return Err(Rejection::at(6, Reason::SessionStageHashNotExact));
-- 
2.54.0 (Apple Git-157)


From 2060e7de7ac49f3c29209e0a55739d15faf9fb79 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:53:32 -0500
Subject: [PATCH 16/29] neural: the binding check reads the descriptor, the C
 ABI reports the path, and CI runs the exact decoder
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

check_reconstruction_binding compared only the hash, so a binding of another
decoderId, format or size than the exact decoder computes was reported as
Differs. reconstruction-binding §2 forbids reporting a binding as failed when
the verifier decoded some other way than its descriptor says. Such a binding
is now BindingCheck::OtherReconstruction, with the path that ran; for the
exact decoder that is any decoderId but 2.

wai_neural_decode_ex dropped the path and the capability, so a native sink
could not state its decoder entry's variant. wai_decoded_t gains `path`
(WAI_DECODE_PATH_SOLE, _EXACT or _FLOAT) and `fallback`, 1 when a WAI1
envelope decoded through its fallback. include/wai.h is regenerated.

The neural gates counted the two JPEG-AI cases' skips as passes, so CI ran
only the one-path case. That step now selects its two committed-input cases by
name, one of them a binding over int_mlicpp checked under its descriptor. A new
job in the 24.04 image fetches the pinned jpegai/v1 bundle and runs the exact
decoder's case with WAI_REQUIRE_SOTA_VECTORS=1, requiring it by name and
refusing a skip. The float decode needs the reference-software assets, which
no job holds, so its case runs locally only; the test header and
determinism-tiers §5 say so. The C ABI step runs the path and fallback test.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-neural-gates.yml   | 86 ++++++++++++++++++++++--
 wai/extensions/determinism-tiers.md      |  7 +-
 wai/extensions/reconstruction-binding.md |  5 +-
 wai/wai-rs/include/wai.h                 | 32 ++++++++-
 wai/wai-rs/src/ffi_neural.rs             | 44 +++++++++---
 wai/wai-rs/src/neural/mod.rs             | 29 ++++++--
 wai/wai-rs/tests/ffi_neural_pin.rs       | 42 ++++++++++--
 wai/wai-rs/tests/jpegai_decode_path.rs   | 51 ++++++++++++--
 8 files changed, 258 insertions(+), 38 deletions(-)

diff --git a/.github/workflows/wai-neural-gates.yml b/.github/workflows/wai-neural-gates.yml
index 56742d8d7..2c5b440cc 100644
--- a/.github/workflows/wai-neural-gates.yml
+++ b/.github/workflows/wai-neural-gates.yml
@@ -28,6 +28,14 @@ name: WAI neural sink gates
 # registry given a key log refuses the committed int_mlicpp set once the log revokes
 # its digest, and when the sink holds no fresh status, and decodes it while a grant
 # holds it.
+#
+# The dispatch reports the decode path that ran (determinism-tiers §5). The tests
+# with one path run on committed inputs in the first job. A second job fetches the
+# pinned jpegai/v1 bundle and runs wai.image.jpegai's exact decoder through the
+# dispatch, with WAI_REQUIRE_SOTA_VECTORS=1, so a missing input is a failure. The
+# float decode needs the JPEG-AI reference-software assets, which are not in the
+# repository, so its test runs locally only, as tests/jpegai_live.rs does
+# (wai-jpegai-e2e.yml).
 
 # One run per branch at a time. A push supersedes the run before it.
 #
@@ -67,6 +75,15 @@ on:
       - 'wai/wai-rs/tests/ffi_neural_pin.rs'
       - 'wai/wai-rs/tests/neural_cost_limit.rs'
       - 'wai/wai-rs/tests/jpegai_decode_path.rs'
+      # The exact JPEG-AI decoder the second job runs through the dispatch, and the
+      # fetch of its pinned inputs.
+      - 'wai/wai-rs/src/jpegai_*.rs'
+      - 'wai/wai-rs/src/float_repro.rs'
+      - 'wai/wai-rs/src/int_kernels.rs'
+      - 'wai/tools/wai_jpegai_fetch_e2e_inputs.sh'
+      - 'wai/tools/jpegai-e2e-v1.sha256'
+      - 'wai/wai-rs/tests/vectors/jpegai/**'
+      - 'wai/wai-rs/tests/support/**'
       - 'wai/wai-rs/src/session_recorder.rs'
       - 'wai/wai-rs/src/reconstruction_binding.rs'
       - 'wai/wai-rs/src/claim_receipt/**'
@@ -193,17 +210,20 @@ jobs:
           grep -q "test result: ok. 3 passed; 0 failed" "$RUNNER_TEMP/neural-cost.log"
       # The dispatch reports the capability it decoded through and the path that ran
       # (extensions/determinism-tiers.md §5): a capability with one path says so, through
-      # its own capability and through its fallback. The JPEG-AI float and exact paths need
-      # inputs no job here provides (the local asset directory and the jpegai/v1 bundle),
-      # so in this step they skip loudly; the count still requires all three to be present.
+      # its own capability and through its fallback, and a reconstruction binding over a
+      # decode-equivalence capability is checked under its descriptor. Committed inputs
+      # only. The exact JPEG-AI decoder runs in the next job; the float decode runs locally
+      # only (see the header).
       - name: The neural dispatch reports the decode path that ran
         working-directory: wai/wai-rs
         shell: bash
         run: |
           set -euo pipefail
-          cargo test --no-default-features --features neural,receipt_claims,video_int --test jpegai_decode_path \
+          cargo test --no-default-features --features neural,receipt_claims,video_int --test jpegai_decode_path -- \
+            --exact a_capability_with_one_path_reports_it_and_the_capability_it_decoded_through \
+            a_binding_over_a_decode_equivalence_capability_is_checked_under_its_descriptor \
             2>&1 | tee "$RUNNER_TEMP/neural-path.log"
-          grep -q "test result: ok. 3 passed; 0 failed" "$RUNNER_TEMP/neural-path.log"
+          grep -q "test result: ok. 2 passed; 0 failed" "$RUNNER_TEMP/neural-path.log"
       - name: A key log refuses a revoked pinned set at the neural sink (SPEC §3.1 check 5)
         working-directory: wai/wai-rs
         shell: bash
@@ -240,5 +260,59 @@ jobs:
           cargo test --no-default-features --features neural,ffi --test ffi_neural_pin -- \
             --exact tampered_model_is_wai_err_prior strict_replicate_is_wai_err_unsound_replicate \
             wai2_decode_ex_picks_the_held_pin failed_decode_ex_clears_out \
+            decode_ex_reports_the_path_and_the_fallback \
             2>&1 | tee "$RUNNER_TEMP/neural-ffi.log"
-          grep -q "test result: ok. 4 passed; 0 failed" "$RUNNER_TEMP/neural-ffi.log"
+          grep -q "test result: ok. 5 passed; 0 failed" "$RUNNER_TEMP/neural-ffi.log"
+
+  # wai.image.jpegai's exact decoder through the dispatch: the result says
+  # DecodePath::Exact, a session claim recorded from it states `variant` `exact` and
+  # verifies, and the binding check holds over its pixels, differs over a changed one,
+  # and reports a binding of another decoderId as another reconstruction
+  # (tests/jpegai_decode_path.rs). Its inputs are the pinned jpegai/v1 bundle, fetched
+  # and checked against wai/tools/jpegai-e2e-v1.sha256 as wai-jpegai-e2e.yml does; an
+  # outage of the models root fails the job. The `neural` feature links the prebuilt
+  # ONNX Runtime, so this job runs in the 24.04 image, as the job above does.
+  jpegai-decode-path:
+    needs: relevant
+    if: needs.relevant.outputs.run == 'true'
+    name: the exact JPEG-AI decoder reports its path (pinned inputs, required)
+    runs-on: linux
+    timeout-minutes: 60
+    container:
+      image: catthehacker/ubuntu:act-24.04
+    env:
+      WAI_REQUIRE_SOTA_VECTORS: '1'
+    steps:
+      - uses: actions/checkout@v5
+      # Bootstrap rustup: the docker runner image ships no Rust.
+      - name: Install Rust 1.98.0
+        run: |
+          export PATH="$HOME/.cargo/bin:$PATH"
+          command -v rustup >/dev/null 2>&1 || \
+            curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y --profile minimal --default-toolchain 1.98.0
+          rustup toolchain install 1.98.0 --profile minimal
+          rustup default 1.98.0
+          echo "$HOME/.cargo/bin" >> "$GITHUB_PATH"
+      - name: Fetch the JPEG-AI inputs and check each against the pinned manifest
+        working-directory: wai
+        shell: bash
+        run: tools/wai_jpegai_fetch_e2e_inputs.sh "$RUNNER_TEMP/jpegai_e2e"
+      - name: The exact decoder reports its path, in the result, the claim and the binding check (missing input = failure)
+        working-directory: wai/wai-rs
+        shell: bash
+        run: |
+          set -euo pipefail
+          [ "${WAI_REQUIRE_SOTA_VECTORS:-}" = 1 ] \
+            || { echo "::error::WAI_REQUIRE_SOTA_VECTORS must be 1 in this job"; exit 1; }
+          export WAI_JPEGAI_WEIGHTS_DIR="$RUNNER_TEMP/jpegai_e2e"
+          export WAI_JPEGAI_NORM="$RUNNER_TEMP/jpegai_e2e/norm.json"
+          cargo test --release --no-default-features --features neural,receipt_claims,video_int --test jpegai_decode_path -- \
+            --exact the_jpegai_exact_decoder_reports_itself --show-output 2>&1 | tee "$RUNNER_TEMP/jpegai-path.log"
+          grep -q "test result: ok. 1 passed; 0 failed" "$RUNNER_TEMP/jpegai-path.log"
+          grep -q "^test the_jpegai_exact_decoder_reports_itself \.\.\. ok$" "$RUNNER_TEMP/jpegai-path.log"
+          # A skip also prints `test X ... ok`; every skip writes a banner with
+          # "SKIPPED" and "NOT VERIFIED" (tests/support/require_vectors.rs).
+          if grep -nE 'SKIPPED|NOT VERIFIED' "$RUNNER_TEMP/jpegai-path.log"; then
+            echo "::error::the exact JPEG-AI decode skipped instead of checking its input"
+            exit 1
+          fi
diff --git a/wai/extensions/determinism-tiers.md b/wai/extensions/determinism-tiers.md
index 6405d4ffb..eb02575e5 100644
--- a/wai/extensions/determinism-tiers.md
+++ b/wai/extensions/determinism-tiers.md
@@ -541,9 +541,10 @@ path in a `reconstruct-session` claim as its decoder entry's `variant`, `exact`
 `float` ([jwp-receipts](jwp-receipts.md) §2.4.12), and the reference check of a
 reconstruction binding reads it (`neural::check_reconstruction_binding`,
 [reconstruction-binding](reconstruction-binding.md) §2).
-`wai-rs/tests/jpegai_decode_path.rs` runs the float path and requires the result,
-the claim and the binding check to say so; it runs only where the float path's
-model bundle is supplied.
+`wai-rs/tests/jpegai_decode_path.rs` runs each path and requires the result, the
+claim and the binding check to say which ran. CI runs the exact decoder over the
+pinned bundle; the float path's model bundle is not in the repository, so its case
+runs locally only, on a host that holds it.
 
 On both paths the stages the capability's claim covers, the scale path and the two
 entropy decodes, run through the integer operator set, by the lowering `jpegai-int/1`
diff --git a/wai/extensions/reconstruction-binding.md b/wai/extensions/reconstruction-binding.md
index 8757b27a8..fc9f94f9d 100644
--- a/wai/extensions/reconstruction-binding.md
+++ b/wai/extensions/reconstruction-binding.md
@@ -107,7 +107,10 @@ Conformance criterion:
   ([determinism-tiers](determinism-tiers.md) §5), and the reference check
   (`neural::check_reconstruction_binding`) reads it: over the exact decoder it
   reports whether the hash holds, and over the float fallback it reports the
-  binding as not checked, with the path, never as failed.
+  binding as not checked, with the path, never as failed. It reads the descriptor
+  too: a binding whose `decoderId` is not 2, whose `format` is not `RGB8`, or whose
+  `width` and `height` are not the decoded picture's names a reconstruction the
+  exact decoder does not compute, and is reported as such, never as failed.
 
 ## 3. C2PA / JPEG Trust carriage
 
diff --git a/wai/wai-rs/include/wai.h b/wai/wai-rs/include/wai.h
index 81e443baa..311d18605 100644
--- a/wai/wai-rs/include/wai.h
+++ b/wai/wai-rs/include/wai.h
@@ -228,6 +228,24 @@
  */
 #define WAI_DECODE_STRICT 1
 
+/**
+ * The decode path of a [`WaiDecoded`]: the capability has one decode path
+ * (`neural::DecodePath::Sole`). A session claim's decoder entry states `variant` `null`.
+ */
+#define WAI_DECODE_PATH_SOLE 1
+
+/**
+ * The decode path of a [`WaiDecoded`]: `wai.image.jpegai`'s exact decoder
+ * (`neural::DecodePath::Exact`). A session claim's decoder entry states `variant` `exact`.
+ */
+#define WAI_DECODE_PATH_EXACT 2
+
+/**
+ * The decode path of a [`WaiDecoded`]: `wai.image.jpegai`'s float decode
+ * (`neural::DecodePath::Float`). A session claim's decoder entry states `variant` `float`.
+ */
+#define WAI_DECODE_PATH_FLOAT 3
+
 #define SECTION_CONTRACT 1
 
 #define SECTION_GRAPH 2
@@ -606,10 +624,16 @@ typedef struct wai_int_declared_t {
  * `WAI_DECODED_IMAGE` (`data` holds RGB8, `width` and `height` set) or
  * `WAI_DECODED_VIDEO` (`data` holds the frames' RGB8 concatenated, `width`,
  * `height`, `n_frames` and `fps_x_1000` set). `index` is the `WAI2` rendition
- * picked (0 for a `WAI1` envelope). Free `data` with `wai_buffer_free`.
+ * picked (0 for a `WAI1` envelope). `path` is the decode path that ran
+ * (`WAI_DECODE_PATH_SOLE`, `WAI_DECODE_PATH_EXACT` or `WAI_DECODE_PATH_FLOAT`,
+ * extensions/determinism-tiers.md §5), which a receipt of the decode states as its
+ * decoder entry's `variant` (extensions/jwp-receipts.md §2.4.12). `fallback` is 1 when
+ * a `WAI1` envelope decoded through its manifest's `fallback` capability and 0 when
+ * through its `capability`; for a `WAI2` envelope `index` names the rendition, and with
+ * it the capability, and `fallback` is 0. Free `data` with `wai_buffer_free`.
  * `wai_neural_decode_ex` writes the whole struct on every call that passes a
- * non-null `out`: on failure `kind` is 0 and `data` is empty, so `data` is always
- * safe to free.
+ * non-null `out`: on failure `kind`, `path` and `fallback` are 0 and `data` is empty,
+ * so `data` is always safe to free.
  */
 typedef struct wai_decoded_t {
   uint32_t kind;
@@ -620,6 +644,8 @@ typedef struct wai_decoded_t {
   uint32_t sample_rate;
   uint32_t n_frames;
   uint32_t fps_x_1000;
+  uint32_t path;
+  uint32_t fallback;
 } wai_decoded_t;
 
 
diff --git a/wai/wai-rs/src/ffi_neural.rs b/wai/wai-rs/src/ffi_neural.rs
index 47ba5e828..c9c05648c 100644
--- a/wai/wai-rs/src/ffi_neural.rs
+++ b/wai/wai-rs/src/ffi_neural.rs
@@ -34,7 +34,7 @@ use std::slice;
 use crate::container::{Wai, WaiMulti};
 use crate::ffi::{set_err, WaiBuffer, WAI_ERR_CODEC, WAI_ERR_CONTAINER, WAI_ERR_INVALID_ARG,
                  WAI_ERR_NO_RENDITION, WAI_ERR_PANIC, WAI_ERR_PRIOR, WAI_ERR_UNSOUND_REPLICATE, WAI_OK};
-use crate::neural::{decode_envelope, decode_envelope_strict, decode_multi, decode_multi_strict, decodes_to_int,
+use crate::neural::{decode_envelope, decode_envelope_strict, decode_multi, decode_multi_strict, decodes_to_int, DecodePath,
                     DecodeError, Decoded, ModelRegistry};
 
 unsafe fn input_envelope<'a>(bytes: *const u8, len: usize) -> Option<&'a [u8]> {
@@ -192,16 +192,32 @@ pub const WAI_DECODED_VIDEO: u32 = 3;
 /// `wai_neural_decode_ex` flag: apply SPEC §7's replicate rule (strict).
 pub const WAI_DECODE_STRICT: u32 = 1;
 
+/// The decode path of a [`WaiDecoded`]: the capability has one decode path
+/// (`neural::DecodePath::Sole`). A session claim's decoder entry states `variant` `null`.
+pub const WAI_DECODE_PATH_SOLE: u32 = 1;
+/// The decode path of a [`WaiDecoded`]: `wai.image.jpegai`'s exact decoder
+/// (`neural::DecodePath::Exact`). A session claim's decoder entry states `variant` `exact`.
+pub const WAI_DECODE_PATH_EXACT: u32 = 2;
+/// The decode path of a [`WaiDecoded`]: `wai.image.jpegai`'s float decode
+/// (`neural::DecodePath::Float`). A session claim's decoder entry states `variant` `float`.
+pub const WAI_DECODE_PATH_FLOAT: u32 = 3;
+
 /// What `wai_neural_decode_ex` decoded.
 ///
 /// `kind` is `WAI_DECODED_AUDIO` (`data` holds `f32` samples, `sample_rate` set),
 /// `WAI_DECODED_IMAGE` (`data` holds RGB8, `width` and `height` set) or
 /// `WAI_DECODED_VIDEO` (`data` holds the frames' RGB8 concatenated, `width`,
 /// `height`, `n_frames` and `fps_x_1000` set). `index` is the `WAI2` rendition
-/// picked (0 for a `WAI1` envelope). Free `data` with `wai_buffer_free`.
+/// picked (0 for a `WAI1` envelope). `path` is the decode path that ran
+/// (`WAI_DECODE_PATH_SOLE`, `WAI_DECODE_PATH_EXACT` or `WAI_DECODE_PATH_FLOAT`,
+/// extensions/determinism-tiers.md §5), which a receipt of the decode states as its
+/// decoder entry's `variant` (extensions/jwp-receipts.md §2.4.12). `fallback` is 1 when
+/// a `WAI1` envelope decoded through its manifest's `fallback` capability and 0 when
+/// through its `capability`; for a `WAI2` envelope `index` names the rendition, and with
+/// it the capability, and `fallback` is 0. Free `data` with `wai_buffer_free`.
 /// `wai_neural_decode_ex` writes the whole struct on every call that passes a
-/// non-null `out`: on failure `kind` is 0 and `data` is empty, so `data` is always
-/// safe to free.
+/// non-null `out`: on failure `kind`, `path` and `fallback` are 0 and `data` is empty,
+/// so `data` is always safe to free.
 #[repr(C)]
 pub struct WaiDecoded {
     pub kind: u32,
@@ -212,6 +228,8 @@ pub struct WaiDecoded {
     pub sample_rate: u32,
     pub n_frames: u32,
     pub fps_x_1000: u32,
+    pub path: u32,
+    pub fallback: u32,
 }
 
 impl WaiDecoded {
@@ -219,7 +237,7 @@ impl WaiDecoded {
     fn empty() -> Self {
         WaiDecoded {
             kind: 0, index: 0, data: WaiBuffer::empty(),
-            width: 0, height: 0, sample_rate: 0, n_frames: 0, fps_x_1000: 0,
+            width: 0, height: 0, sample_rate: 0, n_frames: 0, fps_x_1000: 0, path: 0, fallback: 0,
         }
     }
 }
@@ -285,20 +303,26 @@ pub unsafe extern "C" fn wai_neural_decode_ex(
             let (code, msg) = decode_status(e);
             if left.is_empty() { (code, msg) } else { (code, format!("{msg}; {}", no_kind(&left).1)) }
         };
-        let (index, decoded) = if version == 1 {
+        let (index, fallback, d) = if version == 1 {
             let env = Wai::from_bytes(bytes).map_err(|e| (WAI_ERR_CONTAINER, format!("envelope parse: {e}")))?;
             let d = if strict { decode_envelope_strict(&env, &reg) } else { decode_envelope(&env, &reg) };
             let named = [Some(env.capability()), env.fallback()].into_iter().flatten().collect();
-            (0, d.map_err(|e| explain(named, e))?.output)
+            let d = d.map_err(|e| explain(named, e))?;
+            (0, d.capability != env.capability(), d)
         } else {
             let m = WaiMulti::from_bytes(bytes).map_err(|e| (WAI_ERR_CONTAINER, format!("envelope parse: {e}")))?;
             let r = if strict { decode_multi_strict(&m, &reg) } else { decode_multi(&m, &reg) };
             let named = m.manifest.renditions.iter().map(|r| r.capability.as_str()).collect();
             let (i, d) = r.map_err(|e| explain(named, e))?;
-            (i, d.output)
+            (i, false, d)
         };
-        let mut o = WaiDecoded { index: index as u32, ..WaiDecoded::empty() };
-        match decoded {
+        let path = match d.path {
+            DecodePath::Exact => WAI_DECODE_PATH_EXACT,
+            DecodePath::Float => WAI_DECODE_PATH_FLOAT,
+            DecodePath::Sole => WAI_DECODE_PATH_SOLE,
+        };
+        let mut o = WaiDecoded { index: index as u32, path, fallback: u32::from(fallback), ..WaiDecoded::empty() };
+        match d.output {
             Decoded::Audio(a) => {
                 o.kind = WAI_DECODED_AUDIO;
                 o.sample_rate = a.sample_rate;
diff --git a/wai/wai-rs/src/neural/mod.rs b/wai/wai-rs/src/neural/mod.rs
index 1ae7c97b5..cdda7c666 100644
--- a/wai/wai-rs/src/neural/mod.rs
+++ b/wai/wai-rs/src/neural/mod.rs
@@ -1052,6 +1052,11 @@ pub fn decode_multi_strict(m: &WaiMulti, registry: &ModelRegistry) -> Result<(us
     decode_multi(m, registry)
 }
 
+/// The `decoderId` of the reconstruction `wai.image.jpegai`'s exact decoder computes: its
+/// high-operation-point synthesis (extensions/reconstruction-binding.md §1).
+#[cfg(feature = "provenance")]
+const JPEGAI_EXACT_DECODER_ID: u8 = 2;
+
 /// What checking a reconstruction binding against a decode found
 /// ([`check_reconstruction_binding`], extensions/reconstruction-binding.md §2).
 #[cfg(feature = "provenance")]
@@ -1068,16 +1073,25 @@ pub enum BindingCheck {
     /// `decode-equivalence` nor that capability's exact decoder. The binding was not
     /// checked, and a mismatch would not be evidence that the asset changed.
     NotChecked(DecodePath),
+    /// The path that ran computes a reconstruction, but not the one the binding's
+    /// descriptor names: another `decoderId` (`wai.image.jpegai`'s exact decoder computes
+    /// `decoderId` 2), a `format` other than `RGB8`, or a `width` and `height` other than
+    /// the decoded picture's. The binding was not checked: the verifier decoded some other
+    /// way than the descriptor says (extensions/reconstruction-binding.md §2).
+    OtherReconstruction(DecodePath),
 }
 
 /// Check a reconstruction binding over the decode of a parsed `WAI1` envelope
 /// (extensions/reconstruction-binding.md §2): decode it as [`decode_envelope`] does, then
 /// compare its canonical `RGB8` buffer with the binding only where the path that ran
 /// computes the reconstruction the binding is over: a capability whose registry class
-/// gives `decode-equivalence`, or `wai.image.jpegai`'s exact decoder. Over any other
-/// path the binding is reported as not checked, with that path, never as failed. A
-/// decode that is not a picture is [`DecodeError::InvalidPayload`]: the binding hashes a
-/// picture.
+/// gives `decode-equivalence`, or `wai.image.jpegai`'s exact decoder, under a descriptor
+/// that names what that path computed. Over any other path the binding is reported as not
+/// checked, with that path, never as failed; under a descriptor that names another
+/// reconstruction (for the exact decoder, a `decoderId` other than 2; for either, a format
+/// other than `RGB8` or another picture size) it is reported as
+/// [`BindingCheck::OtherReconstruction`], never as failed. A decode that is not a picture
+/// is [`DecodeError::InvalidPayload`]: the binding hashes a picture.
 #[cfg(feature = "provenance")]
 pub fn check_reconstruction_binding(
     env: &Wai, registry: &ModelRegistry, binding: &crate::reconstruction_binding::ReconstructionBinding,
@@ -1094,6 +1108,13 @@ pub fn check_reconstruction_binding(
     if !exact {
         return Ok(BindingCheck::NotChecked(d.path));
     }
+    // The descriptor names the reconstruction the hash is over; a verifier compares only
+    // under it. A single-branch capability's descriptor fixes no branch.
+    let r = &binding.descriptor;
+    let branch = d.path != DecodePath::Exact || r.decoder_id == JPEGAI_EXACT_DECODER_ID;
+    if !branch || r.format != "RGB8" || (r.width, r.height) != (img.width, img.height) {
+        return Ok(BindingCheck::OtherReconstruction(d.path));
+    }
     Ok(if binding.verify(&img.rgb) { BindingCheck::Holds } else { BindingCheck::Differs })
 }
 
diff --git a/wai/wai-rs/tests/ffi_neural_pin.rs b/wai/wai-rs/tests/ffi_neural_pin.rs
index 571ae06c0..625e0e515 100644
--- a/wai/wai-rs/tests/ffi_neural_pin.rs
+++ b/wai/wai-rs/tests/ffi_neural_pin.rs
@@ -1,7 +1,8 @@
 //! The neural C ABI reports a refused pin as `WAI_ERR_PRIOR` with its reason first
 //! in `wai_last_error`, an unsound `replicate` under the strict flag as
-//! `WAI_ERR_UNSOUND_REPLICATE`, and decodes a `WAI2` envelope through the
-//! rendition whose pin the registry holds — all on the committed int_mlicpp
+//! `WAI_ERR_UNSOUND_REPLICATE`, decodes a `WAI2` envelope through the
+//! rendition whose pin the registry holds, and reports the decode path that ran and
+//! whether a `WAI1` decoded through its fallback — all on the committed int_mlicpp
 //! vector, with no model to fetch.
 #![cfg(all(feature = "neural", feature = "ffi"))]
 
@@ -26,7 +27,7 @@ fn last_error() -> String {
 fn decoded() -> WaiDecoded {
     WaiDecoded {
         kind: 0, index: 0, data: WaiBuffer { data: ptr::null_mut(), len: 0, cap: 0 },
-        width: 0, height: 0, sample_rate: 0, n_frames: 0, fps_x_1000: 0,
+        width: 0, height: 0, sample_rate: 0, n_frames: 0, fps_x_1000: 0, path: 0, fallback: 0,
     }
 }
 
@@ -64,11 +65,11 @@ fn failed_decode_ex_clears_out() {
     let empty = CString::new("{}").unwrap();
     let mut out = WaiDecoded {
         kind: 7, index: 9, data: WaiBuffer { data: 0x5a5a as *mut u8, len: 5, cap: 5 },
-        width: 1, height: 1, sample_rate: 1, n_frames: 1, fps_x_1000: 1,
+        width: 1, height: 1, sample_rate: 1, n_frames: 1, fps_x_1000: 1, path: 3, fallback: 1,
     };
     let code = unsafe { wai_neural_decode_ex(env.as_ptr(), env.len(), empty.as_ptr(), 0, &mut out) };
     assert_ne!(code, WAI_OK);
-    assert_eq!((out.kind, out.index, out.width, out.height), (0, 0, 0, 0));
+    assert_eq!((out.kind, out.index, out.width, out.height, out.path, out.fallback), (0, 0, 0, 0, 0, 0));
     assert!(out.data.data.is_null() && out.data.len == 0 && out.data.cap == 0);
     unsafe { wai_buffer_free(out.data) };
     // A bad argument after the null check clears it too.
@@ -88,9 +89,38 @@ fn strict_replicate_is_wai_err_unsound_replicate() {
     let code = unsafe { wai_neural_decode_ex(bytes.as_ptr(), bytes.len(), reg.as_ptr(), WAI_DECODE_STRICT, &mut out) };
     assert_eq!(code, WAI_ERR_UNSOUND_REPLICATE);
     assert!(last_error().contains("replicate_unpinned"), "{}", last_error());
-    // Without the strict flag the unpinned envelope decodes.
+    // Without the strict flag the unpinned envelope decodes, through its own capability's
+    // one decode path, and says so.
     assert_eq!(unsafe { wai_neural_decode_ex(bytes.as_ptr(), bytes.len(), reg.as_ptr(), 0, &mut out) }, WAI_OK);
     assert_eq!((out.kind, out.width, out.height), (WAI_DECODED_IMAGE, 64, 64));
+    assert_eq!((out.path, out.fallback), (WAI_DECODE_PATH_SOLE, 0));
+    unsafe { wai_buffer_free(out.data) };
+}
+
+/// `wai_neural_decode_ex` reports the decode path that ran and whether a `WAI1` envelope
+/// decoded through its fallback (extensions/determinism-tiers.md §5), which a native sink
+/// states in its receipt (extensions/jwp-receipts.md §2.4.12, a decoder entry's `variant`
+/// and `path`).
+#[test]
+fn decode_ex_reports_the_path_and_the_fallback() {
+    let mut env = Wai::from_bytes(&std::fs::read(dir().join("stream.wai")).unwrap()).unwrap();
+    env.manifest.conditioning.prior = None;
+    let reg = CString::new(format!(r#"{{"wai.neural.int_mlicpp":["{}"]}}"#, dir().join("model.wmb").display())).unwrap();
+    let own = env.to_bytes().unwrap();
+    let mut out = decoded();
+    assert_eq!(unsafe { wai_neural_decode_ex(own.as_ptr(), own.len(), reg.as_ptr(), 0, &mut out) }, WAI_OK);
+    assert_eq!((out.kind, out.path, out.fallback), (WAI_DECODED_IMAGE, WAI_DECODE_PATH_SOLE, 0));
+    let want = fnv_of(&out);
+    unsafe { wai_buffer_free(out.data) };
+    // The same payload under a capability the registry does not hold, with int_mlicpp as
+    // its fallback: the same picture, through the fallback, and the result says so.
+    env.manifest.model_requirement.capability = "wai.neural.int_hyper".into();
+    env.manifest.model_requirement.fallback = Some("wai.neural.int_mlicpp".into());
+    let via = env.to_bytes().unwrap();
+    let mut out = decoded();
+    assert_eq!(unsafe { wai_neural_decode_ex(via.as_ptr(), via.len(), reg.as_ptr(), 0, &mut out) }, WAI_OK, "{}", last_error());
+    assert_eq!((out.kind, out.path, out.fallback), (WAI_DECODED_IMAGE, WAI_DECODE_PATH_SOLE, 1));
+    assert_eq!(fnv_of(&out), want);
     unsafe { wai_buffer_free(out.data) };
 }
 
diff --git a/wai/wai-rs/tests/jpegai_decode_path.rs b/wai/wai-rs/tests/jpegai_decode_path.rs
index ae23d7642..8ae0ce704 100644
--- a/wai/wai-rs/tests/jpegai_decode_path.rs
+++ b/wai/wai-rs/tests/jpegai_decode_path.rs
@@ -4,8 +4,10 @@
 //! (extensions/reconstruction-binding.md §2).
 //!
 //! - A capability with one decode path reports `DecodePath::Sole`, and the capability it
-//!   decoded through: the manifest's `capability`, or its `fallback`. Committed inputs
-//!   only; it never skips.
+//!   decoded through: the manifest's `capability`, or its `fallback`. A binding over a
+//!   `decode-equivalence` capability's picture is checked: it holds, a changed hash
+//!   differs, and a descriptor naming another picture is another reconstruction, never a
+//!   failure. Committed inputs only; neither test skips.
 //! - `wai.image.jpegai` without the integer weight bundle runs its float decode through
 //!   ONNX Runtime. The result says `DecodePath::Float`; a session claim recorded from it
 //!   says `"variant":"float"` and verifies, and the same decode named under a
@@ -13,11 +15,14 @@
 //!   reconstruction binding over its own pixels is reported as not checked, never as
 //!   holding. Needs the float path's model bundle (`WAI_JPEGAI_ASSETS`, the local
 //!   JPEG-AI asset directory exported from the ISO/IEC 6048-1 reference software; not in
-//!   the repository).
+//!   the repository). That directory is in no CI job, so this test runs locally only, on
+//!   a host that holds it, as `tests/jpegai_live.rs` does (wai-jpegai-e2e.yml).
 //! - With the integer weight bundle the exact decoder runs. The result says
 //!   `DecodePath::Exact`, the claim `"variant":"exact"`, and the binding check reports
-//!   whether the hash holds. Needs the bundle (`WAI_JPEGAI_WEIGHTS_DIR`, `WAI_JPEGAI_NORM`,
-//!   as `tests/jpegai_e2e_full.rs` reads them).
+//!   whether the hash holds, and a binding of another `decoderId` as another
+//!   reconstruction. Needs the bundle (`WAI_JPEGAI_WEIGHTS_DIR`, `WAI_JPEGAI_NORM`, as
+//!   `tests/jpegai_e2e_full.rs` reads them), which `tools/wai_jpegai_fetch_e2e_inputs.sh`
+//!   fetches; wai-neural-gates.yml runs it so, with `WAI_REQUIRE_SOTA_VECTORS=1`.
 //!
 //! Without their inputs the two JPEG-AI tests skip loudly, and with
 //! `WAI_REQUIRE_SOTA_VECTORS=1` they fail (support/require_vectors.rs).
@@ -107,6 +112,38 @@ fn a_capability_with_one_path_reports_it_and_the_capability_it_decoded_through()
     assert!(matches!(via.output, Decoded::Int(_)));
 }
 
+/// A binding over the picture of a `decode-equivalence` capability with one decode path
+/// (`wai.neural.int_mlicpp`, the committed vector): the check decodes and compares, and a
+/// descriptor that names another picture than the one decoded is reported as another
+/// reconstruction, not as a failure.
+#[test]
+fn a_binding_over_a_decode_equivalence_capability_is_checked_under_its_descriptor() {
+    let d = Path::new(env!("CARGO_MANIFEST_DIR")).join("tests/vectors/int_mlicpp");
+    let env = Wai::from_bytes(&std::fs::read(d.join("stream.wai")).unwrap()).unwrap();
+    let reg = ModelRegistry::new().register("wai.neural.int_mlicpp", d.join("model.wmb"));
+    let out = decode_envelope(&env, &reg).expect("decodes");
+    assert_eq!((out.capability.as_str(), out.path), ("wai.neural.int_mlicpp", DecodePath::Sole));
+    let Decoded::Image(img) = &out.output else { panic!("not a picture") };
+    let at = |width, height, format: &str| ReconstructionDescriptor {
+        decoder_id: 0,
+        profile: "int_mlicpp".into(),
+        level: "-".into(),
+        width,
+        height,
+        format: format.into(),
+        color_space: "bt709".into(),
+    };
+    let binding = ReconstructionBinding::bind(&img.rgb, at(img.width, img.height, "RGB8"));
+    assert_eq!(check_reconstruction_binding(&env, &reg, &binding).unwrap(), BindingCheck::Holds);
+    let mut changed = binding.clone();
+    changed.hash[0] ^= 1;
+    assert_eq!(check_reconstruction_binding(&env, &reg, &changed).unwrap(), BindingCheck::Differs);
+    for other in [at(img.width / 2, img.height * 2, "RGB8"), at(img.width, img.height, "RGBA8")] {
+        let b = ReconstructionBinding { descriptor: other, ..binding.clone() };
+        assert_eq!(check_reconstruction_binding(&env, &reg, &b).unwrap(), BindingCheck::OtherReconstruction(DecodePath::Sole));
+    }
+}
+
 /// The float decode of `wai.image.jpegai` says so in the result, in the claim and in the
 /// binding check.
 #[test]
@@ -169,5 +206,9 @@ fn the_jpegai_exact_decoder_reports_itself() {
     other[0] ^= 1;
     let changed = ReconstructionBinding::bind(&other, descriptor());
     assert_eq!(check_reconstruction_binding(&env, &reg, &changed).unwrap(), BindingCheck::Differs);
+    // A binding of another synthesis branch is over a reconstruction this decoder does not
+    // compute: not checked, and not failed, though its hash is this decode's.
+    let branch_1 = ReconstructionBinding::bind(&img.rgb, ReconstructionDescriptor { decoder_id: 1, ..descriptor() });
+    assert_eq!(check_reconstruction_binding(&env, &reg, &branch_1).unwrap(), BindingCheck::OtherReconstruction(DecodePath::Exact));
     std::fs::remove_dir_all(&root).ok();
 }
-- 
2.54.0 (Apple Git-157)


From 3c2081f6c228212756c94101b90e4e10d33c45fc Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:53:44 -0500
Subject: [PATCH 17/29] CHANGELOG: the decode-path entry names its breaks, and
 the research notes record the path
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Reason gains three variants and has no #[non_exhaustive], so the API check
counts them as a break, and the entry named none. It now acknowledges them,
the five neural dispatch functions' new return type (the neural feature is
outside the checked API, so the check does not report it), and wai_decoded_t's
two new fields. The entry also states the exact path, the per-object stage-hash
check, where each decode-path case runs, and that §1 records the five routes
while §7.1 cites the two float-decode records.

syfai-input's G11 and the September sweep's G9 said the result does not record
which path ran. Each gains a dated status note pointing to determinism-tiers §5
and neural::Dispatched::path.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md                     | 59 +++++++++++++++++++---------
 wai/research/2026-09-global-sweep.md |  2 +-
 wai/research/syfai-input.md          |  4 ++
 3 files changed, 45 insertions(+), 20 deletions(-)

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index 3bffbc49a..aa5a032c3 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -12,34 +12,55 @@ wire format may change and such changes are recorded here.
   §5). `neural::decode_envelope`, `decode_envelope_strict`, `decode_multi`, `decode_multi_strict` and `decode_payload`
   return a `Dispatched`: the output, the capability the payload decoded through (its own or the fallback) and the path
   of it that ran, `DecodePath::Sole`, `Exact` or `Float`. `wai.image.jpegai`'s float decode through ONNX Runtime and its
-  exact decoder now say which ran. `neural::check_reconstruction_binding` reads the path
-  ([extensions/reconstruction-binding.md](extensions/reconstruction-binding.md) §2): over the exact decoder it reports
-  whether the hash holds, and over the float decode it reports the binding as not checked, never as failed. The notes
-  that the dispatch did not report the path are gone from both texts. `tests/jpegai_decode_path.rs` runs the float path
-  and requires the result, the session claim and the binding check to say so; its JPEG-AI cases need inputs outside the
-  repository and skip without them.
-- **No tier between the two** (determinism-tiers §3, normative). The tier vocabulary stays closed. A decode of a
+  exact decoder now say which ran. The C ABI's `wai_decoded_t` carries the path (`WAI_DECODE_PATH_SOLE`, `_EXACT`,
+  `_FLOAT`) and whether a `WAI1` envelope decoded through its `fallback`, so a native sink states them in its receipt.
+  `neural::check_reconstruction_binding` reads the path and the binding's descriptor
+  ([extensions/reconstruction-binding.md](extensions/reconstruction-binding.md) §2): over a `decode-equivalence`
+  capability or the exact decoder it reports whether the hash holds; over the float decode it reports the binding as not
+  checked, and under a descriptor naming another reconstruction (another `decoderId`, format or size) as another
+  reconstruction, never as failed. The notes that the dispatch did not report the path are gone from both texts.
+- **Where the decode path is tested.** `tests/jpegai_decode_path.rs` requires the result, the session claim and the
+  binding check to state the path. Its cases with one path run on committed inputs in `wai-neural-gates.yml`; a job
+  there fetches the pinned `jpegai/v1` bundle and runs the exact decoder with `WAI_REQUIRE_SOTA_VECTORS=1`. The float
+  decode needs the JPEG-AI reference-software assets, which are not in the repository, so its test runs locally only,
+  on a host that holds them, as `tests/jpegai_live.rs` does.
+- **No tier between the two** (determinism-tiers §3, normative). A pin's tier vocabulary stays closed. A decode of a
   `NeuralIntegerExact` capability in reduced precision or floating point is not a decode of it; a capability whose
   payload holds several frame types takes the class of its weakest; and a clip digest, a stage hash or a reconstruction
-  binding over a sequence covers only frames each decoded at `decode-equivalence`.
+  binding over a sequence covers only frames each decoded at `decode-equivalence`, which a receipt checks per object.
+  These rules govern decoded media; generation-equivalence carries its own digest of the frames generated.
 - **A session claim states its decode path** ([extensions/jwp-receipts.md](extensions/jwp-receipts.md) §2.4.12). A
-  decoder entry's `variant` is `exact`, `float`, or `null` for a capability with one path. Rule 13 refuses a
-  `NeuralIntegerExact` capability named with another path (`session_variant_not_exact`); rule 14 refuses a `float` path
-  under a pin of `decode-equivalence` (`session_tier_exceeds_variant`); and against the stage claims, a stage hash over
-  a base an entry decoded at a weaker step, or over a base or layer decoded by a path that is not exact, is refused
-  (`session_stage_hash_not_exact`). `claim-conformance/` gains a vector for each, beside accepted ones for both
-  `wai.image.jpegai` paths, appended after the earlier vectors, which a test pins by digest; `claim_verify.py` and
-  `differential.py` read the same rules.
-- **determinism-tiers §1 and §7.1** record the five published routes to a reproducible stage, from the paper sections
-  read, and a line that reaches no tier: float synchronised on a tested hardware class (arXiv 2608.00483). The research
-  notes `syfai-input.md` §5 and `2026-09-global-sweep.md` (its row of arXiv 2608.00483, G7 and F5) are corrected against
-  the papers' bodies.
+  decoder entry's `variant` is `exact`, `float`, or `null` for a capability with one path; a `NeuralIntegerExact`
+  capability's integer path may also be named by the lowering its registration names, as opus-dred §7.2 names
+  `dred-int/1`. Those three are the exact paths. Rule 13 refuses a `NeuralIntegerExact` capability named with a path
+  that is not exact (`session_variant_not_exact`); rule 14 refuses any path that is not exact under a pin of
+  `decode-equivalence` (`session_tier_exceeds_variant`); and against the stage claims, a stage hash over a base an entry
+  decoded at a weaker step, or over a base or layer an entry decoded by a path that is not exact, is refused
+  (`session_stage_hash_not_exact`), whichever units that entry presented. `claim-conformance/` gains vectors for each,
+  beside accepted ones for both `wai.image.jpegai` paths and for a registered lowering, appended after the earlier
+  vectors, which a test pins by digest; `claim_verify.py` and `differential.py` read the same rules.
+- **determinism-tiers §1** records the five published routes to a reproducible stage, from the paper sections read, and
+  a line that reaches no tier: float synchronised on a tested hardware class (arXiv 2608.00483). **§7.1** cites the two
+  published records of float decode across devices (arXiv 2608.00483, 2606.28027) beside WAI's record. The research
+  notes `syfai-input.md` §5 and `2026-09-global-sweep.md` (its row of arXiv 2608.00483, G7, G9 and F5) are corrected
+  against the papers' bodies and the dispatch.
 - **A catalogue-prior record** ([research/catalogue-prior.md](research/catalogue-prior.md)): a linear basis learned on
   openly licensed films of one studio, measured against AV1, with its readings, its procedure, its decision rule
   and when that rule was fixed. The rule refutes a linear or spectral catalogue basis and exact recurrence across titles
   for a studio's catalogue; a learned prior is not measured. `tools/wai_catalogue_prior_verify.py` recomputes every
   BD-rate and verdict from the readings and checks every figure the page quotes, in the conformance workflow.
 
+- **Breaking:** `api:rust:wai::claim_receipt::Reason::SessionVariantNotExact`,
+  `api:rust:wai::claim_receipt::Reason::SessionTierExceedsVariant` and
+  `api:rust:wai::claim_receipt::Reason::SessionStageHashNotExact`: new variants of an exhaustive enum, which an
+  exhaustive match over `Reason` needs.
+- **Breaking:** `api:rust:wai::neural::decode_envelope`, `api:rust:wai::neural::decode_envelope_strict`,
+  `api:rust:wai::neural::decode_multi`, `api:rust:wai::neural::decode_multi_strict` and
+  `api:rust:wai::neural::decode_payload` return a `Dispatched` in place of a `Decoded`; the output is its `output`.
+  The `neural` feature is outside the checked API, so the check does not report these.
+- **Breaking:** `api:c:wai_decoded_t`: two fields after `fps_x_1000`, `path` and `fallback`, so the struct is larger. A
+  caller built against the old header is rebuilt against the new one.
+
 ### CI: a full runner disk fails a byte-equality leg as infrastructure, not as a divergence
 
 - **CI:** each leg of the two cross-architecture byte-equality matrices (`wai-byte-equality.yml`, and
diff --git a/wai/research/2026-09-global-sweep.md b/wai/research/2026-09-global-sweep.md
index f04cde830..6c09e3736 100644
--- a/wai/research/2026-09-global-sweep.md
+++ b/wai/research/2026-09-global-sweep.md
@@ -692,7 +692,7 @@ review.
 | G6 | **No WAI receipt says who can check the energy figure, and the base JWP receipt does not carry even the acquisition class that `energy-measurement.md` requires.** The precedents in §5.1 differ exactly on who can check: instrument-signed (OCMF), peer-reviewed but unsigned, log-registered (SCITT), TEE-attested with a derived trust level (the family's `sandbox-tee` `AttestTrust`), or producer-asserted. **The base JWP receipt** has no acquisition-class field; its group signature covers `total_joules_micro` and no class, and the per-object `joules_micro` is in neither the Merkle leaf nor the signature. `energy-measurement.md` §2 requires a class inside the signature for any non-zero figure, while its header says it adds "no new wire format". **The video receipt** signs its class. **The energy binding** carries its class in a C2PA companion assertion signed as part of whatever manifest carries it. **The OER receipt** carries an evidential grade and a ledger identity, and no class. 2030 §8 item 4 queues typed receipt classes. | `extensions/energy-measurement.md` header, §1, §2 and §5; `extensions/jwp-receipts.md` §2.1–§2.4; `joule-code/joule-code-rs/crates/joule-moq/src/receipt.rs` (`ObjectReceipt`, `GroupReceipt::signing_payload`: no class); `wai-rs/src/energy_binding.rs`, `wai-rs/src/video_receipt.rs`; `extensions/oer-receipts.md` §2–§3, `wai-rs/src/oer.rs`; no reference to the family ladder under `wai/` (grep, as in G4) | **Partly closed at link 10 (PR #112); the rest open.** Every receipt within energy-measurement's scope that `wai-rs` and `wai-quantum` seal can carry the class (link 08, PR #109). jwp-receipts §2.4.3–§2.4.7 add a `claim` kind with typed classes. An `energy` claim (§2.4.5) restates a JWP object's figure beside its class, the class's evidence, a coverage statement and a grade; a verifier holding the object receipt rejects the pair if the figures differ, and one holding the OER receipt the claim names checks that grade and figure agree. `claim_receipt.rs` (feature `receipt_claims`) implements it, with vectors in `claim-conformance/`. Open, as energy-measurement §1 records: the JWP object receipt has no member of its own for the class; the group total has none, and nothing carries one for it; the `WAI_RECEIPT` property's class and evidence sit outside any signature; an OER/2 receipt has no member for the class. On who can check: a claim carries its signer's authority and no more (§2.4.4); key discovery and an authority class for who may sign which class are not done (CHANGELOG, v0.4.0 entry for typed receipt classes); no WAI receipt references `sandbox-tee`'s ladder or `efficiency/1`'s attestation (grep). *Status after the v0.4.0 change "Energy reports, labelled object leaves, and what the signatures cover":* the object's class rides in its labelled leaf, a group total can carry a signed class of its own, and a `WAI_RECEIPT` property's class counts only where a signed statement restates it; joule-moq verifies both labelled forms. *Status after the v0.4.0 change "Key discovery, signing authority and module revocation":* key discovery and an authority class for who may sign which class are jwp-receipts §7 (`wai-rs` `key_log`, feature `key_discovery`; `signer-conformance/`); a hardware-class energy authority is confirmed only under a pinned anchor, and a verifier reports whether the key also holds `delivery` |
 | G7 | **Explicit parameter transmission appears in `determinism-tiers.md` only as a generic route. Its first published cross-NPU instance (arXiv 2606.28027) and its cost are not recorded.** Also unrecorded: the mixed-precision entropy-synchronisation line (arXiv 2608.00483: FP16/FP32 across GPUs, no bit-exactness claim); the quantised learned-coding line (arXiv 2007.04684, 2111.09348); and IEEE 1857.11's use of *reproducibility cross-checks* rather than bit-exact decode (**R** ⚠), a reproducibility notion the tier vocabulary does not name. Both 2606.28027 and 2608.00483 target a synchronised entropy stage, so at most `entropy-consistency`, and 2608.00483 only over the GPUs it tested. On `main`, whether such a decode may back `replicate` depends on which of G10's conditions is read. *Corrected 2026-11-03, body read:* arXiv 2608.00483 makes no bit-exactness claim in its abstract, though its body claims bit-exact hashes within a GPU generation (§III-B; its row above). | `extensions/determinism-tiers.md` §1–§3; `SPEC.md` §7 (read) | **Partly closed at link 02 (PR #103).** determinism-tiers §1 names arXiv 2606.28027 as the explicit-parameter instance and says the side information costs bitrate (no magnitude), and names a quantised decoder, arXiv 2312.11209, as the route to a reproducible reconstruction. Such a decode reaches `entropy-consistency` and does not back `replicate` (G10). Not in the tree at link 10 (grep): arXiv 2608.00483, the quantised line 2007.04684 / 2111.09348, and IEEE 1857.11's reproducibility cross-checks; the last is described in [`syfai-input.md`](syfai-input.md) §5, committed with this record. *Since 2026-11-03:* determinism-tiers §1 records arXiv 2608.00483 as float synchronised on a tested hardware class, which reaches no tier (its Table IV and §III-B), and the integer routes of arXiv 2502.20762 (§4.5) and of an integer-trained codec (OpenReview `KCQo0fXtFH`, abstract); the quantised line 2007.04684 / 2111.09348 is still not in the tree |
 | G8 | **Standards inputs not tracked in the tree:** SyfAI; 3GPP TR 26.942; DVB S100 and CM-EE; the beyond-VVC CfP; Gaussian Splat Coding Exploration 45; IEEE 1857.11 / P1857.11.1; UWA 042; India D2M; the JPEG AI core experiments on energy consumption and bit-exact reconstruction. | a case-insensitive, word-bounded `git grep -P` over tracked `*.md`, `*.rs`, `*.json`, `*.toml`, `*.yml`, `*.yaml`, `*.html`, `*.ts` and `*.js` files finds none of these designations outside this record. The one other hit for `uwa` is the Hausa word in `aac/aac-rs/crates/aac-core/src/translations_data.json` (`"ha":"uwa"`), which is not the organisation. MPEG's splat work *is* tracked, though not Exploration 45 by name: `media-codec-registry.md` and `codecs/mod.rs` register `wai.splat.sog` against "MPEG WG4" work in progress, `3d-neural-coding-landscape.md` records splat coding under standardisation, and 2030 §6.4 records an NNC amendment covering Gaussian splats (G11) | **Partly closed by the two notes committed with this record.** [`syfai-input.md`](syfai-input.md) tracks SyfAI, IEEE 1857.11 and UWA 042; [`energy-exposure-mapping.md`](energy-exposure-mapping.md) tracks TR 26.942, DVB S100 and CM-EE. The same grep over the tree at link 10 finds the beyond-VVC CfP, Exploration 45, India D2M and the JPEG AI core experiments in no document but this record. Those four stay open |
-| G9 | **WAI's texts and code class `wai.image.jpegai` in ways that contradict one another, and SPEC contradicts itself on it.** SPEC §5, the capability's registration (`SPEC.md`:197), describes the decode as "integer entropy front end (mANS) + integer dequant/value path + fixed-order synthesis and post-filters" and declares "**decode-equivalence**: one pinned reconstruction … reproducible on every machine (a portable decoded-pixel hash)". `reconstruction-binding.md` §1 (:30–31) says the decode to RGB is "byte-identical on every machine", and §2 (:72) sets the binding's criterion at `decode-equivalence`; `determinism-tiers.md` §2–§3 list the capability there. SPEC §7 lets a neural capability back `replicate` only "with no float operation anywhere" (:445–448). `capability_replicate_class` returns `StandardDefined` for it through the default arm, and SPEC §7 defines that class as "reconstruction is fixed by the named codec/standard … the neural float hazard does not apply"; neither half holds for JPEG AI (§2.2; its decode runs the model). The strict `replicate` check reads only the envelope's prior (`neural/mod.rs`:282–292). **Loose ends:** (i) `jpegai_synthesis.rs` and `jpegai_mans.rs` say they back `wai.neural.jpeg_ai`, a string registered nowhere; (ii) the decode has f32 steps between its integer stages — its dequantisation among them, although SPEC §5 calls it integer; (iii) the byte-for-byte claim is made normatively, in SPEC §5 and `reconstruction-binding.md` §1, and rests on an `exp()` from the platform maths library and on no measurement on a second platform; (iv) the reference sink's dispatch (`neural/mod.rs`:205–216) uses the exact decoder only when the integer weight bundle is present, and otherwise falls back to a float ONNX Runtime decode. | `SPEC.md` §5 (:197) and §7 (:445–448, the `StandardDefined` bullet at :476–478, :492–495); `extensions/reconstruction-binding.md` §1 (:21–35) and §2 (:72); `extensions/determinism-tiers.md` §2–§3; `wai-rs/src/codecs/mod.rs` (the default arm and the test asserting `StandardDefined`); `wai-rs/src/jpegai_decode.rs`, `jpegai_synthesis.rs`, `jpegai_mans.rs`, `jpegai_pipeline.rs`; `wai-rs/src/neural/mod.rs`:205–216, :282–292; `git grep -n decode-equivalence` over `wai/` | **Closed at link 02 (PR #103), with the maths library removed at link 03 (PR #104).** SPEC §5's row registers `entropy-consistency` (registry class `NeuralEntropyExact`), describes dequantisation, latent refinement, post-filters and colour conversion as IEEE-754 float with a fixed evaluation order, and says "**Downgraded in v0.4.0:** this row previously claimed decode-equivalence". `reconstruction-binding.md` §1–§2 make the pinned reconstruction a property of the exact decoder, reproducible under that decoder's build and floating-point conditions. SPEC §7 states the new class and the reason. Loose ends: (i) fixed — the module docs name `wai.image.jpegai`, and `wai.neural.jpeg_ai` appears nowhere else in the tree (grep); (ii) the float steps remain and are now stated as float; (iii) link 03 builds `exp` and rounding from IEEE-754 basic operations (`float_repro`), guards the decode path with clippy's `disallowed_methods` and a linked-symbol check (`tests/jpegai_decode_symbols.rs`), and pins nine hashes for one codestream; determinism-tiers §6 records them identical on macOS arm64, wasm32-wasip2, Linux arm64 and Linux x86_64 under user-mode emulation, and not measured on Windows, macOS x86_64 or x86_64 hardware; (iv) the float fallback remains, and SPEC §5, determinism-tiers §5 and reconstruction-binding §1 now state it, and that the result does not record which path ran. Still true: a `norm.json` with a missing or mistyped key panics in the loaders instead of falling back, as the `try_jpegai_exact` doc comment says |
+| G9 | **WAI's texts and code class `wai.image.jpegai` in ways that contradict one another, and SPEC contradicts itself on it.** SPEC §5, the capability's registration (`SPEC.md`:197), describes the decode as "integer entropy front end (mANS) + integer dequant/value path + fixed-order synthesis and post-filters" and declares "**decode-equivalence**: one pinned reconstruction … reproducible on every machine (a portable decoded-pixel hash)". `reconstruction-binding.md` §1 (:30–31) says the decode to RGB is "byte-identical on every machine", and §2 (:72) sets the binding's criterion at `decode-equivalence`; `determinism-tiers.md` §2–§3 list the capability there. SPEC §7 lets a neural capability back `replicate` only "with no float operation anywhere" (:445–448). `capability_replicate_class` returns `StandardDefined` for it through the default arm, and SPEC §7 defines that class as "reconstruction is fixed by the named codec/standard … the neural float hazard does not apply"; neither half holds for JPEG AI (§2.2; its decode runs the model). The strict `replicate` check reads only the envelope's prior (`neural/mod.rs`:282–292). **Loose ends:** (i) `jpegai_synthesis.rs` and `jpegai_mans.rs` say they back `wai.neural.jpeg_ai`, a string registered nowhere; (ii) the decode has f32 steps between its integer stages — its dequantisation among them, although SPEC §5 calls it integer; (iii) the byte-for-byte claim is made normatively, in SPEC §5 and `reconstruction-binding.md` §1, and rests on an `exp()` from the platform maths library and on no measurement on a second platform; (iv) the reference sink's dispatch (`neural/mod.rs`:205–216) uses the exact decoder only when the integer weight bundle is present, and otherwise falls back to a float ONNX Runtime decode. | `SPEC.md` §5 (:197) and §7 (:445–448, the `StandardDefined` bullet at :476–478, :492–495); `extensions/reconstruction-binding.md` §1 (:21–35) and §2 (:72); `extensions/determinism-tiers.md` §2–§3; `wai-rs/src/codecs/mod.rs` (the default arm and the test asserting `StandardDefined`); `wai-rs/src/jpegai_decode.rs`, `jpegai_synthesis.rs`, `jpegai_mans.rs`, `jpegai_pipeline.rs`; `wai-rs/src/neural/mod.rs`:205–216, :282–292; `git grep -n decode-equivalence` over `wai/` | **Closed at link 02 (PR #103), with the maths library removed at link 03 (PR #104).** SPEC §5's row registers `entropy-consistency` (registry class `NeuralEntropyExact`), describes dequantisation, latent refinement, post-filters and colour conversion as IEEE-754 float with a fixed evaluation order, and says "**Downgraded in v0.4.0:** this row previously claimed decode-equivalence". `reconstruction-binding.md` §1–§2 make the pinned reconstruction a property of the exact decoder, reproducible under that decoder's build and floating-point conditions. SPEC §7 states the new class and the reason. Loose ends: (i) fixed — the module docs name `wai.image.jpegai`, and `wai.neural.jpeg_ai` appears nowhere else in the tree (grep); (ii) the float steps remain and are now stated as float; (iii) link 03 builds `exp` and rounding from IEEE-754 basic operations (`float_repro`), guards the decode path with clippy's `disallowed_methods` and a linked-symbol check (`tests/jpegai_decode_symbols.rs`), and pins nine hashes for one codestream; determinism-tiers §6 records them identical on macOS arm64, wasm32-wasip2, Linux arm64 and Linux x86_64 under user-mode emulation, and not measured on Windows, macOS x86_64 or x86_64 hardware; (iv) the float fallback remains, and SPEC §5, determinism-tiers §5 and reconstruction-binding §1 now state it, and that the result does not record which path ran. Still true: a `norm.json` with a missing or mistyped key panics in the loaders instead of falling back, as the `try_jpegai_exact` doc comment says. *Since 2026-11-04:* the result records which path ran, against (iv): the dispatch returns it (`neural::Dispatched::path`, determinism-tiers §5), a session claim's decoder entry states it as its `variant` (jwp-receipts §2.4.12), and the reference check of a reconstruction binding reads it (reconstruction-binding §2) |
 | G10 | **WAI's texts give three conditions for a neural decode to back `replicate` or a portable reconstruction hash, and SPEC states all three.** (1) **All-integer:** SPEC §7 (`SPEC.md`:445–448, :492–495), enforced in code on the envelope's declared prior (`replicate_is_sound` requires `DeterminismClass::IntegerExact`, a substring heuristic over `Prior.determinism`). (2) **`decode-equivalence` with fixed-order float admitted:** `determinism-tiers.md` §2 defines the tier over "integer / fixed-order arithmetic end to end" (:57–58), and SPEC §5 registers `wai.image.jpegai`, whose float stages run in a fixed order, at that tier. (3) **`entropy-consistency` as the floor:** SPEC §7's closing paragraph (:497–505) calls it "the floor for a sound neural `replicate`", and `determinism-tiers.md` §2–§3 repeat it; §3 also says "a float neural decode meets neither tier", which does not square with the explicit-parameter route. A decode with integer entropy and fixed-order float synthesis meets (2) and (3) but not (1). | `SPEC.md` §5 (:197) and §7 (:445–448, :492–495, :497–505); `extensions/determinism-tiers.md` §1–§3; `extensions/reconstruction-binding.md` §1–§2; `wai-rs/src/container.rs` (`Prior::determinism_class` at :108–117, `is_cross_hardware_exact`, `replicate_is_sound`) | **Closed at link 02 (PR #103), on condition (1).** SPEC §7: a neural capability backs `replicate` only if its entire decode path is integer / fixed-point with no float operation, "the only construction by which a neural capability is registered at `decode-equivalence`"; `entropy-consistency` ("necessary for a sound neural `replicate` … and not sufficient") and a reconstruction pinned by one implementation's fixed-order float stages MUST NOT back it. determinism-tiers §2 defines `decode-equivalence` over "integer / fixed-point arithmetic end to end", and §3 says a decode that derives its entropy parameters in float meets neither tier while one that reads them from the codestream, or derives them in integer arithmetic, and runs later stages in float meets `entropy-consistency` only. The strict check reads the registry class before the prior. Remaining: the prior check still reads a sender-declared string by substring, which determinism-tiers §3 states |
 | G11 | **Three volumetric registrations sit against MPEG's 2026 splat and point-cloud work, and none of the sweep's rows was mapped to them.** `wai.splat.sog` is a candidate "registered so a capability exists ahead of an MPEG freeze" (registry: "MPEG WG4 WIP", "RF (open work)"). `wai.pc.gpcc` (23090-9) and `wai.pc.vpcc` (23090-5) are registered POOL. None has a decoder inside `wai-rs`; the registry records demonstrated sink-side decodes via `tools/wai_carriage_demo.sh` (not re-run here, and not in CI). The sweep found a V-PCC Gaussian-splat CDAM (MPEG 153) *(2026-09-30: at Draft Amendment since MPEG 154, WG 4 N 837)*, a G-PCC Gaussian-splat Amd 1 at CDAM (MPEG 154), Enhanced G-PCC ISO/IEC 23090-38 at FDIS (MPEG 155), and Exploration 45's document series up to a call for content at MPEG 155, all P✓; its tracks and dates are **R**. Questions: **(a)** whether what MPEG standardises is what `wai.splat.sog` names, which cites Self-Organizing Gaussians / SOGS while the part page names no candidate technology; **(b)** whether its royalty class holds (`capability_license` returns `RoyaltyFree` through the default arm; nothing the sweep read establishes the licensing of a codec MPEG selects); **(c)** whether 23090-38 is a payload `wai.pc.gpcc` admits or needs its own capability, and whether the splat amendments change what the two PCC capabilities admit; **(d)** the replicate class: all three fall through to `StandardDefined`, although `wai.splat.sog` has no frozen standard. The tree's "WG4" attribution was not checked (the Exploration 45 page names no working group), and `3d-neural-coding-landscape.md` describes a "future `wai.pointcloud.*`" without mentioning the registered `wai.pc.*` strings. | `research/media-codec-registry.md`; `wai-rs/src/codecs/mod.rs` (`CAP_PC_GPCC`, `CAP_PC_VPCC`, `CAP_SPLAT_SOG` and their arms); `tools/wai_carriage_demo.sh`; `wai-rs/tests/carriage.rs`; `research/3d-neural-coding-landscape.md`; 2030 §6.4 | **(d) closed at link 02 (PR #103); (b) made explicit at link 04 (PR #105); (a), (c), the "WG4" attribution and the `wai.pointcloud.*` sentence open.** All three are `StandardUnclassified`, the class for a standard whose decoder-conformance clauses have not been checked (SPEC §7 notes that `wai.splat.sog` names no frozen standard yet). `capability_license` lists `wai.splat.sog` explicitly as royalty-free ("SOG open work"), which makes the class explicit and does not establish it. *(2026-09-30: `capability_license` now classifies `wai.splat.sog` patent-pool, conservatively and provisionally while it is a candidate; SPEC §5, Licence classes)* `codecs/mod.rs` and `media-codec-registry.md` still say "MPEG WG4", and `3d-neural-coding-landscape.md` still describes a "future `wai.pointcloud.*`". Exploration 45 is untracked (G8) |
 
diff --git a/wai/research/syfai-input.md b/wai/research/syfai-input.md
index f0d83952b..a7443a2e8 100644
--- a/wai/research/syfai-input.md
+++ b/wai/research/syfai-input.md
@@ -826,6 +826,10 @@ defines the whole decode in CI (G8). What remains is stated in the texts, not
 contradicted by them: the float fallback exists and the result does not record
 which path ran, and a `norm.json` with a missing or mistyped key panics in the
 loaders instead of falling back (the `try_jpegai_exact` doc comment says so).
+*Since 2026-11-04:* the result records which path ran. The dispatch returns it
+(`neural::Dispatched::path`, determinism-tiers §5), a session claim's decoder
+entry states it as its `variant` (jwp-receipts §2.4.12), and the reference check
+of a reconstruction binding reads it (reconstruction-binding §2).
 
 **G12: No normative behaviour for a parameter-set mismatch, and a
 per-rendition pin the code carries but never checks.** UC-2 sets out the
-- 
2.54.0 (Apple Git-157)


From bbe1c06fd0066f9793096d0a399778c6c9019e9f Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 03:09:49 -0500
Subject: [PATCH 18/29] claim-conformance: the differential reaches rule 14 and
 the stage-hash refusal

In the sessions CI's run of the differential generates (seed 1) neither
session_tier_exceeds_variant nor session_stage_hash_not_exact was the verdict
once. An entry naming a stage set's base almost never had the unit the session
names the set at, so session_first_unit_mismatch came first, and a path that
is not exact under a pin of decode-equivalence rarely met a capability with no
class. Entries naming the base now take that unit, as an honest sink states
it, four times in five, and the path defect moves its entry to another
capability more often, an unclassified one half the time.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/claim-conformance/differential.py | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/wai/claim-conformance/differential.py b/wai/claim-conformance/differential.py
index 1ed7a7b59..d7e25cccf 100644
--- a/wai/claim-conformance/differential.py
+++ b/wai/claim-conformance/differential.py
@@ -733,6 +733,13 @@ def sessions(ins, rng, n):
                 w = w.replace('"tier":"none"', '"tier":"nonе"')
             stage_wires.append(w)
             stage_sets.append({"first_unit": f, "claim": CV.link_hash(json.loads(w)) if rng.random() < 0.95 else "00" * 32})
+            # An entry naming the set's base has the unit the session names the set at, as an
+            # honest sink states it, so a stage hash is checked against the step and the path
+            # of each entry that decoded the base (item 5) and not refused before it (item 4).
+            if base_hash == entry and rng.random() < 0.8:
+                for d in decoders:
+                    if d["object"] == entry:
+                        d["first_unit"] = f
             # The set's base and its layer object among the objects received, or not.
             if rng.random() < 0.85:
                 objects.append(obj(base_hash, counted=4000))
@@ -851,8 +858,9 @@ def sessions(ins, rng, n):
                 d["variant"] = rng.choice(["float", "exact", "fp16", "dred-int/1"])
                 d["pin"], d["tier"] = rng.choice([(CV.blake3(b"differential: a pin").hex(), "decode-equivalence"),
                                                   (CV.blake3(b"differential: a pin").hex(), "entropy-consistency"), (None, None)])
-                if rng.random() < 0.3:
-                    d["capability"] = rng.choice(["org.example.picture", "wai.image.jpegai", "wai.audio.opus.dred.int"])
+                if rng.random() < 0.6:
+                    d["capability"] = rng.choice(["org.example.picture", "org.example.picture", "wai.image.jpegai",
+                                                  "wai.audio.opus.dred.int"])
         if rng.random() < 0.5:
             ctx["redecoded"] = [{"index": j, "out_digest": rng.choice([p["out_digest"], CV.blake3(b"differential: redecoded").hex()])}
                                 for j, p in enumerate(ps) if rng.random() < 0.3]
-- 
2.54.0 (Apple Git-157)


From 8390a36dcd7ac2ca7bdf5a92640154ca48198f1d Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 03:22:46 -0500
Subject: [PATCH 19/29] claim-conformance: a decode-path sweep in the
 differential, and stage hashes over stage sets that pass

The previous change did not reach either rule: a planted regression of rule 14
to `float` alone still left the differential with nothing to report, since the
random sessions are refused by an earlier rule first. session_decode_paths now
sweeps one picture through one entry over five capabilities (classified at each
tier, or not at all), seven terms for the path, three pins and both kinds of
decoder, where nothing else decides, so rules 6, 13 and 14 each give the
verdict. session_stage_bounds, whose sessions pass every earlier rule, states a
stage hash a third of the time, and then half the time unpins the base's entry
or names another path on an entry.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/claim-conformance/differential.py | 31 ++++++++++++++++++++++++++-
 1 file changed, 30 insertions(+), 1 deletion(-)

diff --git a/wai/claim-conformance/differential.py b/wai/claim-conformance/differential.py
index d7e25cccf..c2b120f90 100644
--- a/wai/claim-conformance/differential.py
+++ b/wai/claim-conformance/differential.py
@@ -925,8 +925,18 @@ def session_stage_bounds(ins, rng, n):
         if rng.random() < 0.15:
             # The base's entry, or its layer's, at another first unit than the set's.
             rng.choice(decoders)["first_unit"] += 1
+        # At decode-equivalence the claim states a stage hash (item 5), which holds while the
+        # base's entry decodes it at decode-equivalence and each entry runs an exact path;
+        # now and then the base's entry is unpinned, or an entry names another path.
+        tier = rng.choice(["none", "none", "decode-equivalence"])
+        if tier == "decode-equivalence" and rng.random() < 0.5:
+            d = rng.choice(decoders)
+            if d is decoders[0] and rng.random() < 0.5:
+                d["pin"], d["tier"] = None, None
+            else:
+                d["variant"] = rng.choice(["float", "fp16", "exact", "whm1/1"])
         w = session_stage_claim(rng, n_set, rng.choice([at1, at1, max(0, at1 - 1), min(n_set, at1 + 1)]), VIEWER, base_hash,
-                                late=late if late and at1 + late <= n_set else 0)
+                                late=late if late and at1 + late <= n_set else 0, tier=tier)
         objects = []
         for h, cnt in [(base_hash, 4000), (CV.blake3(b"differential: a stage layer").hex(), 700),
                        (CV.blake3(b"differential: a late stage layer").hex(), 650)]:
@@ -1606,6 +1616,24 @@ def session_caps(ins):
                     {"outputs": copy.deepcopy(ps)})
 
 
+def session_decode_paths(ins):
+    """§2.4.12 rules 6, 13 and 14 where nothing else decides: one picture through one entry,
+    for every capability below (classified at each tier, or not at all), decoded by a
+    builtin or a module, on a path named by every term below (exact, float, a lowering a
+    registration names, one it names for another capability, one no revision registers, or
+    none), under a pin of either tier or none."""
+    base = ins.receipt("session_variant_float_accept")
+    pin = CV.blake3(b"differential: a pin").hex()
+    for cap in ["org.example.picture", "wai.image.jpegai", "wai.neural.int_hyper", "wai.audio.opus.dred.int", "wai.neural.dcvc_rt"]:
+        for variant in [None, "exact", "float", "fp16", "dred-int/1", "whm1/1", "jpegai-int/1"]:
+            for tier in [None, "entropy-consistency", "decode-equivalence"]:
+                for decoder, status in [("builtin:wai@0.4.0", None), ("module:sha256:" + "5e" * 32, "pinned")]:
+                    d = dict(base["body"]["decoders"][0], capability=cap, variant=variant, pin=pin if tier else None, tier=tier,
+                             decoder=decoder, module_status=status)
+                    body = dict(base["body"], decoders=[d])
+                    ins.add(f"session_decode_path_{cap}_{variant}_{tier}_{status}", CV.jcs(sign(dict(base, body=body), VIEWER)))
+
+
 def figures(ins, rng, n):
     oer_t = ins.by_name["energy_matches_oer_accept"]["context"]["oer"]["receipt"]
     claim_t = ins.receipt("energy_matches_oer_accept")
@@ -2119,6 +2147,7 @@ def main(argv):
     stages(ins, random.Random(f"stages {a.seed}"), a.random)  # its own stream, as classes()
     sessions(ins, random.Random(f"sessions {a.seed}"), a.random)  # its own stream, as classes()
     session_caps(ins)
+    session_decode_paths(ins)
     session_stage_bounds(ins, random.Random(f"session bounds {a.seed}"), a.random)  # its own stream, as classes()
     session_stage_epochs(ins, random.Random(f"session epochs {a.seed}"), a.random)  # its own stream, as classes()
     session_held(ins, random.Random(f"session held {a.seed}"), a.random)  # its own stream, as classes()
-- 
2.54.0 (Apple Git-157)


From 7ca9e889b9e81965c77abf48e1e2b430b5474a65 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Tue, 3 Nov 2026 23:13:40 -0500
Subject: [PATCH 20/29] =?UTF-8?q?SPEC=20=C2=A73.1:=20pin=20by=20default;?=
 =?UTF-8?q?=20wai=20verify=20reports=20an=20unpinned=20envelope,=20and=20a?=
 =?UTF-8?q?=20decode's=20record=20names=20the=20set=20it=20read?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

An encoder SHOULD pin the parameter set a payload was coded against wherever a
capability that decodes it is registered as decoding against one, whatever the
intent (SPEC §3.1, Pin by default).
An envelope that pins nothing there still conforms in 1.x; a verifier reports it
(prior_unpinned), informatively: Wai::check_unpinned, WaiMulti::check_unpinned and
codecs::pin_by_default. wai verify now reads the envelopes under each path, refusing
one whose pin breaks SPEC §3.1 and noting prior_unpinned; wai inspect and wai wrap note
it too. pin-conformance/unpinned.json holds the vectors, reached by wai_pin_vectors and
by pin_verify.py, which reads the classes from the published registry export.

A reconstruct-session decoder entry's pin (jwp-receipts §2.4.12) now names the set an
unpinned decode read, by the digest a pin over those bytes names, with the tier none.
The reference sink's base decode gives it (staged::BaseDecode::set, from
IntParams::read_pin), and stage_measure's sessions carry it.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md                        |  21 ++
 wai/SPEC.md                             |  24 ++-
 wai/catalogue/ops.json                  |   8 +-
 wai/extensions/jwp-receipts.md          |   4 +-
 wai/extensions/staged-measurement.md    |   6 +-
 wai/pin-conformance/README.md           |  18 +-
 wai/pin-conformance/pin_verify.py       |  43 +++-
 wai/pin-conformance/unpinned.json       | 265 ++++++++++++++++++++++++
 wai/wai-rs/src/bin/wai.rs               |  28 ++-
 wai/wai-rs/src/bin/wai_cli/dx/verify.rs |  66 +++++-
 wai/wai-rs/src/bin/wai_pin_vectors.rs   |  42 +++-
 wai/wai-rs/src/catalogue.rs             |   8 +-
 wai/wai-rs/src/codecs/mod.rs            |  11 +
 wai/wai-rs/src/container.rs             |  89 +++++++-
 wai/wai-rs/src/int_payload/mod.rs       |  17 ++
 wai/wai-rs/src/stage_measure/decode.rs  |  11 +-
 wai/wai-rs/src/stage_measure/session.rs |  40 ++++
 wai/wai-rs/src/staged/mod.rs            |  18 +-
 wai/wai-rs/tests/cli_dx.rs              |  50 ++++-
 19 files changed, 726 insertions(+), 43 deletions(-)
 create mode 100644 wai/pin-conformance/unpinned.json

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index aa5a032c3..ff63e017e 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -6,6 +6,27 @@ wire format may change and such changes are recorded here.
 
 ## Unreleased
 
+### Pin by default: an encoder pins the set a payload was coded against, and a verifier reports an envelope that does not
+
+- **SPEC §3.1, *Pin by default*.** An encoder SHOULD pin the parameter set a payload was coded against wherever a
+  capability that decodes it is registered as decoding against one, whatever the intent. Only `replicate` over a
+  `NeuralIntegerExact` capability required a pin, so every other unpinned decode ran against whatever set the sink
+  held, and a set other than the payload's gave plausible output, not an error. An envelope that pins nothing there
+  still conforms in `1.x`, and a sink decodes it as before; a verifier reports it (`prior_unpinned`), informatively.
+- **The report.** `Wai::check_unpinned`, `WaiMulti::check_unpinned` and `codecs::pin_by_default` make it. `wai
+  verify` now reads the envelopes (`*.wai`) under each path: it refuses one whose pin a verifier can tell is wrong
+  without holding a set (SPEC §3.1, §7), and notes `prior_unpinned`, which refuses nothing. `wai inspect` notes it,
+  and so does `wai wrap`, which pins no prior. `pin-conformance/unpinned.json` holds its vectors, reached by
+  `wai_pin_vectors verify` and by `pin_verify.py`, which reads the classes from the published registry export.
+- **A record of a decode names the set it read.** A `reconstruct-session` decoder entry's `pin` (jwp-receipts
+  §2.4.12) now names the set an unpinned decode read, by the digest a pin over those bytes names, with the tier
+  `none`; it was `null`. The reference sink's base decode gives it (`staged::BaseDecode::set`, from
+  `int_payload::IntParams::read_pin`), and the sessions `stage_measure` seals carry it. Every committed session and
+  golden pins its bases, so none changes.
+
+- **Breaking:** `api:rust:wai::staged::BaseDecode::set`: a new member of a public struct, the set the decode read; a
+  struct literal or a pattern that lists every member adds it.
+
 ### A decode reports the path that ran, and no tier lies between the two
 
 - **The reference dispatch reports the decode path** ([extensions/determinism-tiers.md](extensions/determinism-tiers.md)
diff --git a/wai/SPEC.md b/wai/SPEC.md
index 882d7e08c..1b5f86246 100644
--- a/wai/SPEC.md
+++ b/wai/SPEC.md
@@ -227,7 +227,9 @@ An envelope delivered over HTTP is one object of the proposed media type
   payload (§4), where that capability decodes against a parameter set. Absent, or
   `null`, the key is not written, and a decode through such a capability is
   unpinned: it runs against whatever parameter set the sink holds, and it cannot
-  back `intent = replicate` over a `NeuralIntegerExact` capability (§7).
+  back `intent = replicate` over a `NeuralIntegerExact` capability (§7). An encoder
+  SHOULD write it wherever such a capability decodes the payload, whatever the intent
+  (§3.1, *Pin by default*).
 - `security` — OPTIONAL. Absent, `null`, or `encryption: "none"` ⇒ the
   payload is plaintext (the default). Otherwise the payload is encrypted
   and MUST be recovered before dispatch; see §6. The capability, media,
@@ -424,6 +426,26 @@ decodes against a parameter set, and every such capability is registered, means
 nothing: an encoder MUST NOT write one, a verifier reports it
 (`prior_not_applicable`), and a sink ignores it.
 
+**Pin by default.** A pin is OPTIONAL in `1.x`, and only `replicate` over a
+`NeuralIntegerExact` capability requires one (§7). Elsewhere an unpinned decode runs
+against whatever parameter set the sink holds, and a set other than the one the payload
+was coded against gives plausible output, not an error (§4). So:
+
+1. An encoder SHOULD pin the parameter set a payload was coded against wherever a
+   capability that decodes the payload — `model_requirement.capability`,
+   `model_requirement.fallback`, or a `WAI2` entry's capability — is registered as
+   decoding against one, whatever the envelope's `intent`. `wai.generate.spectral` is
+   the exception above: its payload names its set.
+2. An envelope that pins nothing there conforms. A sink decodes it as before.
+3. A verifier reports it (`prior_unpinned`). The report is informative: it is not a
+   finding that the envelope does not conform, and it is made whatever else the verifier
+   finds. A capability this revision does not register is not reported, since this
+   revision does not know what it decodes against.
+4. A sink that records a decode names the parameter set it read, pinned or not, by the
+   digest a pin over those bytes names ([jwp-receipts](extensions/jwp-receipts.md)
+   §2.4.12, a decoder entry's `pin`), so that a decode against another set is visible
+   after the fact.
+
 **Sinks of earlier revisions.** A sink that implements an earlier text of this
 specification may do one of two things with a pin (§8). It may reject the
 envelope as carrying a member it does not understand, which fails closed. Or it
diff --git a/wai/catalogue/ops.json b/wai/catalogue/ops.json
index e2dacb4b3..93346d068 100644
--- a/wai/catalogue/ops.json
+++ b/wai/catalogue/ops.json
@@ -418,12 +418,12 @@
     },
     {
       "id": "receipt.verify",
-      "summary": "Every receipt index (*.wai.receipts) under each path, its group receipts and claims checked offline, and every conformance result and fleet report bundle there",
+      "summary": "Every receipt index (*.wai.receipts) under each path, its group receipts and claims checked offline, and every conformance result, fleet report bundle and envelope there",
       "cli": "wai verify",
       "aliases": [],
       "usage": [
-        "wai verify <path>...   the receipts (receipt indexes *.wai.receipts), conformance results (*.wai-conform.json)",
-        "    and fleet report bundles (*.wai-fleet.json) under each path, checked offline"
+        "wai verify <path>...   the receipts (receipt indexes *.wai.receipts), conformance results (*.wai-conform.json),",
+        "    fleet report bundles (*.wai-fleet.json) and envelopes' pins (*.wai) under each path, checked offline"
       ],
       "placement": "local",
       "effect": "reads",
@@ -445,7 +445,7 @@
             "items": {
               "type": "string"
             },
-            "description": "Files or directories holding receipt indexes, conformance results or fleet report bundles.",
+            "description": "Files or directories holding receipt indexes, conformance results, fleet report bundles or envelopes.",
             "x-cli": {
               "positional": true
             }
diff --git a/wai/extensions/jwp-receipts.md b/wai/extensions/jwp-receipts.md
index 3a78a28bd..49908bb1c 100644
--- a/wai/extensions/jwp-receipts.md
+++ b/wai/extensions/jwp-receipts.md
@@ -1036,8 +1036,8 @@ A **decoder entry**:
 | `path` | string | `primary` or `fallback`. |
 | `decoder` | string | `builtin:<crate>@<version>` — `<crate>` 1 to 64 of `a`–`z`, `0`–`9`, `_` and `-`, `<version>` 1 to 64 of ASCII letters, digits, `.`, `+` and `-`, beginning with a digit — or `module:sha256:` followed by a module's digest (§7.9), 64 lowercase hexadecimal digits. |
 | `variant` | term or `null` | The decode path within the decoder that ran, where the capability has more than one: `exact` for the path that computes the reconstruction the capability registers (for `wai.image.jpegai`, its exact decoder), or `float` for a path that runs a stage in floating point the registration does not fix (for `wai.image.jpegai`, its float decode) ([determinism-tiers](determinism-tiers.md) §5). `null` for a capability with one path. A `NeuralIntegerExact` capability's integer path may also be named by the lowering its registration names ([integer-operators](integer-operators.md) §11, its `ops` member's `lowering`), as a `present` entry of [opus-dred](opus-dred.md) §7.2 names `dred-int/1` ([synthesis](synthesis.md) §10.3). The entry states an **exact path** when its `variant` is `null`, `exact`, or, for a capability of the replicate class `neural-int-exact`, that lowering; any other term, `float` among them, names a path that is not exact. A sink states the path that ran, as the reference dispatch reports it, not the tier its capability is registered at. |
-| `pin` | 32 B hex or `null` | The `sha256` of the parameter-set pin the sink verified (SPEC §3.1). |
-| `tier` | string or `null` | That pin's declared tier as SPEC §3.1 reads it: `decode-equivalence`, `entropy-consistency` or `none`. `null` exactly when `pin` is. |
+| `pin` | 32 B hex or `null` | The parameter set the decode read, by the digest a pin over it names (SPEC §3.1, *Pin by default*): the `sha256` of the pin the sink verified; for a decode that verified no pin, the digest a pin over the bytes it read would name, which is the SHA-256 of the one file of a capability registered with a one-file set, and otherwise the SHA-256 of the listing (`wai.prior.set`) of the files it read. `null` exactly when the decode read no parameter set. |
+| `tier` | string or `null` | That pin's declared tier as SPEC §3.1 reads it: `decode-equivalence`, `entropy-consistency` or `none`; `none` for a decode that verified no pin, which nothing declared. `null` exactly when `pin` is. |
 | `module_status` | string or `null` | For a `module:` decoder, the anchor its grant was confirmed under (§7.9): `pinned` or `fetched`. `null` for a `builtin:` decoder. |
 | `leaves` | integer | The presentations decoded through the entry. |
 
diff --git a/wai/extensions/staged-measurement.md b/wai/extensions/staged-measurement.md
index bfad27ae5..8fb268145 100644
--- a/wai/extensions/staged-measurement.md
+++ b/wai/extensions/staged-measurement.md
@@ -607,8 +607,10 @@ node = SHA-256("wai:out-node\x01" ‖ left ‖ right)
 - `decoder` is the index of the decoder entry that produced the output: entries are
   (capability, pin, object) triples in first-use order, presentations taken by (layers,
   unit), the object being the content hash ([jwp-receipts](jwp-receipts.md) §2.1) of
-  the envelope decoded. For a title, each base's capability with the pin it verified and
-  its envelope, then each layer's refinement capability and envelope. For a session
+  the envelope decoded. For a title, each base's capability with the parameter set its
+  decode read, named as a decoder entry's `pin` names it (jwp-receipts §2.4.12: the pin it
+  verified, where it verified one), and its envelope, then each layer's refinement
+  capability and envelope. For a session
   claim, the entries are its decoder entries (jwp-receipts §2.4.12), in the same order.
 - Leaves are ordered by (unit, layers), and a (unit, layers) given twice is refused.
 - The root over `n ≥ 1` leaves is RFC 6962's Merkle tree hash: one leaf is itself;
diff --git a/wai/pin-conformance/README.md b/wai/pin-conformance/README.md
index 081e7cf75..0c43ba044 100644
--- a/wai/pin-conformance/README.md
+++ b/wai/pin-conformance/README.md
@@ -31,6 +31,11 @@ for every case:
   `tier_unregistered`, `supersedes_malformed`, `form_not_registered`,
   `form_required`, `tier_exceeds_class`, a `replicate_*` reason), or `ok`. For a
   `WAI2` manifest the finding is `[entry index, reason]`.
+- `unpinned.json` — the report a verifier makes about a manifest that pins no
+  parameter set where a capability it decodes through is registered as decoding
+  against one (SPEC §3.1, *Pin by default*): `prior_unpinned`, or `none`. It is
+  informative, and made whatever `static.json` finds. For a `WAI2` manifest, the index
+  of each entry it is made about.
 
 **Listings.** Every `listings/<name>.sha256` is, byte for byte, the listing of
 `sets/<name>/`: regular files, names that begin with `.` left out, sorted by path as
@@ -44,9 +49,12 @@ Files are compared as bytes; `.gitattributes` turns off line-ending conversion.
 
 - `wai_pin_vectors verify pin-conformance` (wai-rs) runs every file through the
   reference implementation's in-memory pin core (`wai::pin::verify_bytes`), its
-  selection (`WaiMulti::pick`) and its static checks (`Wai::check_pin`,
-  `WaiMulti::check_pins`). It runs in both corpus loops of the conformance workflow.
+  selection (`WaiMulti::pick`), its static checks (`Wai::check_pin`,
+  `WaiMulti::check_pins`) and its report (`Wai::check_unpinned`,
+  `WaiMulti::check_unpinned`). It runs in both corpus loops of the conformance workflow.
   `wai_pin_vectors generate pin-conformance` rewrites the listings from the sets.
-- `pin_verify.py` checks `cases.json` and `listings/` with the Python standard
-  library alone, re-implementing the listing rules and the ordered checks from the
-  SPEC text. It shares no code with wai-rs.
+- `pin_verify.py` checks `cases.json`, `unpinned.json` and `listings/` with the Python
+  standard library alone, re-implementing the listing rules, the ordered checks and the
+  report from the SPEC text, and reading which capabilities decode against a parameter
+  set from the published registry export (`../registry/capabilities.json`). It shares no
+  code with wai-rs.
diff --git a/wai/pin-conformance/pin_verify.py b/wai/pin-conformance/pin_verify.py
index 600945515..00e5b97a5 100644
--- a/wai/pin-conformance/pin_verify.py
+++ b/wai/pin-conformance/pin_verify.py
@@ -8,13 +8,18 @@ library only, and checks that
   - every `listings/<name>.sha256` is byte for byte the listing of `sets/<name>/`
     (regular files, names that begin with `.` skipped), and every file under
     `listings/bad/` breaks a listing rule;
-  - every case in `cases.json` gets its declared reason.
+  - every case in `cases.json` gets its declared reason;
+  - every manifest in `unpinned.json` gets its declared report (SPEC.md §3.1,
+    Pin by default): `prior_unpinned` where it pins no parameter set and a
+    capability it decodes through is registered as decoding against one.
 
   python3 pin-conformance/pin_verify.py [corpus-dir]
 
 The registry facts it needs are restated here from SPEC.md §5 and §3.1: the
 registered prior forms, the forms the reference sink reads, and the capabilities
-registered with a one-file parameter set.
+registered with a one-file parameter set. Which capabilities decode against a
+parameter set it reads from the published registry export,
+`../registry/capabilities.json`: those of the four classes SPEC.md §3.1 names.
 """
 import hashlib
 import json
@@ -28,6 +33,10 @@ ONE_FILE = {
     'wai.neural.int_synth', 'wai.neural.int_mlicpp', 'wai.neural.encodec32', 'wai.neural.dac',
     'wai.neural.mimi', 'wai.neural.wavtokenizer', 'wai.neural.snac', 'wai.neural.video_bmshj2018',
 }
+# SPEC.md §3.1: the registry classes whose capabilities decode against a parameter set.
+SET_CLASSES = {'neural-int-exact', 'neural-entropy-exact', 'neural-float', 'generative-exact'}
+# SPEC.md §3.1: the one capability whose payload names its own parameter set.
+NAMES_ITS_SET = 'wai.generate.spectral'
 HEX64 = re.compile(r'^[0-9a-f]{64}$')
 SEGMENT = re.compile(r'^[A-Za-z0-9_-]([A-Za-z0-9_.-]*[A-Za-z0-9_-])?$')
 
@@ -145,6 +154,29 @@ def load(corpus, held):
     return out
 
 
+def registry_classes():
+    path = os.path.join(os.path.dirname(os.path.abspath(__file__)), '..', 'registry', 'capabilities.json')
+    with open(path, encoding='utf-8') as f:
+        return {c['capability']: c['replicate'] for c in json.load(f)['capabilities']}
+
+
+def pin_by_default(cap, classes):
+    """Whether an encoder SHOULD pin the set a payload of `cap` was coded against."""
+    return classes.get(cap) in SET_CLASSES and cap != NAMES_ITS_SET
+
+
+def unpinned(manifest, classes):
+    """The report of SPEC.md §3.1, Pin by default: a WAI1 manifest's, or each WAI2 entry's."""
+    if manifest['wai'] == '1.0':
+        if manifest['conditioning'].get('prior') is not None:
+            return 'none'
+        mr = manifest['model_requirement']
+        caps = [mr['capability']] + ([mr['fallback']] if mr.get('fallback') is not None else [])
+        return 'prior_unpinned' if any(pin_by_default(c, classes) for c in caps) else 'none'
+    return [i for i, e in enumerate(manifest['renditions'])
+            if e.get('prior') is None and pin_by_default(e['capability'], classes)]
+
+
 def main():
     corpus = sys.argv[1] if len(sys.argv) > 1 else os.path.dirname(os.path.abspath(__file__))
     fails, n = [], 0
@@ -174,6 +206,13 @@ def main():
         got = verify(c['capability'], pin, load(corpus, c['held']), c.get('readable', READABLE))
         if got != c['expect']:
             fails.append(f"cases.json {c['name']}: expected {c['expect']}, got {got}")
+    classes = registry_classes()
+    with open(os.path.join(corpus, 'unpinned.json')) as f:
+        for c in json.load(f)['cases']:
+            n += 1
+            got = unpinned(c['manifest'], classes)
+            if got != c['expect']:
+                fails.append(f"unpinned.json {c['name']}: expected {c['expect']}, got {got}")
     for f in fails:
         print('FAIL', f)
     print(f'pin_verify.py: {n - len(fails)} of {n} checks as declared')
diff --git a/wai/pin-conformance/unpinned.json b/wai/pin-conformance/unpinned.json
new file mode 100644
index 000000000..5c61d6af6
--- /dev/null
+++ b/wai/pin-conformance/unpinned.json
@@ -0,0 +1,265 @@
+{
+ "about": "SPEC.md \u00a73.1, Pin by default: the report a verifier makes about a manifest that pins no parameter set where a capability it decodes through is registered as decoding against one (\"prior_unpinned\"), or \"none\". It is informative: every manifest here conforms but the one that claims replicate over an integer-exact capability unpinned, which static.json refuses. For a WAI2 manifest, the index of each entry it is made about.",
+ "cases": [
+  {
+   "name": "integer_exact_improve",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "improve",
+    "model_requirement": {
+     "capability": "wai.neural.int_mlicpp"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "prior_unpinned"
+  },
+  {
+   "name": "integer_exact_replicate",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "replicate",
+    "model_requirement": {
+     "capability": "wai.neural.int_synth"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "prior_unpinned"
+  },
+  {
+   "name": "float_improve",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "improve",
+    "model_requirement": {
+     "capability": "wai.neural.bmshj2018"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "prior_unpinned"
+  },
+  {
+   "name": "entropy_exact_improve",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "improve",
+    "model_requirement": {
+     "capability": "wai.image.jpegai"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "prior_unpinned"
+  },
+  {
+   "name": "generator_with_a_set_create",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "create",
+    "model_requirement": {
+     "capability": "wai.avatar.face"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "prior_unpinned"
+  },
+  {
+   "name": "null_prior_is_absent",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "improve",
+    "model_requirement": {
+     "capability": "wai.neural.int_mlicpp"
+    },
+    "conditioning": {
+     "kind": "k",
+     "prior": null
+    },
+    "target": null
+   },
+   "expect": "prior_unpinned"
+  },
+  {
+   "name": "the_fallback_decodes_against_a_set",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "improve",
+    "model_requirement": {
+     "capability": "wai.image.png",
+     "fallback": "wai.neural.bmshj2018"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "prior_unpinned"
+  },
+  {
+   "name": "pinned",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "replicate",
+    "model_requirement": {
+     "capability": "wai.neural.int_mlicpp"
+    },
+    "conditioning": {
+     "kind": "k",
+     "prior": {
+      "id": "int_mlicpp/synthetic-s20260929",
+      "sha256": "7ef44f3b1fc4a2df6c3e3b68460f595cae3d31cb7f9018699e29e738340272a4",
+      "determinism": "int-rans/p16+int-mlicpp/i64"
+     }
+    },
+    "target": null
+   },
+   "expect": "none"
+  },
+  {
+   "name": "standard_codec",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "replicate",
+    "model_requirement": {
+     "capability": "wai.image.png"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "none"
+  },
+  {
+   "name": "generator_names_its_model",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "create",
+    "model_requirement": {
+     "capability": "wai.generate.spectral"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "none"
+  },
+  {
+   "name": "unregistered_capability",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "improve",
+    "model_requirement": {
+     "capability": "wai.neural.future"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "none"
+  },
+  {
+   "name": "reads_no_set",
+   "manifest": {
+    "wai": "1.0",
+    "media": "image",
+    "intent": "replicate",
+    "model_requirement": {
+     "capability": "wai.video.int_motion"
+    },
+    "conditioning": {
+     "kind": "k"
+    },
+    "target": null
+   },
+   "expect": "none"
+  },
+  {
+   "name": "wai2_each_entry",
+   "manifest": {
+    "wai": "1.1",
+    "media": "image",
+    "intent": "improve",
+    "renditions": [
+     {
+      "capability": "wai.image.png",
+      "kind": "png"
+     },
+     {
+      "capability": "wai.neural.int_mlicpp",
+      "kind": "int_mlicpp"
+     },
+     {
+      "capability": "wai.neural.int_mlicpp",
+      "kind": "int_mlicpp",
+      "prior": {
+       "id": "int_mlicpp/synthetic-s20260929",
+       "sha256": "7ef44f3b1fc4a2df6c3e3b68460f595cae3d31cb7f9018699e29e738340272a4",
+       "determinism": "int-rans/p16+int-mlicpp/i64"
+      }
+     },
+     {
+      "capability": "wai.neural.bmshj2018",
+      "kind": "bmshj2018"
+     }
+    ],
+    "target": null
+   },
+   "expect": [
+    1,
+    3
+   ]
+  },
+  {
+   "name": "wai2_none",
+   "manifest": {
+    "wai": "1.1",
+    "media": "image",
+    "intent": "replicate",
+    "renditions": [
+     {
+      "capability": "wai.neural.int_mlicpp",
+      "kind": "int_mlicpp",
+      "prior": {
+       "id": "int_mlicpp/synthetic-s20260929",
+       "sha256": "7ef44f3b1fc4a2df6c3e3b68460f595cae3d31cb7f9018699e29e738340272a4",
+       "determinism": "int-rans/p16+int-mlicpp/i64"
+      }
+     },
+     {
+      "capability": "wai.image.png",
+      "kind": "png"
+     }
+    ],
+    "target": null
+   },
+   "expect": []
+  }
+ ]
+}
diff --git a/wai/wai-rs/src/bin/wai.rs b/wai/wai-rs/src/bin/wai.rs
index 9192f43d6..58ce08b65 100644
--- a/wai/wai-rs/src/bin/wai.rs
+++ b/wai/wai-rs/src/bin/wai.rs
@@ -424,6 +424,10 @@ fn inspect(path: &str) -> Result<(), String> {
             Ok(()) => println!("    pin check:       \u{2713} no static finding (SPEC \u{a7}3.1, \u{a7}7)"),
             Err(f) => println!("    pin check:       \u{2717} {f}"),
         }
+        // Informative: the envelope conforms without the pin (SPEC §3.1, Pin by default).
+        if let Some(f) = w.check_unpinned() {
+            println!("    note:            {f}");
+        }
         match w.refines() {
             None => {}
             Some(Ok(r)) => println!("  refines:           base {}…  layer {}{}", &r.base[..16], r.layer,
@@ -541,6 +545,9 @@ fn inspect(path: &str) -> Result<(), String> {
             Ok(()) => println!("  pin check:         \u{2713} no static finding (SPEC \u{a7}3.1, \u{a7}7)"),
             Err((i, f)) => println!("  pin check:         \u{2717} [{i}] {f}"),
         }
+        for (i, f) in m.check_unpinned() {
+            println!("  note:              [{i}] {f}");
+        }
         match m.check_component_roles() {
             Ok(()) => println!("  component roles:   \u{2713} conform (SPEC \u{a7}5, \u{a7}7.1)"),
             Err(e) => println!("  component roles:   \u{2717} {e}"),
@@ -598,9 +605,11 @@ fn license_str(cap: &str) -> &'static str {
 /// AC-4 frame — WAI carries it, the sink decodes it). The bytes survive
 /// byte-identically (§2): `wrap` then `unwrap` returns the exact input.
 ///
-/// The envelope pins no prior. Its intent is `--intent` when given, else
-/// `replicate`, and `replicate` is written only where a sink accepts it unpinned
-/// (SPEC §7, `codecs::unpinned_envelope_intent`): over the standard classes. Over
+/// The envelope pins no prior, and where its capability decodes against a parameter
+/// set `wrap` says so (`prior_unpinned`, SPEC §3.1, *Pin by default*). Its intent is
+/// `--intent` when given, else `replicate`, and `replicate` is written only where a
+/// sink accepts it unpinned (SPEC §7, `codecs::unpinned_envelope_intent`): over the
+/// standard classes. Over
 /// any other capability (`wai.image.jpegai`, a float or integer-exact neural codec,
 /// a generator, a behavioral path) `wrap` refuses and writes nothing unless another
 /// intent is given.
@@ -685,11 +694,14 @@ fn wrap(cap: &str, input: &str, out: &str, fallback: Option<&str>,
     out::write(out, &bytes).map_err(|e| format!("{out}: {e}"))?;
     let n = bytes.len();
     let tier = wai::codecs::capability_tier(cap).map(|t| format!("{t:?}")).unwrap_or_else(|| "?".into());
+    let mut lines = vec![
+        format!("wrapped → {out}  ({n} B envelope, {n_in} B payload)"),
+        format!("  capability {cap}  ·  media {}  ·  {tier}  ·  {}  ·  intent {intent}", media_for(cap), license_str(cap)),
+    ];
+    // `wrap` pins no prior: where SPEC §3.1 asks an encoder to pin one, it says so.
+    lines.extend(envelope.check_unpinned().map(|f| format!("  note {f}")));
     Ok(Wrapped {
-        lines: vec![
-            format!("wrapped → {out}  ({n} B envelope, {n_in} B payload)"),
-            format!("  capability {cap}  ·  media {}  ·  {tier}  ·  {}  ·  intent {intent}", media_for(cap), license_str(cap)),
-        ],
+        lines,
         facts: serde_json::json!({
             "output": out, "format": "WAI1", "envelope_bytes": n, "intent": intent, "media": media_for(cap),
             "components": [{"index": 0, "capability": cap, "role": "rendition", "bytes": n_in, "licence": license_str(cap), "tier": tier}],
@@ -801,6 +813,8 @@ fn wrap_with_companions(cap: &str, payload: Vec<u8>, out: &str, fallback: Option
     let mut lines = vec![format!("wrapped → {out}  ({n} B WAI2 envelope, {} components, intent {intent})",
                                  envelope.renditions.len())];
     lines.extend(component_lines(&envelope));
+    // `wrap` pins no prior: where SPEC §3.1 asks an encoder to pin one, it says so.
+    lines.extend(envelope.check_unpinned().into_iter().map(|(i, f)| format!("  note [{i}] {f}")));
     let components: Vec<serde_json::Value> = envelope.manifest.renditions.iter().zip(envelope.renditions.iter()).enumerate()
         .map(|(i, (meta, p))| serde_json::json!({
             "index": i, "capability": meta.capability, "bytes": p.len(), "licence": license_str(&meta.capability),
diff --git a/wai/wai-rs/src/bin/wai_cli/dx/verify.rs b/wai/wai-rs/src/bin/wai_cli/dx/verify.rs
index 3672eb232..1be238415 100644
--- a/wai/wai-rs/src/bin/wai_cli/dx/verify.rs
+++ b/wai/wai-rs/src/bin/wai_cli/dx/verify.rs
@@ -12,6 +12,11 @@
 //! conform`) under each path: its signature, and that its summary and verdict are what its
 //! cases say.
 //!
+//! And every envelope (`*.wai`) under each path: what a verifier finds about its pins
+//! without holding a parameter set (SPEC §3.1, §7), which makes it non-conforming, and the
+//! informative report `prior_unpinned` where it pins nothing that a capability it decodes
+//! through is registered as decoding against (SPEC §3.1, *Pin by default*), which does not.
+//!
 //! A fleet report's bundle (`*.wai-fleet.json`, extensions/fleet.md §7) under a path is verified
 //! too, in a build with `fleet`: the report recomputed from its enrolments and records.
 
@@ -36,7 +41,7 @@ pub fn main(args: &[String]) -> Result<bool, OpError> {
     fn walk(p: &Path, out: &mut Vec<PathBuf>, bundles: &mut Vec<PathBuf>, depth: usize) -> Result<(), String> {
         if p.is_file() {
             let n = p.to_string_lossy();
-            if n.ends_with(".wai.receipts") || n.ends_with(super::conform::record::SUFFIX) {
+            if n.ends_with(".wai.receipts") || n.ends_with(super::conform::record::SUFFIX) || n.ends_with(".wai") {
                 out.push(p.to_path_buf());
             } else if p.to_string_lossy().ends_with(".wai-fleet.json") {
                 bundles.push(p.to_path_buf());
@@ -65,17 +70,20 @@ pub fn main(args: &[String]) -> Result<bool, OpError> {
     }
     if indexes.is_empty() && bundles.is_empty() {
         return Err(OpError::from(format!(
-            "no receipt index (*.wai.receipts), conformance result (*.wai-conform.json) or fleet bundle (*.wai-fleet.json) under {}",
+            "no receipt index (*.wai.receipts), conformance result (*.wai-conform.json), fleet bundle (*.wai-fleet.json) or envelope (*.wai) under {}",
             paths.join(", ")
         )));
     }
     let mut all = true;
-    let (mut n_ix, mut n_rec) = (0, 0);
+    let (mut n_ix, mut n_rec, mut n_env) = (0, 0, 0);
     for ix in &indexes {
         println!("{}", ix.display());
         let (ok, lines) = if ix.to_string_lossy().ends_with(super::conform::record::SUFFIX) {
             n_rec += 1;
             super::conform::record::check(&std::fs::read(ix).map_err(|e| format!("{}: {e}", ix.display()))?)
+        } else if ix.to_string_lossy().ends_with(".wai") {
+            n_env += 1;
+            check_envelope(&std::fs::read(ix).map_err(|e| format!("{}: {e}", ix.display()))?)
         } else {
             n_ix += 1;
             check_index_file(ix)?
@@ -101,6 +109,9 @@ pub fn main(args: &[String]) -> Result<bool, OpError> {
     if n_rec > 0 {
         what.push(format!("{} conformance result{}", n_rec, if n_rec == 1 { "" } else { "s" }));
     }
+    if n_env > 0 {
+        what.push(format!("{} envelope{}", n_env, if n_env == 1 { "" } else { "s" }));
+    }
     if m > 0 {
         what.push(format!("{m} fleet bundle{}", if m == 1 { "" } else { "s" }));
     }
@@ -108,6 +119,55 @@ pub fn main(args: &[String]) -> Result<bool, OpError> {
     Ok(all)
 }
 
+/// One envelope's pins, as a verifier holding no parameter set reads them (SPEC §3.1, §7):
+/// whether it conforms on that count — a static finding makes it non-conforming — and a line
+/// for each thing found, the informative `prior_unpinned` (SPEC §3.1, *Pin by default*)
+/// among them, which does not.
+#[cfg(feature = "quickstart")]
+pub(crate) fn check_envelope(bytes: &[u8]) -> (bool, Vec<String>) {
+    use wai::container::{Wai, WaiMulti};
+    let pin = |p: Option<&wai::container::Prior>| p.map(|p| format!("pinned {}", wai::pin::describe_pin(p))).unwrap_or_else(|| "no pin".into());
+    match bytes.get(..4) {
+        Some(b"WAI1") => match Wai::from_bytes(bytes) {
+            Err(e) => (false, vec![format!("FAIL not an envelope that parses: {e}")]),
+            Ok(w) => {
+                let mut lines = Vec::new();
+                let ok = match w.check_pin() {
+                    Ok(()) => {
+                        lines.push(format!("ok   WAI1 {}, {}: no static finding on its pin (SPEC \u{a7}3.1, \u{a7}7)", w.capability(), pin(w.prior())));
+                        true
+                    }
+                    Err(f) => {
+                        lines.push(format!("FAIL WAI1 {}: {f}", w.capability()));
+                        false
+                    }
+                };
+                lines.extend(w.check_unpinned().map(|f| format!("note {f}")));
+                (ok, lines)
+            }
+        },
+        Some(b"WAI2") => match WaiMulti::from_bytes(bytes) {
+            Err(e) => (false, vec![format!("FAIL not an envelope that parses: {e}")]),
+            Ok(m) => {
+                let mut lines = Vec::new();
+                let ok = match m.check_pins() {
+                    Ok(()) => {
+                        lines.push(format!("ok   WAI2, {} entries: no static finding on their pins (SPEC \u{a7}3.1, \u{a7}7)", m.manifest.renditions.len()));
+                        true
+                    }
+                    Err((i, f)) => {
+                        lines.push(format!("FAIL WAI2 entry [{i}] {}: {f}", m.manifest.renditions[i].capability));
+                        false
+                    }
+                };
+                lines.extend(m.check_unpinned().into_iter().map(|(i, f)| format!("note [{i}] {f}")));
+                (ok, lines)
+            }
+        },
+        _ => (false, vec!["FAIL not an envelope: it begins with neither WAI1 nor WAI2".into()]),
+    }
+}
+
 /// One fleet report bundle (extensions/fleet.md §7): whether it verified, and a line for its
 /// report and for each node, room and scope of energy.
 #[cfg(all(feature = "quickstart", feature = "fleet"))]
diff --git a/wai/wai-rs/src/bin/wai_pin_vectors.rs b/wai/wai-rs/src/bin/wai_pin_vectors.rs
index ed105e8c7..591de5de2 100644
--- a/wai/wai-rs/src/bin/wai_pin_vectors.rs
+++ b/wai/wai-rs/src/bin/wai_pin_vectors.rs
@@ -15,11 +15,14 @@
 //!   first component-role finding.
 //! - `static.json`: the first finding a verifier makes without holding a parameter
 //!   set ([`wai::container::Wai::check_pin`], [`wai::container::WaiMulti::check_pins`]).
+//! - `unpinned.json`: the informative report `prior_unpinned` (SPEC §3.1, *Pin by
+//!   default*; [`wai::container::Wai::check_unpinned`],
+//!   [`wai::container::WaiMulti::check_unpinned`]).
 //! - `listings/`: each `<name>.sha256` is the listing of `sets/<name>/`;
 //!   `listings/bad/*` each break a listing rule (`set_malformed`).
 //!
-//! `pin-conformance/pin_verify.py` checks `cases.json` and `listings/` with code that
-//! shares nothing with this crate.
+//! `pin-conformance/pin_verify.py` checks `cases.json`, `unpinned.json` and `listings/`
+//! with code that shares nothing with this crate.
 
 use std::collections::BTreeMap;
 use std::path::Path;
@@ -171,6 +174,22 @@ fn verify(dir: &Path) -> i32 {
         Err(e) => fail(e),
     }
 
+    // unpinned.json
+    match read_json(&dir.join("unpinned.json")) {
+        Ok(doc) => {
+            for c in doc["cases"].as_array().into_iter().flatten() {
+                checked += 1;
+                let name = c["name"].as_str().unwrap_or("?");
+                match run_unpinned(&c["manifest"]) {
+                    Ok(got) if got == c["expect"] => {}
+                    Ok(got) => fail(format!("unpinned.json {name}: expected {}, got {got}", c["expect"])),
+                    Err(e) => fail(format!("unpinned.json {name}: {e}")),
+                }
+            }
+        }
+        Err(e) => fail(e),
+    }
+
     if failures.is_empty() {
         println!("pin-conformance: {checked} checks, all as declared");
         0
@@ -285,6 +304,25 @@ fn run_selection(dir: &Path, c: &Value) -> Result<Value, String> {
     Ok(Value::Object(out))
 }
 
+/// The report `prior_unpinned` (SPEC §3.1, *Pin by default*): for a `WAI1` manifest,
+/// `prior_unpinned` or `none`; for a `WAI2` manifest, the index of every entry it is made
+/// about.
+fn run_unpinned(manifest: &Value) -> Result<Value, String> {
+    match manifest["wai"].as_str() {
+        Some("1.0") => {
+            let m: Manifest = serde_json::from_value(manifest.clone()).map_err(|e| format!("manifest: {e}"))?;
+            Ok(Wai::new(m, Vec::new()).check_unpinned().map_or("none", |f| f.code).into())
+        }
+        Some("1.1") => {
+            let m: ManifestV2 = serde_json::from_value(manifest.clone()).map_err(|e| format!("manifest: {e}"))?;
+            let n = m.renditions.len();
+            let w = WaiMulti::from_parts(m, vec![Vec::new(); n]).map_err(|e| e.to_string())?;
+            Ok(serde_json::json!(w.check_unpinned().iter().map(|(i, _)| *i).collect::<Vec<_>>()))
+        }
+        other => Err(format!("unknown wai version {other:?}")),
+    }
+}
+
 fn run_static(manifest: &Value) -> Result<Value, String> {
     match manifest["wai"].as_str() {
         Some("1.0") => {
diff --git a/wai/wai-rs/src/catalogue.rs b/wai/wai-rs/src/catalogue.rs
index 34596a962..2b82c9396 100644
--- a/wai/wai-rs/src/catalogue.rs
+++ b/wai/wai-rs/src/catalogue.rs
@@ -406,11 +406,11 @@ pub const OPS: &[Op] = &[
     },
     Op {
         id: "receipt.verify",
-        summary: "Every receipt index (*.wai.receipts) under each path, its group receipts and claims checked offline, and every conformance result and fleet report bundle there",
+        summary: "Every receipt index (*.wai.receipts) under each path, its group receipts and claims checked offline, and every conformance result, fleet report bundle and envelope there",
         cli: Some("verify"),
-        usage: "verify <path>...   the receipts (receipt indexes *.wai.receipts), conformance results (*.wai-conform.json)
-    and fleet report bundles (*.wai-fleet.json) under each path, checked offline",
-        input: &[pos("paths", Ty::Strings, true, "Files or directories holding receipt indexes, conformance results or fleet report bundles.")],
+        usage: "verify <path>...   the receipts (receipt indexes *.wai.receipts), conformance results (*.wai-conform.json),
+    fleet report bundles (*.wai-fleet.json) and envelopes' pins (*.wai) under each path, checked offline",
+        input: &[pos("paths", Ty::Strings, true, "Files or directories holding receipt indexes, conformance results, fleet report bundles or envelopes.")],
         needs: Needs::All(&["quickstart"]),
         mcp: true,
         ..OP
diff --git a/wai/wai-rs/src/codecs/mod.rs b/wai/wai-rs/src/codecs/mod.rs
index e32c4e4b3..050af46e5 100644
--- a/wai/wai-rs/src/codecs/mod.rs
+++ b/wai/wai-rs/src/codecs/mod.rs
@@ -1649,6 +1649,17 @@ pub fn decodes_against_parameter_set(cap: &str) -> bool {
     )
 }
 
+/// Whether an encoder SHOULD pin the parameter set a payload of `cap` was coded against
+/// (SPEC §3.1, *Pin by default*): `cap` is registered as decoding against one
+/// ([`decodes_against_parameter_set`]), and its payload does not name its own set, as the
+/// payload of `wai.generate.spectral` names its model. `false` for an unregistered
+/// capability: this revision does not know what it decodes against. An envelope that pins
+/// nothing where this holds still conforms in `1.x`, and a verifier reports it
+/// (`prior_unpinned`).
+pub fn pin_by_default(cap: &str) -> bool {
+    decodes_against_parameter_set(cap) && cap != CAP_GENERATE_SPECTRAL
+}
+
 /// The registered shape of a capability's parameter set (SPEC §5): one file, or
 /// several. A form-less pin is admissible only over a one-file set (SPEC §3.1,
 /// `form_required`); a set of several files is pinned through its listing
diff --git a/wai/wai-rs/src/container.rs b/wai/wai-rs/src/container.rs
index dd1ef9378..fa9bf9a79 100644
--- a/wai/wai-rs/src/container.rs
+++ b/wai/wai-rs/src/container.rs
@@ -697,7 +697,8 @@ pub fn replicate_is_sound(intent: &str, prior: Option<&Prior>) -> bool {
 pub struct PinFinding {
     /// The reason: `prior_not_applicable`, `pin_malformed`, `tier_unregistered`,
     /// `supersedes_malformed`, `form_not_registered`, `form_required`,
-    /// `tier_exceeds_class`, or a `replicate_*` reason.
+    /// `tier_exceeds_class`, or a `replicate_*` reason; or, from `check_unpinned`, the
+    /// informative report `prior_unpinned` ([`PRIOR_UNPINNED`]).
     pub code: &'static str,
     pub detail: String,
 }
@@ -722,6 +723,28 @@ pub const PIN_FINDING_CODES: &[&str] = &[
     "replicate_form_not_exact",
 ];
 
+/// The report a verifier makes about an envelope that pins no parameter set where a
+/// capability it decodes through is registered as decoding against one (SPEC §3.1, *Pin by
+/// default*; [`Wai::check_unpinned`], [`WaiMulti::check_unpinned`]). It is informative:
+/// such an envelope conforms in `1.x`, so the code is not one of [`PIN_FINDING_CODES`].
+pub const PRIOR_UNPINNED: &str = "prior_unpinned";
+
+/// [`PRIOR_UNPINNED`] for a pin `prior` over `caps`: the first capability of `caps` whose
+/// set an encoder SHOULD pin (`codecs::pin_by_default`), when `prior` is absent.
+fn unpinned_report(caps: &[&str], prior: Option<&Prior>) -> Option<PinFinding> {
+    if prior.is_some() {
+        return None;
+    }
+    let c = caps.iter().find(|c| crate::codecs::pin_by_default(c))?;
+    Some(PinFinding {
+        code: PRIOR_UNPINNED,
+        detail: format!(
+            "{c} decodes against a parameter set and the envelope pins none, so a sink decodes it against \
+             whatever set it holds; an encoder SHOULD pin the set the payload was coded against (SPEC \u{a7}3.1)"
+        ),
+    })
+}
+
 /// The static pin findings for a pin that would apply to the capabilities `caps`
 /// (the first is the one decoded through first), under `intent`.
 fn static_pin_findings(caps: &[&str], prior: Option<&Prior>, intent: &str) -> Result<(), PinFinding> {
@@ -1441,6 +1464,14 @@ impl Wai {
         static_pin_findings(&caps, self.prior(), self.intent())
     }
 
+    /// [`PRIOR_UNPINNED`] when this envelope pins no parameter set and its capability, or
+    /// else its fallback, is one whose set an encoder SHOULD pin (SPEC §3.1, *Pin by
+    /// default*). Informative: it is reported whatever [`Self::check_pin`] finds.
+    pub fn check_unpinned(&self) -> Option<PinFinding> {
+        let caps: Vec<&str> = [Some(self.capability()), self.fallback()].into_iter().flatten().collect();
+        unpinned_report(&caps, self.prior())
+    }
+
     pub fn media(&self) -> &str {
         &self.manifest.media
     }
@@ -3162,6 +3193,18 @@ impl WaiMulti {
         Ok(())
     }
 
+    /// [`PRIOR_UNPINNED`] for each entry, with its index, that pins no parameter set while
+    /// its capability is one whose set an encoder SHOULD pin (SPEC §3.1, *Pin by default*),
+    /// in table order. Informative, as [`Wai::check_unpinned`].
+    pub fn check_unpinned(&self) -> Vec<(usize, PinFinding)> {
+        self.manifest
+            .renditions
+            .iter()
+            .enumerate()
+            .filter_map(|(i, m)| unpinned_report(&[m.capability.as_str()], m.prior.as_ref()).map(|f| (i, f)))
+            .collect()
+    }
+
 
     /// The refinements of the entry at `index`, in table order: each entry with
     /// `role: "refinement"` whose `refines.base` names that entry's `id`, and whose
@@ -4717,6 +4760,50 @@ pub(crate) mod tests {
         }
     }
 
+    /// SPEC §3.1, *Pin by default*: an envelope that pins nothing where its capability, or
+    /// its fallback, decodes against a parameter set is reported `prior_unpinned`, whatever
+    /// its intent and whatever else is found; a pinned one, one over a capability that reads
+    /// no set, the generator whose payload names its model, and an unregistered capability
+    /// are not.
+    #[test]
+    fn an_unpinned_envelope_is_reported_where_it_decodes_against_a_set() {
+        let on = |cap: &str, fb: Option<&str>, intent: &str, p: Option<Prior>| {
+            let mut c = Conditioning::new("k");
+            c.prior = p;
+            let r = match fb {
+                Some(f) => ModelRequirement::new(cap).with_fallback(f),
+                None => ModelRequirement::new(cap),
+            };
+            Wai::new(Manifest::new("image", intent, r, c), vec![])
+        };
+        let report = |w: &Wai| w.check_unpinned().map(|f| f.code);
+        let pin = Prior::new("m", MODEL_SHA, "int");
+        for intent in ["improve", "create", "replicate"] {
+            assert_eq!(report(&on("wai.neural.int_mlicpp", None, intent, None)), Some(PRIOR_UNPINNED), "{intent}");
+            assert_eq!(report(&on("wai.neural.bmshj2018", None, intent, None)), Some(PRIOR_UNPINNED), "{intent}");
+        }
+        // Reported beside the finding that makes the envelope non-conforming.
+        let unsound = on("wai.neural.int_synth", None, "replicate", None);
+        assert_eq!(unsound.check_pin().err().map(|f| f.code), Some("replicate_unpinned"));
+        assert_eq!(report(&unsound), Some(PRIOR_UNPINNED));
+        assert_eq!(report(&on("wai.neural.int_mlicpp", None, "improve", Some(pin.clone()))), None);
+        assert_eq!(report(&on("wai.image.png", None, "replicate", None)), None);
+        assert_eq!(report(&on("wai.image.png", Some("wai.neural.bmshj2018"), "improve", None)), Some(PRIOR_UNPINNED),
+                   "the fallback decodes the same payload against a set");
+        assert_eq!(report(&on("wai.generate.spectral", None, "create", None)), None, "its payload names its model");
+        assert_eq!(report(&on("wai.neural.future", None, "improve", None)), None, "unregistered: its set is not known");
+        assert!(!PIN_FINDING_CODES.contains(&PRIOR_UNPINNED), "informative, not a finding of non-conformance");
+
+        let m = WaiMulti::new("image", "improve", Value::Null, vec![
+            Rendition::media("wai.image.png", "png", b"PNG".to_vec()),
+            Rendition::media("wai.neural.int_mlicpp", "int_mlicpp", b"M".to_vec()),
+            Rendition::media("wai.neural.int_mlicpp", "int_mlicpp", b"N".to_vec()).with_prior(pin),
+            Rendition::media("wai.neural.bmshj2018", "bmshj2018", b"B".to_vec()),
+        ]);
+        let entries: Vec<(usize, &str)> = m.check_unpinned().iter().map(|(i, f)| (*i, f.code)).collect();
+        assert_eq!(entries, [(1, PRIOR_UNPINNED), (3, PRIOR_UNPINNED)]);
+    }
+
     // ---- component identity (SPEC §7.1) ----
 
     #[test]
diff --git a/wai/wai-rs/src/int_payload/mod.rs b/wai/wai-rs/src/int_payload/mod.rs
index a3504e806..9042e9907 100644
--- a/wai/wai-rs/src/int_payload/mod.rs
+++ b/wai/wai-rs/src/int_payload/mod.rs
@@ -184,6 +184,23 @@ impl<'a> IntParams<'a> {
         }
     }
 
+    /// The parameter set `cap`'s decode reads from these parameters, named as a pin over
+    /// exactly those bytes names it (SPEC §3.1): for a capability registered with a one-file
+    /// set, the SHA-256 of that file, which a form-less pin names; for another, the SHA-256
+    /// of the listing of the members read, which a pin in the form `wai.prior.set` over them
+    /// names. `None` where these parameters hold no set, or none of the capability's
+    /// members. A record of an unpinned decode names its set by it (jwp-receipts §2.4.12).
+    pub fn read_pin(&self, cap: &str) -> Option<String> {
+        let read = self.read(cap);
+        if self.is_none() || read.is_empty() {
+            return None;
+        }
+        if crate::container::pin_shape(cap) == crate::codecs::ParameterSetShape::OneFile {
+            return set_members(cap).first().and_then(|m| read.get(*m)).map(|b| crate::pin::sha256_hex(b));
+        }
+        self.read_digest(cap).map(|d| d.iter().map(|b| format!("{b:02x}")).collect())
+    }
+
     /// Whether a pin verified these parameters.
     pub fn is_pinned(&self) -> bool {
         matches!(self.0, Params::Verified(_) | Params::Held { pinned: true, .. })
diff --git a/wai/wai-rs/src/stage_measure/decode.rs b/wai/wai-rs/src/stage_measure/decode.rs
index 31fda6587..fafdc2e04 100644
--- a/wai/wai-rs/src/stage_measure/decode.rs
+++ b/wai/wai-rs/src/stage_measure/decode.rs
@@ -25,8 +25,10 @@ fn invalid(s: impl Into<String>) -> MeasureError {
     MeasureError::Invalid(s.into())
 }
 
-/// A base decoded: its identity as the sink holds it, its units, and the SHA-256 of
-/// the pin its decode verified (`None` when unpinned).
+/// A base decoded: its identity as the sink holds it, its units, and the parameter set its
+/// decode read, named as a pin names it: the SHA-256 of the pin it verified, or, unpinned,
+/// the digest a pin over the bytes it read would name (`staged::BaseDecode::set`); `None`
+/// for a capability that reads no parameter set.
 #[derive(Debug, Clone, PartialEq, Eq)]
 pub struct DecodedBase {
     pub info: BaseInfo,
@@ -67,7 +69,7 @@ fn checked(b: &TitleBase, d: crate::staged::BaseDecode) -> Result<DecodedBase, M
             return Err(invalid(format!("{at}: the decode gives (via, fallback, tier, pin) {got:?}; the case states {want:?}")));
         }
     }
-    Ok(DecodedBase { info, units: d.units, pin: d.pin.map(|p| p.sha256) })
+    Ok(DecodedBase { info, units: d.units, pin: d.set })
 }
 
 /// The chain configuration a replay runs under: the title's chain budget, its cost
@@ -96,7 +98,8 @@ pub struct StageEvent<'a> {
     pub layers: u16,
     /// The capability that produced it: the base's, or the layer's.
     pub capability: &'a str,
-    /// The pin the base's decode verified, for stage 0.
+    /// The parameter set the base's decode read, as a pin names it ([`DecodedBase`]), for
+    /// stage 0.
     pub pin: Option<&'a str>,
     /// The content hash of the envelope that produced it: the base's, or the layer's.
     pub object: [u8; 32],
diff --git a/wai/wai-rs/src/stage_measure/session.rs b/wai/wai-rs/src/stage_measure/session.rs
index e56a4aeed..e0e8803f5 100644
--- a/wai/wai-rs/src/stage_measure/session.rs
+++ b/wai/wai-rs/src/stage_measure/session.rs
@@ -248,6 +248,46 @@ mod tests {
         }
     }
 
+    /// SPEC §3.1, *Pin by default*: a base its encoder left unpinned is still decoded against
+    /// the set the sink holds, and the session's decoder entry names that set by the digest a
+    /// pin over the bytes read names (jwp-receipts §2.4.12), with the tier `none`. The
+    /// fixture's own pin, a listing of the same files, is that digest; the pinned base
+    /// names it as before, with the tier its pin declares.
+    #[test]
+    fn an_unpinned_decode_names_the_set_it_read() {
+        let wai = std::path::Path::new(env!("CARGO_MANIFEST_DIR")).join("..");
+        let key = SigningKey::from_bytes(&[0x44; 32]);
+        let host = SessionHost { profile: "native-ffi".into(), target: None, build: BuildRecord::of_this_build(), threads: 1 };
+        let hex = |b: &[u8; 32]| b.iter().map(|x| format!("{x:02x}")).collect::<String>();
+        let entry_of = |t: &Title| {
+            let e = super::super::run::evaluate(t).unwrap();
+            let (r, entry) = recorder_of(t, &e, host.clone(), [3; 32]).unwrap();
+            let s = r.finish(entry, &key, "did:example:sink", 0, None, "2026-10-06T12:00:00Z").unwrap();
+            s.claim.verify().unwrap();
+            let ClaimBody::ReconstructSession(b) = s.claim.body else { panic!("a session") };
+            let d = b.decoders[0].clone();
+            (d.pin.as_ref().map(hex), d.tier)
+        };
+        let mut t = Title::load(&wai, "staged-measure-conformance/fixtures/still").unwrap();
+        // The base alone: its layers name the pinned base by its content hash, and its source
+        // is the size of their final stage.
+        t.layers.clear();
+        t.references.clear();
+        t.source = None;
+        t.expect = None;
+        let pinned = Wai::from_bytes(&t.bases[0].obj.bytes).unwrap();
+        let pin = pinned.prior().expect("the fixture's base is pinned").sha256.clone();
+        assert_eq!(entry_of(&t), (Some(pin.clone()), Some("decode-equivalence".to_string())));
+
+        let mut unpinned = Wai::from_bytes(&t.bases[0].obj.bytes).unwrap();
+        unpinned.manifest.conditioning.prior = None;
+        unpinned.manifest.intent = "improve".into();
+        assert!(unpinned.check_unpinned().is_some(), "the planted base is reported prior_unpinned");
+        t.bases[0].obj.bytes = unpinned.to_bytes().unwrap();
+        t.bases[0].case = None;
+        assert_eq!(entry_of(&t), (Some(pin), Some("none".to_string())), "the set it read, by the digest its pin would have named");
+    }
+
     /// A metered session, as `wai_stage_measure bench --sign` seals one: the figure a live
     /// from-rest bracket gives under `OnChipCounter`, in the session's `energy`, and the
     /// `measurement` claim on it by the same key, chained after the session and checked
diff --git a/wai/wai-rs/src/staged/mod.rs b/wai/wai-rs/src/staged/mod.rs
index 95996ad2b..02255f33c 100644
--- a/wai/wai-rs/src/staged/mod.rs
+++ b/wai/wai-rs/src/staged/mod.rs
@@ -296,11 +296,17 @@ pub struct BaseDecode {
     /// the decode read only bytes it verified; `None` for an unpinned decode, or a
     /// capability that reads no parameter set.
     pub pin: Option<crate::container::Prior>,
+    /// The parameter set the decode read, named as a pin names it (SPEC §3.1, *Pin by
+    /// default*): the `sha256` of the pin it verified; for an unpinned decode, the digest
+    /// a pin over the bytes it read would name (`int_payload::IntParams::read_pin`). `None`
+    /// for a capability that reads no parameter set. A record of the decode names its set
+    /// by it (jwp-receipts §2.4.12).
+    pub set: Option<String>,
 }
 
-/// [`decode_base_via`], also giving the pin the decode verified: what a record of the
-/// decode names as its parameter set (`extensions/staged-measurement.md` Appendix C,
-/// the decoder entries).
+/// [`decode_base_via`], also giving the pin the decode verified and the set it read: what
+/// a record of the decode names as its parameter set (`extensions/staged-measurement.md`
+/// Appendix C, the decoder entries).
 pub fn decode_base_pinned(
     envelope: &[u8],
     via: Option<&str>,
@@ -379,6 +385,10 @@ pub fn decode_base_pinned(
         };
         let units = units_from_decoded(decoded).filter(|u| !u.is_empty()).ok_or(BaseError::NoUnits)?;
         let pin = if params.is_pinned() { w.prior().cloned() } else { None };
+        let read = match &pin {
+            Some(p) => Some(p.sha256.clone()),
+            None => params.read_pin(&cap),
+        };
         let info = BaseInfo {
             content_hash: crate::jwp_hash::content_hash(envelope),
             tier: tier::base_tier(&cap, pin.as_ref()),
@@ -387,7 +397,7 @@ pub fn decode_base_pinned(
             attested: false,
             first_unit: 0,
         };
-        return Ok(BaseDecode { info, units, pin });
+        return Ok(BaseDecode { info, units, pin, set: read });
     }
 }
 
diff --git a/wai/wai-rs/tests/cli_dx.rs b/wai/wai-rs/tests/cli_dx.rs
index 92602f8e8..25d90a553 100644
--- a/wai/wai-rs/tests/cli_dx.rs
+++ b/wai/wai-rs/tests/cli_dx.rs
@@ -2,7 +2,9 @@
 //! object, its receipt index and its group's receipt file, and `wai verify` checks them
 //! offline and refuses each when it is altered; `wai create` writes each example of
 //! `wai/examples/` as a project, file for file and byte for byte, and never into a directory
-//! that holds anything.
+//! that holds anything. `wai verify` also reads the envelopes under a path: a planted envelope
+//! that pins nothing where its capability decodes against a parameter set is reported
+//! `prior_unpinned`, and still verifies; one whose pin breaks SPEC §3.1 does not.
 #![cfg(feature = "quickstart")]
 
 use std::path::{Path, PathBuf};
@@ -42,7 +44,8 @@ fn hello_runs_the_four_steps_and_verify_checks_what_it_wrote() {
     let h = d.join("h");
     let v = wai(&d, &["verify", "h"]);
     assert!(v.status.success(), "{}", text(&v));
-    assert!(text(&v).contains("delivered-verified") && text(&v).contains("1 receipt index verified"), "{}", text(&v));
+    // The object is an envelope there too, and its pins are read.
+    assert!(text(&v).contains("delivered-verified") && text(&v).contains("1 receipt index and 1 envelope verified"), "{}", text(&v));
 
     // A byte of the object changed: its content hash is not the receipt's.
     let obj = one(&h, ".wai");
@@ -77,6 +80,49 @@ fn hello_runs_the_four_steps_and_verify_checks_what_it_wrote() {
     let _ = std::fs::remove_dir_all(&d);
 }
 
+/// SPEC §3.1, *Pin by default*: `wai verify` reports `prior_unpinned` on a planted envelope
+/// that pins nothing over a capability that decodes against a parameter set, and the envelope
+/// still verifies; it reports nothing on one over a standard codec or on a pinned one; and it
+/// refuses one whose pin a verifier can tell is wrong without holding a set.
+#[test]
+fn verify_reports_an_unpinned_envelope() {
+    use wai::container::{Conditioning, Manifest, ModelRequirement, Prior, Wai};
+    let d = tmp("unpinned");
+    let p = d.join("planted");
+    std::fs::create_dir_all(&p).unwrap();
+    std::fs::write(d.join("payload.bin"), b"opaque test bytes").unwrap();
+    let wrap = |cap: &str, out: &str| {
+        let o = wai(&d, &["wrap", cap, "payload.bin", "-o", out, "--intent", "improve"]);
+        assert!(o.status.success(), "{}", text(&o));
+        o
+    };
+    let w = wrap("wai.neural.bmshj2018", "planted/unpinned.wai");
+    assert!(text(&w).contains("note prior_unpinned: wai.neural.bmshj2018"), "wrap says it pinned nothing: {}", text(&w));
+    wrap("wai.image.png", "planted/png.wai");
+    let pin = Prior::new("int_mlicpp/test", "7ef44f3b1fc4a2df6c3e3b68460f595cae3d31cb7f9018699e29e738340272a4", "int-rans/p16+int-mlicpp/i64");
+    let pinned = Wai::new(Manifest::new("image", "improve", ModelRequirement::new("wai.neural.int_mlicpp"),
+                                        Conditioning::new("int_mlicpp").with_prior(pin.clone())), b"M".to_vec());
+    std::fs::write(p.join("pinned.wai"), pinned.to_bytes().unwrap()).unwrap();
+
+    let v = wai(&d, &["verify", "planted"]);
+    let t = text(&v);
+    assert!(v.status.success(), "a report is not a refusal: {t}");
+    assert!(t.contains("note prior_unpinned: wai.neural.bmshj2018 decodes against a parameter set"), "{t}");
+    assert_eq!(t.matches("prior_unpinned").count(), 1, "the PNG and the pinned envelope are not reported: {t}");
+    assert!(t.contains("3 envelopes verified"), "{t}");
+    let v = wai(&d, &["verify", "planted/unpinned.wai", "--json"]);
+    assert!(v.status.success() && text(&v).contains("prior_unpinned"), "{}", text(&v));
+
+    // A pin over a capability that reads no set: the envelope does not conform.
+    let bad = Wai::new(Manifest::new("image", "replicate", ModelRequirement::new("wai.image.png"),
+                                     Conditioning::new("png").with_prior(pin)), b"PNG".to_vec());
+    std::fs::write(p.join("bad.wai"), bad.to_bytes().unwrap()).unwrap();
+    let v = wai(&d, &["verify", "planted"]);
+    assert_eq!(v.status.code(), Some(1), "{}", text(&v));
+    assert!(text(&v).contains("FAIL WAI1 wai.image.png: prior_not_applicable"), "{}", text(&v));
+    let _ = std::fs::remove_dir_all(&d);
+}
+
 /// Every file of an example but its `test/` directory, relative to it, sorted.
 fn example_files(dir: &Path) -> Vec<String> {
     fn walk(root: &Path, p: &Path, out: &mut Vec<String>) {
-- 
2.54.0 (Apple Git-157)


From ae84a668a471c18728310115ef5e0bf83f375fd4 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:53:47 -0500
Subject: [PATCH 21/29] wai: the page's recorder names the set an unpinned
 decode read, and a measured title over an unpinned base holds both recorders
 to it
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The browser recorder (wai-web's SessionHandle.decoderFor) still wrote pin and tier null
for every decode that verified no pin, so for one unpinned decode the page and the native
tool sealed different claims, against SPEC §3.1 rule 4 and jwp-receipts §2.4.12. Every
committed fixture pinned its base, which hid it. decoderFor now takes the set the decode
read, a trailing argument, apart from whether it verified a pin; StageBase gives both
(paramSet, pinned) from staged::decode_base_pinned. A decode that verified no pin is named
by the set it read, with the tier none, as stage_measure's recorder names it; a pinned
decode given a set other than its pin's is refused. measure_verify.py names the set as the
text does: the SHA-256 of a one-file capability's file, otherwise of the wai.prior.set
listing of the files read.

staged-measure-conformance/fixtures/unpinned is a title over the staged corpus's unpinned
wai.neural.int_hyper base (image_int_hyper_unpinned), decoded against the same files the
still's pin names, so its decoder entry names that digest; goldens.json regenerated
with wai_stage_measure. The parity test and verify_measure_wasm.mjs pass the verified-pin
flag apart from the triple's pin, so both recorders and both verifiers are compared on it,
and verify_staged_wasm.mjs holds StageBase.pinned and paramSet to every staged case.

jwp-receipts §2.4.12 says that an entry from a sink of the earlier text may carry null for
an unpinned decode that read a set, which a verifier cannot tell from one that read none.
SPEC §3.1 rule 2 says that an unpinned envelope conforms unless §7 requires the pin
(replicate_unpinned). DecoderEntry::pin says what it now names.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md                              |  10 +-
 wai/SPEC.md                                   |   3 +-
 wai/extensions/jwp-receipts.md                |   5 +
 wai/staged-measure-conformance/README.md      |   7 +-
 .../fixtures/unpinned/L01.wai                 | Bin 0 -> 619 bytes
 .../fixtures/unpinned/L02.wai                 | Bin 0 -> 547 bytes
 .../fixtures/unpinned/R0_00.wai               | Bin 0 -> 811 bytes
 .../fixtures/unpinned/expect.json             |  29 ++
 .../fixtures/unpinned/source.bin              | Bin 0 -> 768 bytes
 .../fixtures/unpinned/title.json              |  31 ++
 wai/staged-measure-conformance/goldens.json   | 307 +++++++++++++++++-
 .../measure_verify.py                         |  41 ++-
 wai/wai-rs/src/bin/wai_stage_measure.rs       |   9 +
 wai/wai-rs/src/claim_receipt/session.rs       |   8 +-
 wai/wai-web/demo/verify_measure_wasm.mjs      |  19 +-
 wai/wai-web/demo/verify_staged_wasm.mjs       |  12 +-
 wai/wai-web/src/measure.rs                    |  94 ++++--
 wai/wai-web/src/stage.rs                      |  29 +-
 18 files changed, 555 insertions(+), 49 deletions(-)
 create mode 100644 wai/staged-measure-conformance/fixtures/unpinned/L01.wai
 create mode 100644 wai/staged-measure-conformance/fixtures/unpinned/L02.wai
 create mode 100644 wai/staged-measure-conformance/fixtures/unpinned/R0_00.wai
 create mode 100644 wai/staged-measure-conformance/fixtures/unpinned/expect.json
 create mode 100644 wai/staged-measure-conformance/fixtures/unpinned/source.bin
 create mode 100644 wai/staged-measure-conformance/fixtures/unpinned/title.json

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index ff63e017e..02fe293ba 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -21,8 +21,14 @@ wire format may change and such changes are recorded here.
 - **A record of a decode names the set it read.** A `reconstruct-session` decoder entry's `pin` (jwp-receipts
   §2.4.12) now names the set an unpinned decode read, by the digest a pin over those bytes names, with the tier
   `none`; it was `null`. The reference sink's base decode gives it (`staged::BaseDecode::set`, from
-  `int_payload::IntParams::read_pin`), and the sessions `stage_measure` seals carry it. Every committed session and
-  golden pins its bases, so none changes.
+  `int_payload::IntParams::read_pin`), and the sessions `stage_measure` seals carry it. The page's recorder writes the
+  same: `wai-web`'s `SessionHandle.decoderFor` takes the set the decode read (`StageBase.paramSet`) beside whether
+  it verified a pin (`StageBase.pinned`), and `measure_verify.py` names the set as the text does. The fixture
+  `staged-measure-conformance/fixtures/unpinned`, over the staged corpus's unpinned `wai.neural.int_hyper` base, holds
+  both recorders and both verifiers to it; every other committed session and golden pins its bases, so none of them
+  changes. A decoder entry sealed by a sink of the earlier text may carry `null` for an unpinned decode that read a
+  set, and a verifier cannot tell it from a decode that read none (jwp-receipts §2.4.12).
+- **SPEC §3.1 rule 2** says that an unpinned envelope conforms unless §7 requires the pin (`replicate_unpinned`).
 
 - **Breaking:** `api:rust:wai::staged::BaseDecode::set`: a new member of a public struct, the set the decode read; a
   struct literal or a pattern that lists every member adds it.
diff --git a/wai/SPEC.md b/wai/SPEC.md
index 1b5f86246..9538480bc 100644
--- a/wai/SPEC.md
+++ b/wai/SPEC.md
@@ -436,7 +436,8 @@ was coded against gives plausible output, not an error (§4). So:
    `model_requirement.fallback`, or a `WAI2` entry's capability — is registered as
    decoding against one, whatever the envelope's `intent`. `wai.generate.spectral` is
    the exception above: its payload names its set.
-2. An envelope that pins nothing there conforms. A sink decodes it as before.
+2. An envelope that pins nothing there conforms, unless §7 requires the pin
+   (`replicate_unpinned`). A sink decodes it as before.
 3. A verifier reports it (`prior_unpinned`). The report is informative: it is not a
    finding that the envelope does not conform, and it is made whatever else the verifier
    finds. A capability this revision does not register is not reported, since this
diff --git a/wai/extensions/jwp-receipts.md b/wai/extensions/jwp-receipts.md
index 49908bb1c..71d62b8fc 100644
--- a/wai/extensions/jwp-receipts.md
+++ b/wai/extensions/jwp-receipts.md
@@ -1041,6 +1041,11 @@ A **decoder entry**:
 | `module_status` | string or `null` | For a `module:` decoder, the anchor its grant was confirmed under (§7.9): `pinned` or `fetched`. `null` for a `builtin:` decoder. |
 | `leaves` | integer | The presentations decoded through the entry. |
 
+A decoder entry sealed by a sink of an earlier text of this extension, under which `pin`
+named only a pin the sink verified, may carry `pin` and `tier` `null` for an unpinned
+decode that read a parameter set. A verifier cannot tell such an entry from one whose decode
+read no set.
+
 `energy` is the decode energy of one completion of the presentations the claim commits
 to — the decodes behind its `leaf_count` leaves, the work unit
 [staged-measurement](staged-measurement.md) §8 names (its §8.1) — on the host's
diff --git a/wai/staged-measure-conformance/README.md b/wai/staged-measure-conformance/README.md
index 891e2a01f..ec6af14ed 100644
--- a/wai/staged-measure-conformance/README.md
+++ b/wai/staged-measure-conformance/README.md
@@ -70,13 +70,16 @@ identity (Appendix D). Across the corpus: the record-wide deduplication totals (
 [`../wai-rs/tests/vectors/`](../wai-rs/tests/vectors/), generated too (that directory's
 README states how), as the staged-delivery corpus wraps it: a `wai.neural.int_hyper` picture, a `wai.video.int_hyper` clip, a
 `wai.audio.int_codec` clip and a `wai.splat.int_codec` attribute tensor, each pinned to
-its parameter set. Each source is the base's output with fine detail added by integer
+its parameter set; and the unpinned `wai.neural.int_hyper` picture of
+`image_int_hyper_unpinned`, decoded against the set the sink holds, which its decoder entry
+names by the digest a pin over it names (SPEC §3.1, *Pin by default*): the digest the
+pinned picture's pin names, over the same files. Each source is the base's output with fine detail added by integer
 functions of position, channel and a seed (for audio, two triangle waves), raised to the
 final shape by the layer's own predictor where the title changes the shape. The layers
 are `WIR1` payloads coded toward the source by the reference encoder
 (`staged::wir1::encode`): up2 then identity for the picture; one identity layer per
 frame, twice, for the clip; a rate layer to twice the sample rate, then identity, for
-the audio; identity for the attribute tensor. Each title's single-layer references code
+the audio; identity for the attribute tensor and for the unpinned picture. Each title's single-layer references code
 one layer straight from stage 0 at the final step and, for some titles, at further
 steps (the curve of §6.3).
 
diff --git a/wai/staged-measure-conformance/fixtures/unpinned/L01.wai b/wai/staged-measure-conformance/fixtures/unpinned/L01.wai
new file mode 100644
index 0000000000000000000000000000000000000000..4891b1a682964d0fc44b2409d099e5956ffe5175
GIT binary patch
literal 619
zcmWG{^fYv4WMHUPDo@N*vQjeCGf>h|%1uqlOa$^Wa}(23f&9$8lGMBspg>V-K~83J
zVhKnvH$NpcC%!1Pur#wMH5V*jt(2TtkeHO2lUY&;R1Gvz4{U@U&;X#yw9LFzrCJ@O
z<ovvp%#zIfyv)3GC97(s?999rkn+qTLy$B?VKGQ1DX};eC}W(KY+`C=nQCNWnUrRh
zWS*9kWSnH0W^M=+wXjG^v`Di^GBY+dvq&@milru-8l_koo0}OJo1`WsCZ`zzU0YC;
zT1Fj{auO?3i<GPkfi5jcEJ{xWhD%;)PEM^1Fc`x<gA5sgG!QWeKru6rh5<Gp<M03f
z|2d%otJ#5S6&V=Tm@zQ?I}OAS85mZ6W?=aLAEbbh;RKM3>z&?QBG16Es9=+3?+fYv
zwG+Mv*)My4`TyO2TiYEbEvq~FBc0`n!-d_l?GGiEOECm>`Fr`C-Wq6{@lnt6&`tYu
zJ%&0vZ*@MZf7^BU`PBDiRpGn;JPPREpi^2pZLNrUJky5lYea<e%Ig*vy_vzhMS|(z
za_PKtR+CPgY;L<;&?51)_H^y~s(HLS6e6yO1iI$Ft&4d6tj(aP>P%^6$JAiiV1*hk
qx3AXM6YsrzVd|}M@5x=(fQw!Me^R$IK0O)zOU9b3fwiRS(nJ96@Y-Ym

literal 0
HcmV?d00001

diff --git a/wai/staged-measure-conformance/fixtures/unpinned/L02.wai b/wai/staged-measure-conformance/fixtures/unpinned/L02.wai
new file mode 100644
index 0000000000000000000000000000000000000000..64d88e745902a03dc00a36583beeb7fa52f6d1ae
GIT binary patch
literal 547
zcmXv~J!lj`6rMvww347mAV`?X;y8EqZ*~tXL{QE$=8T9CW_I5kgL`}D-X>>QQ$_Fx
zMX?YS;owq<C>9E$Xd(V#p@@m1Rw{uIf);{BimY#=8RmP#`@U~J9&c=KjqNKDlF_cv
zv`SgUpk>;GPz&sIQ#1g39j6edNF>mXbRg1#Y_>v(#uGR(p%ZAX=@|_~TSy)0bPC-V
zSy_vyU;sLmj)6{>Y0!#8o$6Ms;|8r}^tg`0!mblm$p4I;LXn~as8}kn9oGlb_N8(q
zS5jKiQJf*`c~W@FldfgCo-mMuz%fJL;;v!YAVr`|oVK07BqnhSeC0^yhSCJ)3ow~$
z3U2#;=nDm&?Yi8NGVmQ2Y#spPk_V2BzbwNL4#T8KOhH1cCQh1)qye~>cp{3X=LjKV
zwGmdrb>byk|6>Eb>!;vX^E}_QPMGLLjJ_wNj>z*uD3P;Rcgwl!MRzi@Z!LD~4`v?y
zzPP}?o+ej!+@IOmJ#_czsqeSnb(b2YC$Be_2RgMI^DFlT-c+*lk9$8h|IC)RJtP;r
z?UxQbTj*Q$JyoT@eR_Fic6n)jc4YkH<)Qt<yAE$z=^wjxhkt%s?j5uzzZkvv*Nf**
ThmO1&9XvPqr`5mt0nYpdDJ!&x

literal 0
HcmV?d00001

diff --git a/wai/staged-measure-conformance/fixtures/unpinned/R0_00.wai b/wai/staged-measure-conformance/fixtures/unpinned/R0_00.wai
new file mode 100644
index 0000000000000000000000000000000000000000..4a5775d91a8c244a12a88f22c2366c3999ab59e1
GIT binary patch
literal 811
zcmWG{^fYv4WMHUPDo@N*vQjeCGf>h|%1uqlOa$^Wa}(23f&9$8lGMBspg>V-K~83J
zVhKnvH$NpcC%!1Pur#wMH5V*jt(2TtkeHO2lUY&;R1Gvz4{U@U&;X#yw9LFzrCJ@O
z<ovvp%#zIfyv)3GC97(s?999rkn+qTLy$B?VKGQ1DX};eC}W(KY+`C=nQCNWnUrRh
zWS*9kWSnH0W^M=+wXjG^v`Di^GBY+dvq&@milru-8l_koo0}OJo1`WsCZ`zzU0YC;
zT1Fj{auO?3i<GPkfi5jcEJ{xWhD%;)PEPF|U@(Sz1{pE}X&_<{fMRAK4Fe$Y5C8xF
zSB46#W@lhn1HwSXYk`<aih+R{gcTVWSb#WEgMlFmh&SjlFl+>2GX{oDKpbn&zz_$-
zKMpc5`~>0C3=F@3*#8m(LjVZhW?%>e;y(`=82$or&<h5JU=V)8zz_n&D?c+ZtODZy
z|3MC6WKdvaV7N5t&Yb2CKy%MdzVx1><I8vc%bqpqhpy=T?W=iM?X|x83ESRkn_0iQ
z_Nm2`w9n@_TD~W8^)tU|_Y|0>$1`=<RL{8b|LA^W|HquIvD&OGt6msb-SORhNHF1k
zYlYOj=Kp()zAZD;&9%tAyj_;bP+G*4P3-R>w;6g<BhKVJ(TR58Hn!fEyh+>S*;Mvy
ze$mczle#W=-fvEx%j1%+rQfyeO3v{|&-$(w{EaJKFC%~OePq?f8|yy^XWch=KXdQg
zjOx{P-q-zihWJdr;XOn9)bg#{CTwK)*t+4JX+ZeNF1wc>)4GMyjkp)|9(g3a<wYG|
z(~ia?mnZFc)IZbq%7vB-vyT32x+L4ZXz~Sixy5D8qTzq?c^Tda+;ME(oc;fcQSQCd
FUI0*$A?yGE

literal 0
HcmV?d00001

diff --git a/wai/staged-measure-conformance/fixtures/unpinned/expect.json b/wai/staged-measure-conformance/fixtures/unpinned/expect.json
new file mode 100644
index 000000000..dcc6bd28e
--- /dev/null
+++ b/wai/staged-measure-conformance/fixtures/unpinned/expect.json
@@ -0,0 +1,29 @@
+{
+  "format": "wai-measure-expect/1",
+  "about": "The fixture tool's own reconstruction: the SHA-256 of each unit's canonical buffer at each stage it coded, and of each reference's final units, from the encoder's reconstruction loop.",
+  "presentations": [
+    {
+      "unit": 0,
+      "layers": 1,
+      "out_sha256": "dce7282d46f88f03d5dc0fd42f325756d310fe1fc982b6e1792a1fbfbeb482cf"
+    },
+    {
+      "unit": 0,
+      "layers": 2,
+      "out_sha256": "71639c87086d0fe69eb0e9ba8a976e882cb7ac38056a3c388d901dd3f0ac4fd2"
+    },
+    {
+      "unit": 0,
+      "layers": 3,
+      "out_sha256": "7655e2d7055f1ea34084785b13e0becbec58301023ee9beff3e75a11abb7b621"
+    }
+  ],
+  "references": [
+    {
+      "label": "one layer, q 1",
+      "out_sha256": [
+        "7655e2d7055f1ea34084785b13e0becbec58301023ee9beff3e75a11abb7b621"
+      ]
+    }
+  ]
+}
diff --git a/wai/staged-measure-conformance/fixtures/unpinned/source.bin b/wai/staged-measure-conformance/fixtures/unpinned/source.bin
new file mode 100644
index 0000000000000000000000000000000000000000..183b492610fc1a64c862cb700308f879404883cf
GIT binary patch
literal 768
zcmW-fTZ+XX5JU?JA`-;C%r0jC`w)>JB2sgleB|CFWxcv>zd`lUR@d|{Jye(M-d%^d
ztnab8$H!WhYqxaluX5W4sVAlU0t&_BoH+Pv&_`J+m%~%a-Xn8-+NCQz((7zJYOp`Y
z-*9yfQf%u)sd_oLin7P&ju8W>c<Am&Elmz?Kzx~1`}y?ybL4WL`GiY_l&r=<u8JdH
zYh%XfwjmB~H7|ZscDxu3IU?D2yn1rrR#h5g*^@##(c9tu*_z3X-aOWYoip=JyL<_v
z5=;CR4l1p9&JD57{E8NCU>RDm`z#8pNfCDS=8YH+*M@I%A;hJL=@g4n-NLoMcakZq
zf_>^Jeru#nkwD|6?HDeq%=m_(;Bd;UK13*i`a69v26@49{gJSLGJ)>MOje`g1wEUA
z|C?F~1;wKQ&p$+r_~{sC6%eB6f!q$0HQBTQOBqgIK>$f2R)*-IsWijgkJQD-TYb61
z5{oq=b$z%c=z1UIcWewC2PQJGRrL71cq57Y_n-GJ8E-}lY!$ia44gMdCxxC2DTCgG
MS?O-mO}SqG0ic9^^Z)<=

literal 0
HcmV?d00001

diff --git a/wai/staged-measure-conformance/fixtures/unpinned/title.json b/wai/staged-measure-conformance/fixtures/unpinned/title.json
new file mode 100644
index 000000000..9af53e4d2
--- /dev/null
+++ b/wai/staged-measure-conformance/fixtures/unpinned/title.json
@@ -0,0 +1,31 @@
+{
+  "format": "wai-measure-title/1",
+  "title": "unpinned",
+  "about": "an unpinned wai.neural.int_hyper picture (16×16), decoded against the set the sink holds (SPEC §3.1, Pin by default), its decoder entry naming that set; an identity layer at q 6, then one at q 1; reference: one identity layer at q 1",
+  "unit_us": 40000,
+  "bases": [
+    {
+      "case": "staged-conformance/cases/image_int_hyper_unpinned",
+      "base": 0,
+      "first_unit": 0,
+      "capture_us": null
+    }
+  ],
+  "layers": [
+    "L01.wai",
+    "L02.wai"
+  ],
+  "source": {
+    "file": "source.bin",
+    "sha256": "7655e2d7055f1ea34084785b13e0becbec58301023ee9beff3e75a11abb7b621"
+  },
+  "references": [
+    {
+      "label": "one layer, q 1",
+      "layers": [
+        "R0_00.wai"
+      ]
+    }
+  ],
+  "expect": "expect.json"
+}
diff --git a/wai/staged-measure-conformance/goldens.json b/wai/staged-measure-conformance/goldens.json
index e5d630f33..1a0486c53 100644
--- a/wai/staged-measure-conformance/goldens.json
+++ b/wai/staged-measure-conformance/goldens.json
@@ -1809,14 +1809,313 @@
         "referenced_bytes": 8846,
         "stored_bytes": 6515
       }
+    },
+    {
+      "title": "unpinned",
+      "content_id": "da9fd8c0531e288c5d4666624592ec5fae818a8320ad8e6e4f778ddf70533fd9",
+      "framing": "registered",
+      "unit_class": "rgb8",
+      "unit_us": 40000,
+      "bases": [
+        {
+          "object": "3fc4569e249bf6b7fbb3b5f7145688ba8f8b63368a0714ec52d9376034ebacf2",
+          "capability": "wai.neural.int_hyper",
+          "pin": "13e8b0d18e001eb9f0160309a02755b35a14f71ba0bcaa13f373e32b931ab818",
+          "tier": "none",
+          "first_unit": 0,
+          "capture_us": null,
+          "units": 1,
+          "param_set_bytes": 2676
+        }
+      ],
+      "objects": [
+        {
+          "file": "staged-conformance/cases/image_int_hyper_unpinned/base.wai",
+          "object": "3fc4569e249bf6b7fbb3b5f7145688ba8f8b63368a0714ec52d9376034ebacf2",
+          "capability": "wai.neural.int_hyper",
+          "base": 0,
+          "layer": 0,
+          "units": [
+            0,
+            1
+          ],
+          "counted": {
+            "header": 12,
+            "manifest": 158,
+            "payload": 88,
+            "total": 258
+          }
+        },
+        {
+          "file": "staged-measure-conformance/fixtures/unpinned/L01.wai",
+          "object": "73e9f5b16db2e19ae2162a7499d9afe846675bbc956e47ce17b7c543e9300a51",
+          "capability": "wai.image.int_refine",
+          "base": 0,
+          "layer": 1,
+          "units": [
+            0,
+            1
+          ],
+          "counted": {
+            "header": 12,
+            "manifest": 323,
+            "payload": 284,
+            "total": 619
+          }
+        },
+        {
+          "file": "staged-measure-conformance/fixtures/unpinned/L02.wai",
+          "object": "60c436fc8ff204639290211b6dbf0372c1834a3c0fecda783523280561377bea",
+          "capability": "wai.image.int_refine",
+          "base": 0,
+          "layer": 2,
+          "units": [
+            0,
+            1
+          ],
+          "counted": {
+            "header": 12,
+            "manifest": 323,
+            "payload": 212,
+            "total": 547
+          }
+        }
+      ],
+      "bfpf_min": 258,
+      "prefixes": [
+        {
+          "layers": 1,
+          "stage_counted": 258,
+          "counted": 258,
+          "per_base": [
+            {
+              "stage_counted": 258,
+              "counted": 258
+            }
+          ],
+          "units": 1,
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "39980",
+            "sse_y": "5688",
+            "ssim_y_q32": 4287319116,
+            "windows": 9
+          },
+          "vs_final": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "39980",
+            "sse_y": "5688",
+            "ssim_y_q32": 4287319116,
+            "windows": 9
+          }
+        },
+        {
+          "layers": 2,
+          "stage_counted": 619,
+          "counted": 877,
+          "per_base": [
+            {
+              "stage_counted": 619,
+              "counted": 877
+            }
+          ],
+          "units": 1,
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "1160",
+            "sse_y": "210",
+            "ssim_y_q32": 4294692926,
+            "windows": 9
+          },
+          "vs_final": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "1160",
+            "sse_y": "210",
+            "ssim_y_q32": 4294692926,
+            "windows": 9
+          }
+        },
+        {
+          "layers": 3,
+          "stage_counted": 547,
+          "counted": 1424,
+          "per_base": [
+            {
+              "stage_counted": 547,
+              "counted": 1424
+            }
+          ],
+          "units": 1,
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "0",
+            "sse_y": "0",
+            "ssim_y_q32": 4294967296,
+            "windows": 9
+          },
+          "vs_final": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "0",
+            "sse_y": "0",
+            "ssim_y_q32": 4294967296,
+            "windows": 9
+          }
+        }
+      ],
+      "decoders": [
+        {
+          "capability": "wai.neural.int_hyper",
+          "pin": "13e8b0d18e001eb9f0160309a02755b35a14f71ba0bcaa13f373e32b931ab818",
+          "object": "3fc4569e249bf6b7fbb3b5f7145688ba8f8b63368a0714ec52d9376034ebacf2"
+        },
+        {
+          "capability": "wai.image.int_refine",
+          "pin": null,
+          "object": "73e9f5b16db2e19ae2162a7499d9afe846675bbc956e47ce17b7c543e9300a51"
+        },
+        {
+          "capability": "wai.image.int_refine",
+          "pin": null,
+          "object": "60c436fc8ff204639290211b6dbf0372c1834a3c0fecda783523280561377bea"
+        }
+      ],
+      "presentations": [
+        {
+          "unit": 0,
+          "layers": 1,
+          "decoder": 0,
+          "shape": [
+            16,
+            16,
+            3
+          ],
+          "out_sha256": "dce7282d46f88f03d5dc0fd42f325756d310fe1fc982b6e1792a1fbfbeb482cf",
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "39980",
+            "sse_y": "5688",
+            "ssim_y_q32": 4287319116,
+            "windows": 9
+          },
+          "vs_final": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "39980",
+            "sse_y": "5688",
+            "ssim_y_q32": 4287319116,
+            "windows": 9
+          }
+        },
+        {
+          "unit": 0,
+          "layers": 2,
+          "decoder": 1,
+          "shape": [
+            16,
+            16,
+            3
+          ],
+          "out_sha256": "71639c87086d0fe69eb0e9ba8a976e882cb7ac38056a3c388d901dd3f0ac4fd2",
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "1160",
+            "sse_y": "210",
+            "ssim_y_q32": 4294692926,
+            "windows": 9
+          },
+          "vs_final": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "1160",
+            "sse_y": "210",
+            "ssim_y_q32": 4294692926,
+            "windows": 9
+          }
+        },
+        {
+          "unit": 0,
+          "layers": 3,
+          "decoder": 2,
+          "shape": [
+            16,
+            16,
+            3
+          ],
+          "out_sha256": "7655e2d7055f1ea34084785b13e0becbec58301023ee9beff3e75a11abb7b621",
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "0",
+            "sse_y": "0",
+            "ssim_y_q32": 4294967296,
+            "windows": 9
+          },
+          "vs_final": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "0",
+            "sse_y": "0",
+            "ssim_y_q32": 4294967296,
+            "windows": 9
+          }
+        }
+      ],
+      "output_root": "0e7a20ad5c9593c26900154e2ffb606928445498472264285d13c248d9136ca6",
+      "leaf_count": 3,
+      "overhead": [
+        {
+          "label": "staged",
+          "counted": 1424,
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "0",
+            "sse_y": "0",
+            "ssim_y_q32": 4294967296,
+            "windows": 9
+          },
+          "out_sha256": [
+            "7655e2d7055f1ea34084785b13e0becbec58301023ee9beff3e75a11abb7b621"
+          ]
+        },
+        {
+          "label": "one layer, q 1",
+          "counted": 1069,
+          "vs_source": {
+            "kind": "rgb8",
+            "n_samples": 768,
+            "sse_rgb": "0",
+            "sse_y": "0",
+            "ssim_y_q32": 4294967296,
+            "windows": 9
+          },
+          "out_sha256": [
+            "7655e2d7055f1ea34084785b13e0becbec58301023ee9beff3e75a11abb7b621"
+          ]
+        }
+      ],
+      "dedup": {
+        "objects": 5,
+        "unique": 4,
+        "referenced_bytes": 2493,
+        "stored_bytes": 2235
+      }
     }
   ],
   "record": {
     "dedup": {
-      "objects": 46,
-      "unique": 32,
-      "referenced_bytes": 133542,
-      "stored_bytes": 113587
+      "objects": 51,
+      "unique": 36,
+      "referenced_bytes": 136035,
+      "stored_bytes": 115822
     }
   }
 }
diff --git a/wai/staged-measure-conformance/measure_verify.py b/wai/staged-measure-conformance/measure_verify.py
index cf051ad0e..66c60ed40 100755
--- a/wai/staged-measure-conformance/measure_verify.py
+++ b/wai/staged-measure-conformance/measure_verify.py
@@ -20,7 +20,9 @@ specifications. It implements the rest itself:
 - the quality integers (§4): integer BT.601 luma, SSE, the Q32 structural similarity
   over clipped 8×8 windows at stride 4 with an edge-aligned last window on each axis
   (Python's // floors toward −∞, as the method states), PCM and tensor sums, each pooled over several units;
-- the output tree (Appendix C): RFC 6962's tree hash over SHA-256 leaves;
+- the output tree (Appendix C): RFC 6962's tree hash over SHA-256 leaves, its decoder
+  entries each naming the parameter set its decode read (jwp-receipts §2.4.12): the pin
+  it verified, or, unpinned, the digest a pin over the bytes read would name (SPEC §3.1);
 - the prefixes' counted bytes, first presentable, the layering overhead's points,
   deduplication (§9), and the title's content identity.
 
@@ -70,6 +72,18 @@ UNIT_CLASS = {
 }
 KIND = {"rgb8": "rgb8", "pcm16": "pcm", "tensor-i64": "attr"}
 
+# SPEC §5: the members of the parameter set each base capability the corpus uses decodes
+# against, and those registered with a one-file set, which a pin names by that file's
+# SHA-256 rather than by a listing (SPEC §3.1).
+SET_MEMBERS = {
+    "wai.neural.int_hyper": ["entropy.json", "g_s.whm", "h_s.whm"],
+    "wai.video.int_hyper": ["entropy.json", "g_s.whm", "h_s.whm"],
+    "wai.video.int_motion": [],
+    "wai.audio.int_codec": ["model.wam"],
+    "wai.splat.int_codec": ["g_s.whm"],
+}
+ONE_FILE = {"wai.audio.int_codec", "wai.splat.int_codec"}
+
 # staged-measurement Appendix D: the tools' stated limits, and the order Rust's
 # DeclaredCost::exceeds compares the figures in.
 TOOL_COST = [1 << 25, 1 << 25, 1 << 28, 1 << 38]
@@ -138,6 +152,21 @@ def base_checks(env, cost):
     raise Refusal("missing_capability", "no capability this decode takes")
 
 
+def set_read(cap, held):
+    """The parameter set a decode through `cap` read from `held` (member -> bytes), by the
+    digest a pin over those bytes names (SPEC §3.1, Pin by default; jwp-receipts §2.4.12): the
+    SHA-256 of the one file of a capability registered with a one-file set, and otherwise the
+    SHA-256 of the listing (`wai.prior.set`) of the members read, one line each,
+    `<sha256>  <path>`, in byte order of the paths. None where it read no set."""
+    read = {m: held[m] for m in SET_MEMBERS.get(cap, []) if m in held}
+    if not read:
+        return None
+    if cap in ONE_FILE:
+        return sha(read[SET_MEMBERS[cap][0]])
+    listing = b"".join(sha(read[m]).encode() + b"  " + m.encode() + b"\n" for m in sorted(read, key=lambda x: x.encode()))
+    return sha(listing)
+
+
 def sha(b):
     return hashlib.sha256(b).hexdigest()
 
@@ -411,8 +440,10 @@ def evaluate(f):
         if sha(stage0) != cb["stage0_sha256"]:
             raise ValueError(f"{path}: stage0")
         set_bytes = 0
-        for _, sp in sorted(cb["decode"]["set"].items()):
-            set_bytes += len(f.obj(sp))
+        held = {}
+        for member, sp in sorted(cb["decode"]["set"].items()):
+            held[member] = f.obj(sp)
+            set_bytes += len(held[member])
         units, at = [], 0
         for u in cb["units"]:
             shape = tuple(u["shape"])
@@ -425,8 +456,10 @@ def evaluate(f):
                 at += n
         if sum(len(x) * (2 if u["elem"] == "i16" else 1) for x, u in zip((v for _, v in units), cb["units"])) > budget:
             raise Refusal("chain-over-limit", "the base's units pass the chain budget")
+        # The set the decode read, as a decoder entry names it: the pin it verified, or,
+        # unpinned, the digest a pin over the bytes it read would name.
         prior = (m.get("conditioning") or {}).get("prior")
-        pin = prior["sha256"] if cb["pin_resolved"] and prior else None
+        pin = prior["sha256"] if cb["pin_resolved"] and prior else set_read(cap, held)
         bases.append({"hash": cb["object"], "cap": cap, "pin": pin, "first": b["first_unit"], "units": units})
         base_extra.append({"object": cb["object"], "capability": cap, "pin": pin, "tier": cb["tier"],
                            "first_unit": b["first_unit"], "capture_us": b["capture_us"], "units": len(units),
diff --git a/wai/wai-rs/src/bin/wai_stage_measure.rs b/wai/wai-rs/src/bin/wai_stage_measure.rs
index 74e4d95a3..72e234f5a 100644
--- a/wai/wai-rs/src/bin/wai_stage_measure.rs
+++ b/wai/wai-rs/src/bin/wai_stage_measure.rs
@@ -288,6 +288,15 @@ fn specs() -> Vec<FixtureSpec> {
             layers: vec![l(Identity, 6, (0, 0), false), l(Identity, 1, (0, 0), false)],
             references: vec![("one layer, q 1", l(Identity, 1, (0, 0), false))],
         },
+        FixtureSpec {
+            name: "unpinned",
+            about: "an unpinned wai.neural.int_hyper picture (16×16), decoded against the set the sink holds (SPEC §3.1, Pin by default), its decoder entry naming that set; an identity layer at q 6, then one at q 1; reference: one identity layer at q 1",
+            case: "staged-conformance/cases/image_int_hyper_unpinned",
+            cap: wai::codecs::CAP_IMAGE_INT_REFINE,
+            media: "image",
+            layers: vec![l(Identity, 6, (0, 0), false), l(Identity, 1, (0, 0), false)],
+            references: vec![("one layer, q 1", l(Identity, 1, (0, 0), false))],
+        },
     ]
 }
 
diff --git a/wai/wai-rs/src/claim_receipt/session.rs b/wai/wai-rs/src/claim_receipt/session.rs
index 9b6b2c3c1..53fc6bea6 100644
--- a/wai/wai-rs/src/claim_receipt/session.rs
+++ b/wai/wai-rs/src/claim_receipt/session.rs
@@ -147,9 +147,13 @@ pub struct DecoderEntry {
     /// `NeuralIntegerExact` capability's integer path may also be named by the lowering its
     /// registration names, as opus-dred §7.2 names `dred-int/1`.
     pub variant: Option<String>,
-    /// The SHA-256 of the pin the sink verified.
+    /// The parameter set the decode read, by the digest a pin over it names (SPEC §3.1,
+    /// *Pin by default*): the SHA-256 of the pin the sink verified, or, for a decode that
+    /// verified none, the digest a pin over the bytes it read would name. `None` when the
+    /// decode read no parameter set.
     pub pin: Option<[u8; 32]>,
-    /// That pin's declared tier, as SPEC §3.1 reads it.
+    /// That pin's declared tier, as SPEC §3.1 reads it; `none` for a decode that verified
+    /// no pin.
     pub tier: Option<String>,
     /// For a `module:` decoder, the anchor its grant was confirmed under.
     pub module_status: Option<String>,
diff --git a/wai/wai-web/demo/verify_measure_wasm.mjs b/wai/wai-web/demo/verify_measure_wasm.mjs
index 88416d8f4..855f7b115 100644
--- a/wai/wai-web/demo/verify_measure_wasm.mjs
+++ b/wai/wai-web/demo/verify_measure_wasm.mjs
@@ -219,6 +219,15 @@ const expected = [];
 const flip = (h) => (h[0] === '0' ? '1' : '0') + h.slice(1);
 // The title's unit that is unit 0 of an object's base: the base's own, for a layer too.
 const firstUnitOf = (golden, object) => golden.bases[golden.objects.find((o) => o.object === object).base].first_unit;
+// A decoder entry's pin names the set its decode read, pinned or not (jwp-receipts
+// §2.4.12): the decode verified a pin only where the envelope it decoded pins one.
+const pinnedOf = (envelope, d) => {
+  if (d.pin === null) return false;
+  const e = wai.parse(envelope);
+  const p = e.prior;
+  e.free();
+  return p !== undefined && p !== null;
+};
 for (const [name, { files, t, golden }] of titles) {
   const native = t.session(HOST, NONCE).finish(SEED, undefined, T);
   // The page's recording.
@@ -227,7 +236,8 @@ for (const [name, { files, t, golden }] of titles) {
   const h = wai.sessionBegin(entry, HOST, NONCE, undefined);
   for (const p of golden.presentations) {
     const d = golden.decoders[p.decoder];
-    const handle = h.decoderFor(envelopeOf.get(d.object), d.capability, d.pin !== null, firstUnitOf(golden, d.object));
+    const env = envelopeOf.get(d.object);
+    const handle = h.decoderFor(env, d.capability, pinnedOf(env, d), firstUnitOf(golden, d.object), d.pin);
     h.presented(p.unit, p.layers, handle, p.out_sha256, undefined);
   }
   for (const o of golden.objects) h.received(files.get(o.file));
@@ -279,7 +289,7 @@ for (const [name, { files, t, golden }] of titles) {
   // fresh handle, refused for its own field and not for anything an earlier case left.
   const fresh = () => {
     const x = wai.sessionBegin(env(base), HOST, NONCE, undefined);
-    return [x, x.decoderFor(env(base), golden.decoders[0].capability, golden.decoders[0].pin !== null, 0)];
+    return [x, x.decoderFor(env(base), golden.decoders[0].capability, pinnedOf(env(base), golden.decoders[0]), 0, golden.decoders[0].pin)];
   };
   for (const [what, field, f] of [
     ['unit 0.5', 'unit', (x, dd) => x.presented(0.5, 1, dd, p0.out_sha256, undefined)],
@@ -310,7 +320,8 @@ for (const [name, { files, t, golden }] of titles) {
     const dd = golden.decoders[p.decoder];
     if (dd.object === last.object) continue;
     const key = `${dd.object}`;
-    if (!handles.has(key)) handles.set(key, h.decoderFor(env(golden.objects.find((o) => o.object === dd.object)), dd.capability, dd.pin !== null, firstUnitOf(golden, dd.object)));
+    const de = env(golden.objects.find((o) => o.object === dd.object));
+    if (!handles.has(key)) handles.set(key, h.decoderFor(de, dd.capability, pinnedOf(de, dd), firstUnitOf(golden, dd.object), dd.pin));
     h.presented(p.unit, p.layers, handles.get(key), p.out_sha256, undefined);
   }
   for (const o of golden.objects) h.received(env(o));
@@ -356,7 +367,7 @@ for (const [name, { files, t, golden }] of titles) {
     const w = new Uint8Array(readFileSync(join(corpus, `counting/${which}.wai`)));
     const h = wai.sessionBegin(env, HOST, NONCE, undefined);
     const bd = golden.decoders.find((x) => x.object === base.object);
-    const dd = h.decoderFor(env, bd.capability, bd.pin !== null, firstUnitOf(golden, base.object));
+    const dd = h.decoderFor(env, bd.capability, pinnedOf(env, bd), firstUnitOf(golden, base.object), bd.pin);
     const p0 = golden.presentations.find((x) => golden.decoders[x.decoder].object === base.object && x.layers === 1);
     h.presented(p0.unit, 1, dd, p0.out_sha256, undefined);
     h.received(env);
diff --git a/wai/wai-web/demo/verify_staged_wasm.mjs b/wai/wai-web/demo/verify_staged_wasm.mjs
index 33b84f09c..1b19d1a49 100644
--- a/wai/wai-web/demo/verify_staged_wasm.mjs
+++ b/wai/wai-web/demo/verify_staged_wasm.mjs
@@ -7,10 +7,12 @@
 // A base whose decode is the integer path (case.json `decode.kind` "int") is decoded
 // in wasm32 too, through StageBase.decode with its parameter set, its pin and a cost
 // limit, and must give its stage0.bin byte for byte, the case's tier and the case's
-// fallback flag. A PNG or FLAC base (no decoder in this build) is claimed from
-// stage0.bin (StageBase.fromStage0), whose SHA-256 is checked; every report, unit and
-// stage hash of its chain must then say claimed, and with that mark set aside equal
-// the case's. Each base then takes the case's facts (attested, first unit).
+// fallback flag, report its pin verified exactly where the case states it resolved, and
+// name the parameter set it read (paramSet) exactly where the case gives it one. A PNG
+// or FLAC base (no decoder in this build) is claimed from stage0.bin
+// (StageBase.fromStage0), whose SHA-256 is checked; every report, unit and stage hash
+// of its chain must then say claimed, and with that mark set aside equal the case's.
+// Each base then takes the case's facts (attested, first unit).
 //
 // Then the binding's guards:
 //   - bounds: the bases in wai-rs/tests/vectors/staged_bounds/ declare clips of about
@@ -82,6 +84,8 @@ for (const name of names) {
       if (base.tier !== b.tier) { fail(`${name}: the base tier is ${base.tier}, the case states ${b.tier}`); ok = false; }
       if (base.viaFallback !== b.via_fallback) { fail(`${name}: via fallback ${base.viaFallback}, the case states ${b.via_fallback}`); ok = false; }
       if (base.capability !== b.decoded_via) { fail(`${name}: decoded via ${base.capability}`); ok = false; }
+      if (base.pinned !== b.pin_resolved) { fail(`${name}: pinned ${base.pinned}, the case states pin_resolved ${b.pin_resolved}`); ok = false; }
+      if ((base.paramSet !== undefined) !== (set.size > 0)) { fail(`${name}: paramSet ${base.paramSet}, the case gives ${set.size} set files`); ok = false; }
     } else {
       base = wai.StageBase.fromStage0(env, b.decoded_via, b.tier, shapes, [...elems][0], stage0);
       if (!base.claimed || base.viaFallback !== b.via_fallback || base.tier !== b.tier) { fail(`${name}: the claimed base's facts differ from the case's`); ok = false; }
diff --git a/wai/wai-web/src/measure.rs b/wai/wai-web/src/measure.rs
index 0687eded5..aef37baf3 100644
--- a/wai/wai-web/src/measure.rs
+++ b/wai/wai-web/src/measure.rs
@@ -204,12 +204,19 @@ pub(crate) fn host_of(host_json: &str) -> Result<SessionHost, String> {
 
 /// The decoder entry key for decoding `envelope` here: through `via` (its manifest's
 /// capability where absent; otherwise that capability or the manifest's fallback, and
-/// nothing else, SPEC §4), on the primary path, with this build's decoders; `pinned`
-/// when its decode verified the parameter set its manifest pins, whose SHA-256 is then
-/// the entry's `pin` and whose declared tier (SPEC §3.1: an unregistered token, or none,
-/// reads as `none`) its `tier`; `first_unit` the title's unit that is unit 0 of the
-/// object's base (of the object itself for a base, of the base it refines for a layer).
-pub(crate) fn decoder_key(envelope: &[u8], via: Option<&str>, pinned: bool, first_unit: u64) -> Result<DecoderKey, String> {
+/// nothing else, SPEC §4), on the primary path, with this build's decoders; `first_unit`
+/// the title's unit that is unit 0 of the object's base (of the object itself for a base,
+/// of the base it refines for a layer).
+///
+/// The entry names the parameter set the decode read (jwp-receipts §2.4.12; SPEC §3.1,
+/// *Pin by default*). `pinned` when the decode verified the set its manifest pins: the
+/// pin's SHA-256 is then the entry's `pin`, and its declared tier (an unregistered token,
+/// or none, reads as `none`) its `tier`. `set` is the set the decode read, by the digest a
+/// pin over it names, as the base's decode gives it (`StageBase::paramSet`,
+/// `staged::BaseDecode::set`): for a decode that verified no pin it is the entry's `pin`,
+/// with the tier `none`; for a pinned decode it is the pin's, where given. `None` with no
+/// pin verified: the decode read no parameter set.
+pub(crate) fn decoder_key(envelope: &[u8], via: Option<&str>, pinned: bool, set: Option<[u8; 32]>, first_unit: u64) -> Result<DecoderKey, String> {
     let w = Wai::from_bytes(envelope).map_err(|e| e.to_string())?;
     if let Some(v) = via {
         if v != w.capability() && w.fallback() != Some(v) {
@@ -219,9 +226,13 @@ pub(crate) fn decoder_key(envelope: &[u8], via: Option<&str>, pinned: bool, firs
     let (pin, tier) = if pinned {
         let p = w.prior().ok_or("pinned, but the envelope pins no parameter set")?;
         let tier = p.tier().and_then(DeclaredTier::from_token).unwrap_or(DeclaredTier::None).token().to_owned();
-        (Some(unhex32(&p.sha256, "the pin's sha256")?), Some(tier))
+        let pin = unhex32(&p.sha256, "the pin's sha256")?;
+        if set.is_some_and(|s| s != pin) {
+            return Err("set: a pinned decode reads the set its pin names, and this is another".into());
+        }
+        (Some(pin), Some(tier))
     } else {
-        (None, None)
+        (set, set.map(|_| DeclaredTier::None.token().to_owned()))
     };
     Ok(DecoderKey {
         capability: via.unwrap_or(w.capability()).to_owned(),
@@ -467,14 +478,18 @@ impl SessionHandle {
     /// manifest's capability where absent; otherwise that capability or its fallback, and
     /// any other is refused); `pinned` when the decode verified the parameter set the
     /// manifest pins; `first_unit`, which the page states, the title's unit that is unit 0
-    /// of the object's base (of the object for a base, of the base it refines for a layer).
-    /// The recorder holds the envelopes, so a `first_unit` that is not a layer's base's, or
-    /// that puts a presentation outside the units its envelope covers, is refused when the
-    /// session is sealed. The same decoder on the same object gives the same handle.
+    /// of the object's base (of the object for a base, of the base it refines for a layer);
+    /// `set_hex` the parameter set the decode read, as `StageBase.paramSet` gives it
+    /// (lowercase hex), or `undefined` for a decode that read none. A decode that verified
+    /// no pin but read a set is named by that set, with the tier `none` (jwp-receipts
+    /// §2.4.12). The recorder holds the envelopes, so a `first_unit` that is not a layer's
+    /// base's, or that puts a presentation outside the units its envelope covers, is refused
+    /// when the session is sealed. The same decoder on the same object gives the same handle.
     #[wasm_bindgen(js_name = decoderFor)]
-    pub fn decoder_for(&mut self, envelope: &[u8], via: Option<String>, pinned: bool, first_unit: f64) -> Result<u32, JsError> {
+    pub fn decoder_for(&mut self, envelope: &[u8], via: Option<String>, pinned: bool, first_unit: f64, set_hex: Option<String>) -> Result<u32, JsError> {
         let first_unit = int(first_unit, "first_unit")?;
-        let k = decoder_key(envelope, via.as_deref(), pinned, first_unit).map_err(js_err)?;
+        let set = set_hex.as_deref().map(|s| unhex32(s, "set")).transpose().map_err(js_err)?;
+        let k = decoder_key(envelope, via.as_deref(), pinned, set, first_unit).map_err(js_err)?;
         Ok(self.rec()?.decoder(k) as u32)
     }
 
@@ -708,8 +723,14 @@ mod tests {
             };
             for p in &m.e.presentations {
                 let (cap, pin, object) = &m.e.decoders[usize::from(p.p.decoder)];
-                let first = m.e.objects.iter().find(|o| o.content_hash == *object).map(|o| m.e.bases[o.base].0.info.first_unit).unwrap();
-                let h = r.decoder(decoder_key(envelope_of(object), Some(cap), pin.is_some(), first).unwrap());
+                let o = m.e.objects.iter().find(|o| o.content_hash == *object).unwrap();
+                let first = m.e.bases[o.base].0.info.first_unit;
+                // The triple's pin names the set the decode read, an unpinned base's too;
+                // whether it verified a pin is apart from that: the case's statement, which
+                // the base's decode was checked against.
+                let set = pin.as_deref().map(|x| unhex32(x, "the set read").unwrap());
+                let pinned = set.is_some() && m.t.bases[o.base].case.as_ref().is_some_and(|c| c.pin_resolved);
+                let h = r.decoder(decoder_key(envelope_of(object), Some(cap), pinned, set, first).unwrap());
                 r.presented(p.p.unit, p.p.layers, h, p.p.out_digest, None).unwrap();
             }
             for o in &m.e.objects {
@@ -753,15 +774,15 @@ mod tests {
         let m = measured(dir, &files).unwrap();
         let base = &m.t.bases[0].obj.bytes;
         let cap = Wai::from_bytes(base).unwrap().capability().to_owned();
-        assert_eq!(decoder_key(base, Some(&cap), true, 0).unwrap().capability, cap);
-        assert_eq!(decoder_key(base, None, true, 0).unwrap().capability, cap);
-        let why = decoder_key(base, Some("wai.audio.int_codec"), true, 0).unwrap_err();
+        assert_eq!(decoder_key(base, Some(&cap), true, None, 0).unwrap().capability, cap);
+        assert_eq!(decoder_key(base, None, true, None, 0).unwrap().capability, cap);
+        let why = decoder_key(base, Some("wai.audio.int_codec"), true, None, 0).unwrap_err();
         assert!(why.contains("neither the envelope's capability"), "{why}");
         let encrypted = std::fs::read(wai_dir().join("staged-measure-conformance/counting/encrypted.wai")).unwrap();
         let p = &m.e.presentations[0];
         let session = |capability: Option<&str>| {
             let (mut r, subject) = begin(base, host_of(HOST).unwrap(), [7; 32], None).unwrap();
-            let mut k = decoder_key(base, None, true, 0).unwrap();
+            let mut k = decoder_key(base, None, true, None, 0).unwrap();
             if let Some(c) = capability {
                 k.capability = c.into();
             }
@@ -781,6 +802,37 @@ mod tests {
         assert!(why.contains("session_source_mismatch"), "{why}");
     }
 
+    /// A decoder entry names the parameter set its decode read, apart from whether the decode
+    /// verified a pin (jwp-receipts §2.4.12; SPEC §3.1, *Pin by default*): the unpinned
+    /// fixture's base, decoded against the set the sink holds, is named by that set with the
+    /// tier `none`; the pinned fixture's by its pin, with the tier the pin declares, and a set
+    /// other than its pin's is refused; a decode that read no set names none. A decode
+    /// reported pinned over an envelope that pins nothing is refused.
+    #[test]
+    fn a_decoder_entry_names_the_set_its_decode_read() {
+        let base_of = |name: &str| {
+            let dir = format!("staged-measure-conformance/fixtures/{name}");
+            let m = measured(&dir, &fetch(&dir)).unwrap();
+            let set = m.e.decoders[0].1.as_deref().map(|x| unhex32(x, "set").unwrap());
+            (m.t.bases[0].obj.bytes.clone(), set)
+        };
+        let (unpinned, read) = base_of("unpinned");
+        assert!(Wai::from_bytes(&unpinned).unwrap().prior().is_none(), "the fixture's base pins nothing");
+        let read = read.expect("its decode read a set");
+        let k = decoder_key(&unpinned, None, false, Some(read), 0).unwrap();
+        assert_eq!((k.pin, k.tier.as_deref()), (Some(read), Some("none")));
+        assert!(decoder_key(&unpinned, None, true, Some(read), 0).unwrap_err().contains("pins no parameter set"));
+        let k = decoder_key(&unpinned, None, false, None, 0).unwrap();
+        assert_eq!((k.pin, k.tier), (None, None), "a decode that read no set");
+        let (pinned, pin) = base_of("still");
+        let pin = pin.expect("the still's base is pinned");
+        assert_eq!(read, pin, "the two bases decode against the same files, which the still's pin names");
+        let k = decoder_key(&pinned, None, true, Some(pin), 0).unwrap();
+        assert_eq!((k.pin, k.tier.as_deref()), (Some(pin), Some("decode-equivalence")));
+        assert_eq!(decoder_key(&pinned, None, true, None, 0).unwrap().pin, Some(pin), "the pin, the set left out");
+        assert!(decoder_key(&pinned, None, true, Some([0; 32]), 0).unwrap_err().starts_with("set:"), "a set other than the pin's");
+    }
+
     /// A page that receives a `WAI2` envelope it does not decode — its payloads plaintext or
     /// encrypted, its rendition table in the clear either way — seals its session: `finish`
     /// states the classical length the envelopes held give (jwp-receipts §2.4.12 rule 9),
@@ -797,7 +849,7 @@ mod tests {
             let w = std::fs::read(wai_dir().join(format!("staged-measure-conformance/counting/{name}.wai"))).unwrap();
             let png = wai::container::WaiMulti::from_bytes(&w).unwrap().renditions[1].len() as u64;
             let (mut r, subject) = begin(base, host_of(HOST).unwrap(), [7; 32], None).unwrap();
-            let h = r.decoder(decoder_key(base, None, true, 0).unwrap());
+            let h = r.decoder(decoder_key(base, None, true, None, 0).unwrap());
             r.presented(p.p.unit, p.p.layers, h, p.p.out_digest, None).unwrap();
             r.received_envelope(base, SessionPath::Primary);
             r.received_envelope(&w, SessionPath::Primary);
diff --git a/wai/wai-web/src/stage.rs b/wai/wai-web/src/stage.rs
index 91d9a5f2a..7def4c134 100644
--- a/wai/wai-web/src/stage.rs
+++ b/wai/wai-web/src/stage.rs
@@ -383,12 +383,16 @@ pub struct StageBase {
     info: BaseInfo,
     units: Vec<UnitBuf>,
     claimed: bool,
+    /// Whether the decode verified the envelope's pin, and the parameter set it read, as
+    /// `staged::BaseDecode` gives them; neither for a claimed base.
+    pinned: bool,
+    set: Option<String>,
 }
 
 #[wasm_bindgen]
 impl StageBase {
     /// Decode a base envelope by SPEC §4 through the integer path
-    /// (`wai::staged::decode_base_via`). `set` maps each parameter file's name (as SPEC
+    /// (`wai::staged::decode_base_pinned`). `set` maps each parameter file's name (as SPEC
     /// §5 names it) to its bytes, and is absent for a capability that reads none
     /// (`wai.video.int_motion`); the envelope's pin is verified against those bytes.
     /// `via` is the capability to decode through: absent to run §4 (the manifest's
@@ -410,8 +414,8 @@ impl StageBase {
         let limit = limit_from(&limit_v).map_err(js_err)?;
         let held = set_from(set)?;
         let files: BTreeMap<String, &[u8]> = held.iter().map(|(k, v)| (k.clone(), v.as_slice())).collect();
-        let (info, units) = staged::decode_base_via(envelope, via.as_deref(), &files, limit.as_ref(), None).map_err(js_err)?;
-        Ok(StageBase { info, units, claimed: false })
+        let d = staged::decode_base_pinned(envelope, via.as_deref(), &files, limit.as_ref(), None).map_err(js_err)?;
+        Ok(StageBase { info: d.info, units: d.units, claimed: false, pinned: d.pin.is_some(), set: d.set })
     }
 
     /// A base the page decoded elsewhere (a codec this build does not run, or a
@@ -424,7 +428,7 @@ impl StageBase {
     #[wasm_bindgen(js_name = fromStage0)]
     pub fn from_stage0(envelope: &[u8], capability: &str, tier_token: &str, shapes: &[u32], elem: &str, stage0: &[u8]) -> Result<StageBase, JsError> {
         let (info, units) = claim_base(envelope, capability, tier_token, shapes, elem, stage0).map_err(js_err)?;
-        Ok(StageBase { info, units, claimed: true })
+        Ok(StageBase { info, units, claimed: true, pinned: false, set: None })
     }
 
     /// The base envelope's content hash, lowercase hex.
@@ -456,6 +460,21 @@ impl StageBase {
     pub fn claimed(&self) -> bool {
         self.claimed
     }
+    /// Whether the decode verified the parameter set the envelope pins (SPEC §3.1).
+    /// `false` for a claimed base, which nothing here decoded.
+    #[wasm_bindgen(getter)]
+    pub fn pinned(&self) -> bool {
+        self.pinned
+    }
+    /// The parameter set the decode read, by the digest a pin over it names (lowercase
+    /// hex; SPEC §3.1, *Pin by default*): the pin's `sha256` where the decode verified one,
+    /// and otherwise the digest a pin over the bytes it read would name. `undefined` for a
+    /// capability that reads no set, and for a claimed base. A session names the decode by
+    /// it (`SessionHandle.decoderFor`, jwp-receipts §2.4.12).
+    #[wasm_bindgen(getter, js_name = paramSet)]
+    pub fn param_set(&self) -> Option<String> {
+        self.set.clone()
+    }
     /// Whether the sink verified a delivery receipt for the base (§10).
     #[wasm_bindgen(getter)]
     pub fn attested(&self) -> bool {
@@ -549,7 +568,7 @@ impl StageSink {
     /// report refuses its envelope `chain-over-limit` (§4.5).
     #[wasm_bindgen(js_name = addBase)]
     pub fn add_base(&mut self, base: StageBase) -> Result<JsValue, JsError> {
-        let StageBase { info, units, claimed } = base;
+        let StageBase { info, units, claimed, .. } = base;
         let h = info.content_hash;
         let held_before = self.sink.chain(&h).is_some();
         let rs = self.sink.add_base(info, units);
-- 
2.54.0 (Apple Git-157)


From 4d0a6251fb8ac2f1af12206f79ba7e6951baf91e Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 02:54:02 -0500
Subject: [PATCH 22/29] wai conform sink: the native sink through the
 conformance runner, and wai verify --conformant admits a result only for a
 run that passed
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

conformance-results §4.1: a relying party that admits an implementation on a
conform-result/1 record requires the record accepted, its run passed (not_conformant), its
suite one whose corpus the party holds (suite_unknown), and its corpus digest that corpus's
(corpus_mismatch). wai verify --conformant [--root <wai-dir>] applies the rules against a
checkout; without it a record of a run that did not pass verifies as before. A sink kind:
wai conform sink --suite native-sink gives a sink each session of
live-conformance/native-sink/ with its answers withheld, and each configuration the corpus
states a session refuses; wai_native_sink_vectors answers [--c-abi] is the reference sink,
through its module or through the C interface the platform integrations link.

wai-breaking.yml's runner job runs the native sink both ways and the score decoder, each
record admitted against the checkout's corpus, and a planted sink that presents an object
whose digest does not match: its run must exit 1 with tampered-fragment failed, and
admission must refuse its record at rule 2. tests/cli_conform_sinks.rs holds the same, and
corpus_mismatch and suite_unknown; it has its [[test]] entry with its required features,
so check-tests-in-ci.py passes. The binaries are built with the test step's features, so
cargo reuses what that step built. The browser sink has no sink suite, and §7 says that
§4.1 has only this reference implementation, with no admission vectors.

The catalogue's receipt.verify (the new options, and the envelopes the pin-by-default commit
added) and conformance.run (the sink kind, the {case} placeholder) are exported again, and
Studio's actions regenerated from the export: actions.mjs --check had failed since that
commit. c_abi::take reads and frees no null buffer.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-breaking.yml            |  54 +++++-
 wai/CHANGELOG.md                              |  21 +++
 wai/README.md                                 |   5 +-
 wai/SPEC.md                                   |   6 +-
 wai/catalogue/ops.json                        |  24 ++-
 wai/extensions/conformance-results.md         |  44 ++++-
 wai/extensions/native-sink.md                 |   5 +
 wai/live-conformance/native-sink/README.md    |  10 ++
 wai/wai-rs/Cargo.toml                         |   7 +
 wai/wai-rs/src/bin/wai_cli/dx/conform/mod.rs  |   4 +-
 .../src/bin/wai_cli/dx/conform/record.rs      |  35 ++++
 .../src/bin/wai_cli/dx/conform/suites.rs      |  50 +++++-
 wai/wai-rs/src/bin/wai_cli/dx/mod.rs          |   6 +-
 wai/wai-rs/src/bin/wai_cli/dx/verify.rs       |  36 +++-
 wai/wai-rs/src/bin/wai_native_sink_vectors.rs | 167 +++++++++++++++++-
 wai/wai-rs/src/catalogue.rs                   |  15 +-
 wai/wai-rs/src/native_sink/corpus.rs          |  24 +++
 wai/wai-rs/tests/cli_conform_sinks.rs         | 163 +++++++++++++++++
 wai/wai-studio/src/actions.generated.js       |   4 +-
 19 files changed, 645 insertions(+), 35 deletions(-)
 create mode 100644 wai/wai-rs/tests/cli_conform_sinks.rs

diff --git a/.github/workflows/wai-breaking.yml b/.github/workflows/wai-breaking.yml
index 8fe49ec75..93d907ce9 100644
--- a/.github/workflows/wai-breaking.yml
+++ b/.github/workflows/wai-breaking.yml
@@ -11,9 +11,13 @@ name: WAI breaking changes and the conformance runner
 #   the room service's OpenAPI document).
 # - runner: the tests of `wai breaking` (each kind of break reported and failing until a
 #   changelog entry names it) and of `wai conform` (the reference implementations conform, a
-#   changed verdict or output byte fails, a changed record is refused by `wai verify`); then
-#   `wai conform` runs the corpora's independent Python verifiers and the reference packager,
-#   and `wai verify` checks every record those runs wrote.
+#   changed verdict or output byte fails, a changed record is refused by `wai verify`, and
+#   `wai verify --conformant` admits a passing run of the native sink and refuses a planted
+#   failing one); then `wai conform` runs the corpora's independent Python verifiers and the
+#   reference packager, and `wai verify` checks every record those runs wrote; and it runs the
+#   native sink, through its module and its C interface, and the score decoder, each record
+#   admitted against this checkout's corpus and a planted failing sink refused
+#   (conformance-results §4.1).
 #
 # Push filters match the push's delta, and the `relevant` job gates the work from the pull
 # request's own diff (scripts/pr-touches-paths.py, reading the list below), as wai-pack.yml
@@ -128,15 +132,18 @@ jobs:
           git config --global --add safe.directory "$GITHUB_WORKSPACE"
           git config --global user.email ci@example.invalid
           git config --global user.name ci
-          cargo test --locked --no-default-features --features quickstart,score,key_discovery \
-            --bin wai --test cli_breaking --test cli_conform
+          cargo test --locked --no-default-features --features quickstart,score,key_discovery,native_sink \
+            --bin wai --test cli_breaking --test cli_conform --test cli_conform_sinks
       - name: Build wai and the reference packager
         shell: bash
         run: |
           set -euo pipefail
           out() { (cd "$1" && cargo metadata --format-version 1 --no-deps --locked | python3 -c 'import json,sys; print(json.load(sys.stdin)["target_directory"])'); }
           mkdir -p "$RUNNER_TEMP/bin"
-          (cd wai/wai-rs && cargo build --locked --no-default-features --features quickstart --bin wai) && cp "$(out wai/wai-rs)/debug/wai" "$RUNNER_TEMP/bin/"
+          # The test step's features, so cargo reuses what it built.
+          (cd wai/wai-rs && cargo build --locked --no-default-features --features quickstart,score,key_discovery,native_sink \
+            --bin wai --bin wai_native_sink_vectors)
+          cp "$(out wai/wai-rs)/debug/wai" "$(out wai/wai-rs)/debug/wai_native_sink_vectors" "$RUNNER_TEMP/bin/"
           (cd wai/wai-pack && cargo build --locked --bin wai-pack) && cp "$(out wai/wai-pack)/debug/wai-pack" "$RUNNER_TEMP/bin/"
       # The second implementations of the claim, JWP and key-discovery corpora, written from the
       # extension texts and sharing no code with wai-rs, run through the conformance runner as
@@ -156,3 +163,38 @@ jobs:
           "$w" conform packager --suite policy --text --root wai \
             --impl "$RUNNER_TEMP/bin/wai-pack policy-verdict {in}" --out "$d/policy.wai-conform.json"
           "$w" verify "$d"
+      # The native sink through the conformance runner, as another implementation would be run:
+      # through its Rust module and through the C interface its platform integrations link; and
+      # the score decoder. Each run must be conformant, and `wai verify
+      # --conformant` must admit each record against this checkout's corpus. A planted sink that
+      # presents an object whose digest does not match must be refused: its run fails, its record
+      # still verifies, and admission refuses it (not_conformant).
+      - name: wai conform with the native sink and the score decoder, each record admitted, a planted failing sink refused
+        shell: bash
+        run: |
+          set -euo pipefail
+          w="$RUNNER_TEMP/bin/wai"; s="$RUNNER_TEMP/bin/wai_native_sink_vectors"; d="$RUNNER_TEMP/sinks"; mkdir -p "$d"
+          "$w" conform sink --suite native-sink --text --root wai --impl "$s answers {case}" --out "$d/native-sink.wai-conform.json"
+          "$w" conform sink --suite native-sink --text --root wai --impl "$s answers --c-abi {case}" --out "$d/native-sink-c-abi.wai-conform.json"
+          "$w" conform decoder --suite score --text --root wai --impl "$w score render {in} --pcm {out}" --out "$d/score.wai-conform.json"
+          "$w" verify --conformant --root wai "$d"
+          # The planted sink's run must fail as a run (exit 1), not as a usage error, a missing
+          # program or an adapter that does not answer, and on the case it tampers with; its
+          # record must be refused at rule 2, not at another.
+          p="$RUNNER_TEMP/planted"; mkdir -p "$p"
+          planted="$s answers {case} | sed '/object_digest_mismatch/s/\"present\":false/\"present\":true/'"
+          set +e
+          "$w" conform sink --suite native-sink --text --root wai --impl "$planted" --out "$p/planted.wai-conform.json" > "$RUNNER_TEMP/planted-run.txt" 2>&1
+          c=$?
+          set -e
+          cat "$RUNNER_TEMP/planted-run.txt"
+          [ "$c" = 1 ] || { echo "the planted failing sink: wai conform exited $c, not 1"; exit 1; }
+          grep -q 'tampered-fragment' "$RUNNER_TEMP/planted-run.txt" || { echo "the planted failing sink: tampered-fragment is not a failed case"; exit 1; }
+          "$w" verify "$p"
+          set +e
+          "$w" verify --conformant --root wai "$p" > "$RUNNER_TEMP/planted-admit.txt" 2>&1
+          c=$?
+          set -e
+          cat "$RUNNER_TEMP/planted-admit.txt"
+          [ "$c" = 1 ] || { echo "the planted failing sink: wai verify --conformant exited $c, not 1"; exit 1; }
+          grep -q 'FAIL admission rule 2 not_conformant' "$RUNNER_TEMP/planted-admit.txt" || { echo "the planted failing sink was not refused at rule 2"; exit 1; }
diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index 02fe293ba..e2ffa2385 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -6,6 +6,27 @@ wire format may change and such changes are recorded here.
 
 ## Unreleased
 
+### The native sink through the conformance runner, and a result that admits an implementation only for a run that passed
+
+- **Admitting an implementation** ([extensions/conformance-results.md](extensions/conformance-results.md) §4.1). A
+  relying party that admits an implementation on a `conform-result/1` record, such as a certification of a sink or a
+  gate in a build, requires the record accepted (§4), its run passed (`not_conformant`), its suite one whose corpus it
+  holds (`suite_unknown`), and its corpus digest that corpus's (`corpus_mismatch`). `wai verify --conformant [--root
+  <wai-dir>]` applies the rules against the corpus of a checkout. A record of a run that did not pass is a genuine
+  record, and `wai verify` without `--conformant` accepts it as before.
+- **A sink kind.** `wai conform sink --suite native-sink` gives a native sink every session of
+  `live-conformance/native-sink/` with its answers withheld, and each configuration the corpus states a session
+  refuses; a case passes when the sink's answers, one RFC 8785 text a line, have the SHA-256 of the answers the corpus
+  states. `wai_native_sink_vectors answers [--c-abi] <case>` is the reference sink, through its module or through the C
+  interface the platform integrations link.
+- **The native sink and the score decoder in CI** (`wai-breaking.yml`, the runner job): the native sink through its
+  module and through its C interface, and the score decoder (`decoder --suite score`), each run conformant and each
+  record admitted against the checkout's corpus. A planted sink that presents an object whose digest does not match
+  fails its run, on that case; its record verifies, and admission refuses it at rule 2. `tests/cli_conform_sinks.rs`
+  holds the same, and refuses a run against a copy of the corpus changed to suit the sink (`corpus_mismatch`) and a
+  record of a suite the build does not run (`suite_unknown`). The browser sink has no `sink` suite:
+  `wai_browser_sink_vectors verify` and `browser_sink_verify.py` hold its corpus, as before.
+
 ### Pin by default: an encoder pins the set a payload was coded against, and a verifier reports an envelope that does not
 
 - **SPEC §3.1, *Pin by default*.** An encoder SHOULD pin the parameter set a payload was coded against wherever a
diff --git a/wai/README.md b/wai/README.md
index 15f655a79..419005774 100644
--- a/wai/README.md
+++ b/wai/README.md
@@ -107,8 +107,9 @@ wai verify claim.wai-conform.json      # the signed result, checked
 
 A break passes only when [`CHANGELOG.md`](CHANGELOG.md) names it: an entry this change adds that
 says **Breaking** and gives the finding's id in backticks. `wai conform --list` names every kind
-(verifier, decoder, packager, origin) and suite, with its adapters: a command line, or an
-`http://` URL.
+(verifier, decoder, packager, origin, sink) and suite, with its adapters: a command line, or an
+`http://` URL. `wai verify --conformant` admits a result only for a run that passed, against the
+corpus of the checkout it runs in.
 
 With the `codecs` feature, `wai-rs/build.rs` finds libFLAC through pkg-config
 and, on macOS without pkg-config, in the Homebrew "flac" formula prefix
diff --git a/wai/SPEC.md b/wai/SPEC.md
index 9538480bc..da00fc368 100644
--- a/wai/SPEC.md
+++ b/wai/SPEC.md
@@ -2645,8 +2645,10 @@ it does for plaintext; in Lane A there is no output for it to cover.
   implementation, is the signed record `conform-result/1` of
   [conformance-results](extensions/conformance-results.md) §2. A verifier checks its signature
   and that its summary and its verdict are what its cases say
-  ([conformance-results](extensions/conformance-results.md) §4). `wai conform` writes such
-  records and `wai verify` checks them.
+  ([conformance-results](extensions/conformance-results.md) §4). A party that admits an
+  implementation on one, a sink among them, also requires that its run passed, against the
+  corpus the party holds (§4.1 there). `wai conform` writes such records, and `wai verify`
+  checks them (`--conformant`: admits them).
 
 ### Replicate-soundness — the integer-entropy contract (normative)
 
diff --git a/wai/catalogue/ops.json b/wai/catalogue/ops.json
index 93346d068..f70d733c8 100644
--- a/wai/catalogue/ops.json
+++ b/wai/catalogue/ops.json
@@ -422,8 +422,9 @@
       "cli": "wai verify",
       "aliases": [],
       "usage": [
-        "wai verify <path>...   the receipts (receipt indexes *.wai.receipts), conformance results (*.wai-conform.json),",
-        "    fleet report bundles (*.wai-fleet.json) and envelopes' pins (*.wai) under each path, checked offline"
+        "wai verify <path>... [--conformant [--root <wai-dir>]]   the receipts (receipt indexes *.wai.receipts), conformance",
+        "    results (*.wai-conform.json), fleet report bundles (*.wai-fleet.json) and envelopes' pins (*.wai) under each path,",
+        "    checked offline; --conformant admits a result only for a run that passed against this checkout's corpus"
       ],
       "placement": "local",
       "effect": "reads",
@@ -449,6 +450,21 @@
             "x-cli": {
               "positional": true
             }
+          },
+          "conformant": {
+            "type": "boolean",
+            "description": "Admit each conformance result only when its run passed, against the corpus of a checkout.",
+            "x-cli": {
+              "flag": "--conformant"
+            }
+          },
+          "root": {
+            "type": "string",
+            "format": "path",
+            "description": "With --conformant, the wai/ directory of the checkout whose corpus a result names.",
+            "x-cli": {
+              "flag": "--root"
+            }
           }
         },
         "required": [
@@ -3466,7 +3482,7 @@
         "properties": {
           "kind": {
             "type": "string",
-            "description": "The kind of implementation (wai conformance list): a decoder, a verifier, an origin, a packager.",
+            "description": "The kind of implementation (wai conformance list): a decoder, a verifier, an origin, a packager, a sink.",
             "x-cli": {
               "positional": true
             }
@@ -3480,7 +3496,7 @@
           },
           "impl": {
             "type": "string",
-            "description": "The implementation: a command with {in}/{out}/{corpus} placeholders, or a URL.",
+            "description": "The implementation: a command with {case}/{in}/{out}/{corpus} placeholders, or a URL.",
             "x-cli": {
               "flag": "--impl"
             }
diff --git a/wai/extensions/conformance-results.md b/wai/extensions/conformance-results.md
index e2ea5200e..677770198 100644
--- a/wai/extensions/conformance-results.md
+++ b/wai/extensions/conformance-results.md
@@ -18,7 +18,8 @@
 ## 1. Scope
 
 In scope: the record (§2), its signing input and signature (§3), what a verifier checks and in
-what order (§4), and the codes it refuses with (§5).
+what order (§4), what a relying party that admits an implementation on a record checks besides
+(§4.1), and the codes a verifier refuses a record with (§5).
 
 Out of scope: how a run reaches an implementation (a command line, a URL), which corpora and
 suites exist, and what makes an implementation conformant beyond the run a record states. A
@@ -34,7 +35,7 @@ Every other number, and every member not listed, makes the record malformed.
 | member | type | value |
 |---|---|---|
 | `wai` | string | `conform-result/1`, the record's kind and version |
-| `kind` | string | the kind of implementation run: `verifier`, `decoder`, `packager`, `origin`, or another kind a later revision names |
+| `kind` | string | the kind of implementation run: `verifier`, `decoder`, `packager`, `origin`, `sink`, or another kind a later revision names |
 | `suite` | string | the suite that ran |
 | `standard` | string | the version of WAI whose corpus ran (`v0.4.0`) |
 | `toolchain` | object | the program that ran the corpus: `name` and `version`, each a string |
@@ -103,9 +104,38 @@ writes its outcome as `accept conformant=true`, `accept conformant=false`, or
 `reject step=<n> reason=<code>`. A record that says its run was not conformant is a genuine
 record of that run, and is accepted as one.
 
+### 4.1 Admitting an implementation
+
+Acceptance shows that a record is intact and consistent; it admits nothing (§8). A relying
+party that admits an implementation on a record, such as a certification of a sink or a gate in
+a build, holds the corpus the record names, and applies these rules in this order. It refuses
+the implementation at the first rule that fails, with the code that rule names:
+
+1. The record is accepted (§4). Otherwise the code of the step of §4 that refused it.
+2. Its `conformant` is `true` (`not_conformant`). A record of a run that did not pass is
+   genuine, and admits nothing.
+3. Its `kind` and `suite` name a suite whose corpus the relying party holds and reads
+   (`suite_unknown`).
+4. Its `corpus.digest` is the corpus digest (§2.1) of that suite's files as the relying party
+   holds them (`corpus_mismatch`). A run against another corpus, an earlier revision's or one
+   changed to suit an implementation, admits nothing.
+
+These rules read the record and the corpus the relying party holds, and nothing that depends on
+the runner, so two relying parties that hold the same corpus admit the same records. They do not
+say who ran the run: a relying party that requires a known runner also requires `key` to be
+`supplied` and confirms `signer_pubkey` (§6). `wai verify --conformant` applies them, against
+the corpus of a checkout of the standard.
+
+| reason | raised at |
+|---|---|
+| `not_conformant` | §4.1 rule 2 |
+| `suite_unknown` | §4.1 rule 3 |
+| `corpus_mismatch` | §4.1 rule 4 |
+
 ## 5. Reasons
 
-The codes this extension defines, each with the step that raises it:
+The codes a verifier refuses a record with, each with the step of §4 that raises it (§4.1
+names its own):
 
 | reason | raised at |
 |---|---|
@@ -139,6 +169,14 @@ digest out of form, bytes that are not JSON), each with the outcome §4 gives it
 `wai_conform_result_vectors verify` (in `wai-rs`, feature `conform_result`) and the independent
 `conform_result_verify.py`, which shares no code with `wai-rs`, reach every vector's outcome.
 
+The rules of §4.1 are held by `wai-rs`'s `tests/cli_conform_sinks.rs`, over records of runs of the
+native sink, through its module and its C interface, and of the score decoder: each is admitted; a
+sink that presents an object whose digest does not match is refused `not_conformant`; the same sink
+run against a copy of the corpus changed to suit it is admitted by a party holding that copy and
+refused `corpus_mismatch` against WAI's; and a record of a suite the build does not run is refused
+`suite_unknown`. §4.1 has that reference implementation only: this corpus holds no admission
+vectors, and `conform_result_verify.py` does not apply its rules.
+
 ## 8. Security and privacy considerations
 
 - **What a record reveals.** The command line and the paths of the files it names are the
diff --git a/wai/extensions/native-sink.md b/wai/extensions/native-sink.md
index 9db88b98a..551c93fa8 100644
--- a/wai/extensions/native-sink.md
+++ b/wai/extensions/native-sink.md
@@ -353,6 +353,11 @@ version.
   and every one of its checks dropped.
 - A native integration's own tests drive it through its platform's player and require that every
   object the player was handed is one the sink presented, and that the bundle verifies.
+- `wai conform sink --suite native-sink` runs the corpus against a sink, each case with its answers
+  withheld, and writes the run's outcome as a signed `conform-result/1` record
+  ([conformance-results](conformance-results.md)). WAI's own sink is run so in CI, through the
+  reference module and through the C interface of §9, and each record must be admitted
+  (conformance-results §4.1).
 
 ## 11. Security and privacy considerations
 
diff --git a/wai/live-conformance/native-sink/README.md b/wai/live-conformance/native-sink/README.md
index a1fbd3523..47c6ea241 100644
--- a/wai/live-conformance/native-sink/README.md
+++ b/wai/live-conformance/native-sink/README.md
@@ -52,3 +52,13 @@ Rebuilding: `cd wai/wai-rs && cargo run --no-default-features --features native_
 wai_native_sink_vectors gen ../live-conformance/native-sink`; `… verify ../live-conformance/native-sink`
 checks it; `python3 -I ../live-conformance/native_sink_verify.py ../live-conformance/native-sink` is
 the second implementation.
+
+**Running a sink against it.** `wai conform sink --suite native-sink --impl "<command> {case}"` gives a
+sink each case with its answers withheld: the bodies its calls name (`objects`, hexadecimal), its
+`config`, `seed` and `start_us`, and each call's `do`; and each configuration under `configs` as a case
+of no call, under a seed of zeros. The sink writes its answer to each call in order, or, refusing the
+configuration, that refusal alone, one JSON text a line in RFC 8785 form. A case passes when what it
+writes has the SHA-256 of the answers this file states, written the same way, and the run's outcome is
+a signed `conform-result/1` record ([conformance-results](../../extensions/conformance-results.md)).
+`wai_native_sink_vectors answers {case}` is the reference sink, and `… answers --c-abi {case}` the same
+sink through its C interface.
diff --git a/wai/wai-rs/Cargo.toml b/wai/wai-rs/Cargo.toml
index 5f750822b..f8e8008ac 100644
--- a/wai/wai-rs/Cargo.toml
+++ b/wai/wai-rs/Cargo.toml
@@ -1610,6 +1610,13 @@ name = "cli_conform"
 path = "tests/cli_conform.rs"
 required-features = ["quickstart", "score", "key_discovery"]
 
+# The native sink through `wai conform`, its reference module and its C interface
+# (wai_native_sink_vectors), and the score decoder.
+[[test]]
+name = "cli_conform_sinks"
+path = "tests/cli_conform_sinks.rs"
+required-features = ["quickstart", "score", "native_sink"]
+
 [[test]]
 name = "cli_fleet"
 path = "tests/cli_fleet.rs"
diff --git a/wai/wai-rs/src/bin/wai_cli/dx/conform/mod.rs b/wai/wai-rs/src/bin/wai_cli/dx/conform/mod.rs
index a90e9a060..6b2809ae3 100644
--- a/wai/wai-rs/src/bin/wai_cli/dx/conform/mod.rs
+++ b/wai/wai-rs/src/bin/wai_cli/dx/conform/mod.rs
@@ -51,7 +51,7 @@ fn cut(s: &str) -> String {
 
 /// The `wai/` directory of a checkout: `--root`, else the first directory up from here
 /// that is one, or holds one.
-fn wai_root(given: Option<&str>) -> Result<PathBuf, OpError> {
+pub(crate) fn wai_root(given: Option<&str>) -> Result<PathBuf, OpError> {
     let is_root = |d: &Path| d.join("SPEC.md").is_file() && d.join("registry").join("capabilities.json").is_file();
     if let Some(g) = given {
         let d = PathBuf::from(g);
@@ -106,7 +106,7 @@ fn now_rfc3339() -> String {
 }
 
 /// The corpus digest (conformance-results §2.1) of the files a run read.
-fn corpus_digest(root: &Path, files: &[PathBuf]) -> Result<String, OpError> {
+pub(crate) fn corpus_digest(root: &Path, files: &[PathBuf]) -> Result<String, OpError> {
     let mut read: Vec<(String, Vec<u8>)> = Vec::new();
     for f in files {
         let b = std::fs::read(f).map_err(|e| OpError::new(EXIT_NOT_FOUND, "corpus-file-missing", format!("{}: {e}", f.display()), "the corpus is incomplete; check out the standard again"))?;
diff --git a/wai/wai-rs/src/bin/wai_cli/dx/conform/record.rs b/wai/wai-rs/src/bin/wai_cli/dx/conform/record.rs
index 3c77c74f3..4154a6f15 100644
--- a/wai/wai-rs/src/bin/wai_cli/dx/conform/record.rs
+++ b/wai/wai-rs/src/bin/wai_cli/dx/conform/record.rs
@@ -45,6 +45,41 @@ pub fn sign(record: Map<String, Value>, key: &SigningKey) -> Result<Map<String,
     conform_result::sign(record, key)
 }
 
+/// A result record checked (§4), then held to §4.1 as a relying party that admits an
+/// implementation on it: the run passed (rule 2), its kind and suite are one this build runs
+/// and whose corpus `root` (a checkout's `wai/` directory) holds (rule 3), and its corpus
+/// digest is that corpus's (rule 4). Whether it is admitted, and a line for each thing
+/// checked.
+pub fn admit(bytes: &[u8], root: &std::path::Path) -> (bool, Vec<String>) {
+    let (_, mut lines) = check(bytes);
+    let Ok(a) = conform_result::check(bytes) else {
+        return (false, lines);
+    };
+    let mut refuse = |rule: u8, code: &str, what: String| {
+        lines.push(format!("FAIL admission rule {rule} {code}: {what}"));
+        (false, std::mem::take(&mut lines))
+    };
+    if !a.conformant {
+        return refuse(2, "not_conformant", format!("the run it records did not pass ({} of {} cases failed)", a.cases - a.passed, a.cases));
+    }
+    let Some(suite) = super::suites::SUITES.iter().find(|s| s.kind == a.kind && s.name == a.suite) else {
+        return refuse(3, "suite_unknown", format!("{} {} is no suite this build runs", a.kind, a.suite));
+    };
+    let files = match (suite.load)(&root.join(suite.dir)) {
+        Ok(l) => l.files,
+        Err(e) => return refuse(3, "suite_unknown", format!("the corpus of {} {} does not read here: {e}", a.kind, a.suite)),
+    };
+    let here = match super::corpus_digest(root, &files) {
+        Ok(d) => d,
+        Err(e) => return refuse(3, "suite_unknown", e.message),
+    };
+    if here != a.corpus_digest {
+        return refuse(4, "corpus_mismatch", format!("it names the corpus {}, and the corpus of {} {} here is {here}", a.corpus_digest, a.kind, a.suite));
+    }
+    lines.push(format!("ok   admitted: {} {} passed against the corpus of {}", a.kind, a.suite, root.display()));
+    (true, lines)
+}
+
 /// A result record checked (§4): whether it holds, and a line for each thing checked.
 pub fn check(bytes: &[u8]) -> (bool, Vec<String>) {
     match conform_result::check(bytes) {
diff --git a/wai/wai-rs/src/bin/wai_cli/dx/conform/suites.rs b/wai/wai-rs/src/bin/wai_cli/dx/conform/suites.rs
index 818f099d1..798e83ddb 100644
--- a/wai/wai-rs/src/bin/wai_cli/dx/conform/suites.rs
+++ b/wai/wai-rs/src/bin/wai_cli/dx/conform/suites.rs
@@ -11,6 +11,7 @@
 //! | decoder | `score` | `score-conformance/` | a `WSCR` score | the rendered `i16le` PCM has the declared SHA-256 |
 //! | packager | `policy` | `pack-conformance/` | a policy file | it is admitted with the declared digest, or refused with the declared reason (packaging §3, §12) |
 //! | origin | `http` | `http-conformance/objects/` | an object | it is served as http-delivery §5.3 and §6.2 require: bytes, headers, ranges, head proof, preflight |
+//! | sink | `native-sink` | `live-conformance/native-sink/vectors.json` | one session, its answers withheld, or one configuration | the answers the sink writes, one a line in RFC 8785 form, have the SHA-256 of the answers the corpus states (native-sink §10) |
 
 use serde_json::Value;
 use std::path::{Path, PathBuf};
@@ -66,7 +67,7 @@ pub struct Suite {
     pub load: fn(&Path) -> Result<Loaded, String>,
 }
 
-pub const KINDS: &[&str] = &["verifier", "decoder", "packager", "origin"];
+pub const KINDS: &[&str] = &["verifier", "decoder", "packager", "origin", "sink"];
 
 pub const SUITES: &[Suite] = &[
     Suite { kind: "verifier", name: "claim", dir: "claim-conformance", about: "typed claim receipts (jwp-receipts §3.1): accept with the link hash, or reject at a step with a reason", http: true, command: true, load: claim },
@@ -76,6 +77,7 @@ pub const SUITES: &[Suite] = &[
     Suite { kind: "decoder", name: "score", dir: "score-conformance", about: "a WSCR score rendered to its canonical i16le PCM (the score extension's mixdown-equivalence)", http: true, command: true, load: score },
     Suite { kind: "packager", name: "policy", dir: "pack-conformance", about: "packaging policies admitted with their digest, or refused with their reason (packaging §3, §12)", http: true, command: true, load: policy },
     Suite { kind: "origin", name: "http", dir: "http-conformance", about: "objects, head proofs, ranges and headers as an origin serves them (http-delivery §5.3, §6.2)", http: true, command: false, load: http_origin },
+    Suite { kind: "sink", name: "native-sink", dir: "live-conformance/native-sink", about: "a native sink's answer to every call of a session, byte for byte, and its refusal of a configuration (native-sink §10)", http: true, command: true, load: native_sink },
 ];
 
 fn read_json(p: &Path) -> Result<Value, String> {
@@ -244,6 +246,52 @@ fn http_origin(dir: &Path) -> Result<Loaded, String> {
     Ok(Loaded { cases, files, corpus_file: None })
 }
 
+/// The native sink's corpus (native-sink §10). Each case is given to the sink with its answers
+/// withheld: the bodies its calls name (`objects`, hexadecimal), its `config`, `seed` and
+/// `start_us`, and each call's `do`. Each configuration the corpus states a session refuses is a
+/// case of no call, under a seed of zeros. The sink opens the session and writes its answer to
+/// each call in order, or, refusing the configuration, that refusal alone (`{"error":
+/// <reason>}`): one JSON text a line, in RFC 8785 form. A case passes when what it writes has the
+/// SHA-256 of the answers the corpus states, written the same way.
+fn native_sink(dir: &Path) -> Result<Loaded, String> {
+    let p = dir.join("vectors.json");
+    let doc = read_json(&p)?;
+    let objects = doc.get("objects").and_then(Value::as_object).ok_or_else(|| format!("{}: no `objects`", p.display()))?;
+    let digest = |answers: &[&Value]| -> Result<String, String> {
+        let mut text = String::new();
+        for a in answers {
+            text.push_str(&wai::jcs::jcs(a).map_err(|e| format!("{}: an answer outside RFC 8785's subset: {e}", p.display()))?);
+            text.push('\n');
+        }
+        Ok(super::adapter::sha256_hex(text.as_bytes()))
+    };
+    let mut cases = Vec::new();
+    for c in doc.get("cases").and_then(Value::as_array).ok_or_else(|| format!("{}: no `cases`", p.display()))? {
+        let calls = c.get("calls").and_then(Value::as_array).ok_or_else(|| format!("{}: {} has no calls", p.display(), name_of(c)))?;
+        let mut named = serde_json::Map::new();
+        for d in calls.iter().map(|x| &x["do"]) {
+            for k in ["body", "receipt"] {
+                if let Some(n) = d.get(k).and_then(Value::as_str) {
+                    let b = objects.get(n).ok_or_else(|| format!("{}: {} names no object {n}", p.display(), name_of(c)))?;
+                    named.insert(n.to_string(), b.clone());
+                }
+            }
+        }
+        let input = serde_json::json!({
+            "objects": named, "config": c["config"], "seed": c["seed"], "start_us": c["start_us"],
+            "calls": calls.iter().map(|x| serde_json::json!({"do": x["do"]})).collect::<Vec<_>>(),
+        });
+        let answers: Vec<&Value> = calls.iter().map(|x| &x["answer"]).collect();
+        cases.push(Case { name: name_of(c), input: Input::Json(input), expect: Expect::Output { sha256: digest(&answers)? } });
+    }
+    for (i, c) in doc.get("configs").and_then(Value::as_array).into_iter().flatten().enumerate() {
+        let input = serde_json::json!({"objects": {}, "config": c["config"], "seed": "0".repeat(64), "start_us": 0, "calls": []});
+        let refused = serde_json::json!({"error": c["refused"]});
+        cases.push(Case { name: format!("configs/{i}"), input: Input::Json(input), expect: Expect::Output { sha256: digest(&[&refused])? } });
+    }
+    Ok(Loaded { cases, files: vec![p], corpus_file: None })
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
diff --git a/wai/wai-rs/src/bin/wai_cli/dx/mod.rs b/wai/wai-rs/src/bin/wai_cli/dx/mod.rs
index b17c6dff9..aed0a6449 100644
--- a/wai/wai-rs/src/bin/wai_cli/dx/mod.rs
+++ b/wai/wai-rs/src/bin/wai_cli/dx/mod.rs
@@ -6,8 +6,10 @@
 //!   (`wai/examples/`), or from an app made with Create (`--from app.oui`, `app.export`);
 //! - `dev.start` (`wai dev`): a relay, a room service and the room page on one origin, on this
 //!   machine;
-//! - `receipt.verify` (`wai verify`): the receipts under a directory, conformance results and
-//!   fleet report bundles, checked offline;
+//! - `receipt.verify` (`wai verify`): the receipts under a directory, conformance results,
+//!   fleet report bundles and envelopes' pins, checked offline; with `--conformant`, each
+//!   conformance result admits its implementation only for a run that passed against a
+//!   checkout's corpus;
 //! - `compat.check` (`wai breaking`): what the working tree breaks against a git ref;
 //! - `conformance.run` (`wai conform`): WAI's conformance corpus run against another
 //!   implementation, its result a signed record; `conformance.list` (`wai conformance list`,
diff --git a/wai/wai-rs/src/bin/wai_cli/dx/verify.rs b/wai/wai-rs/src/bin/wai_cli/dx/verify.rs
index 1be238415..d60824b8f 100644
--- a/wai/wai-rs/src/bin/wai_cli/dx/verify.rs
+++ b/wai/wai-rs/src/bin/wai_cli/dx/verify.rs
@@ -10,7 +10,9 @@
 //!
 //! It also checks every conformance result record (`*.wai-conform.json`, written by `wai
 //! conform`) under each path: its signature, and that its summary and verdict are what its
-//! cases say.
+//! cases say. With `--conformant`, each record must also admit its implementation
+//! (conformance-results §4.1): the run passed, against the corpus of a checkout of the
+//! standard (`--root <wai-dir>`, or found from the current directory), named by its digest.
 //!
 //! And every envelope (`*.wai`) under each path: what a verifier finds about its pins
 //! without holding a parameter set (SPEC §3.1, §7), which makes it non-conforming, and the
@@ -31,10 +33,27 @@ pub fn main(_: &[String]) -> Result<bool, OpError> {
 pub fn main(args: &[String]) -> Result<bool, OpError> {
     use std::path::{Path, PathBuf};
     let usage = super::usage("receipt.verify");
-    if let Some(a) = args.iter().find(|a| a.starts_with("--")) {
-        return Err(OpError::usage(format!("wai verify: {a}: not an option ({usage})"), ""));
+    let mut i = 0;
+    let (mut conformant, mut root) = (false, None);
+    while i < args.len() {
+        match args[i].as_str() {
+            "--conformant" => conformant = true,
+            "--root" => {
+                root = Some(args.get(i + 1).ok_or_else(|| OpError::usage(format!("wai verify: --root takes the wai/ directory of a checkout ({usage})"), ""))?.as_str());
+                i += 1;
+            }
+            a if a.starts_with("--") => return Err(OpError::usage(format!("wai verify: {a}: not an option ({usage})"), "")),
+            _ => {}
+        }
+        i += 1;
     }
-    let paths = super::positional(args, &[]);
+    if root.is_some() && !conformant {
+        return Err(OpError::usage(format!("wai verify: --root names the corpus --conformant holds a record to ({usage})"), ""));
+    }
+    // Admission compares a record's corpus digest with the corpus of a checkout: without
+    // one there is nothing to compare it with.
+    let root = if conformant { Some(super::conform::wai_root(root)?) } else { None };
+    let paths = super::positional(args, &["--root"]);
     if paths.is_empty() {
         return Err(OpError::usage(usage, ""));
     }
@@ -80,7 +99,11 @@ pub fn main(args: &[String]) -> Result<bool, OpError> {
         println!("{}", ix.display());
         let (ok, lines) = if ix.to_string_lossy().ends_with(super::conform::record::SUFFIX) {
             n_rec += 1;
-            super::conform::record::check(&std::fs::read(ix).map_err(|e| format!("{}: {e}", ix.display()))?)
+            let bytes = std::fs::read(ix).map_err(|e| format!("{}: {e}", ix.display()))?;
+            match root.as_deref() {
+                Some(r) => super::conform::record::admit(&bytes, r),
+                None => super::conform::record::check(&bytes),
+            }
         } else if ix.to_string_lossy().ends_with(".wai") {
             n_env += 1;
             check_envelope(&std::fs::read(ix).map_err(|e| format!("{}: {e}", ix.display()))?)
@@ -116,6 +139,9 @@ pub fn main(args: &[String]) -> Result<bool, OpError> {
         what.push(format!("{m} fleet bundle{}", if m == 1 { "" } else { "s" }));
     }
     println!("wai verify: {} {}", what.join(" and "), if all { "verified" } else { "with a refusal (above)" });
+    if let (Some(r), true, true) = (root.as_deref(), all, n_rec > 0) {
+        println!("wai verify: every conformance result admits its implementation (conformance-results \u{a7}4.1), against the corpus of {}", r.display());
+    }
     Ok(all)
 }
 
diff --git a/wai/wai-rs/src/bin/wai_native_sink_vectors.rs b/wai/wai-rs/src/bin/wai_native_sink_vectors.rs
index b22136f0d..7b00f9901 100644
--- a/wai/wai-rs/src/bin/wai_native_sink_vectors.rs
+++ b/wai/wai-rs/src/bin/wai_native_sink_vectors.rs
@@ -4,6 +4,13 @@
 //!   wai_native_sink_vectors gen     <corpus-dir>   write vectors.json, from fixed seeds
 //!   wai_native_sink_vectors verify  <corpus-dir>   vectors.json regenerates byte for byte,
 //!                                                 and every call gets the answer it states
+//!   wai_native_sink_vectors answers [--c-abi] <case.json>
+//!                                                 the sink's answer to each call of one case
+//!                                                 as `wai conform sink --suite native-sink`
+//!                                                 gives it, its answers withheld: one JSON
+//!                                                 text a line, RFC 8785; through the Rust
+//!                                                 module, or with `--c-abi` through the C
+//!                                                 interface a native integration links
 //!
 //! The generator and this verifier are one implementation. The corpus's second
 //! implementation is `live-conformance/native_sink_verify.py`, written from the
@@ -14,6 +21,17 @@ use wai::native_sink::corpus;
 
 fn main() {
     let a: Vec<String> = std::env::args().collect();
+    if a.get(1).map(String::as_str) == Some("answers") {
+        let (abi, file) = match (a.get(2).map(String::as_str), a.get(3)) {
+            (Some("--c-abi"), Some(f)) => (true, f),
+            (Some(f), None) if f != "--c-abi" => (false, &a[2]),
+            _ => {
+                eprintln!("usage: wai_native_sink_vectors answers [--c-abi] <case.json>");
+                std::process::exit(2);
+            }
+        };
+        std::process::exit(answers(Path::new(file), abi));
+    }
     match (a.get(1).map(String::as_str), a.get(2)) {
         (Some("gen"), Some(d)) => {
             let p = Path::new(d).join("vectors.json");
@@ -25,7 +43,7 @@ fn main() {
         }
         (Some("verify"), Some(d)) => std::process::exit(verify(Path::new(d))),
         _ => {
-            eprintln!("usage: wai_native_sink_vectors gen|verify <corpus-dir>");
+            eprintln!("usage: wai_native_sink_vectors gen|verify <corpus-dir> | answers [--c-abi] <case.json>");
             std::process::exit(2);
         }
     }
@@ -52,3 +70,150 @@ fn verify(dir: &Path) -> i32 {
     fail += r.is_err() as i32;
     (fail != 0) as i32
 }
+
+/// One case's answers, each on its own line in RFC 8785 form: the conformance runner hashes
+/// what is written and compares it with the answers the corpus states.
+fn answers(file: &Path, abi: bool) -> i32 {
+    let case: serde_json::Value = match std::fs::read(file).map_err(|e| e.to_string()).and_then(|b| serde_json::from_slice(&b).map_err(|e| e.to_string())) {
+        Ok(v) => v,
+        Err(e) => {
+            eprintln!("{}: {e}", file.display());
+            return 1;
+        }
+    };
+    let got = if abi { c_abi::answer_case(&case) } else { corpus::answer_case(&case) };
+    match got.and_then(|v| v.iter().map(|a| wai::jcs::jcs(a).map_err(|e| e.to_string())).collect::<Result<Vec<_>, _>>()) {
+        Ok(lines) => {
+            for l in lines {
+                println!("{l}");
+            }
+            0
+        }
+        Err(e) => {
+            eprintln!("{}: {e}", file.display());
+            1
+        }
+    }
+}
+
+/// The native sink through its C interface (`ffi_sink`, extensions/native-sink.md §9), as a
+/// native integration links it: every call made through the C symbols, and a refusal read
+/// from `wai_last_error`.
+mod c_abi {
+    use serde_json::{json, Value};
+    use std::ffi::{CStr, CString};
+    use wai::ffi::{wai_buffer_free, wai_last_error, WaiBuffer, WAI_OK};
+    use wai::ffi_sink::*;
+
+    /// A buffer's bytes, the buffer freed. An empty buffer may carry a null pointer, which
+    /// is never read or freed.
+    fn take(b: WaiBuffer) -> Vec<u8> {
+        if b.data.is_null() {
+            return Vec::new();
+        }
+        let v = if b.len == 0 { Vec::new() } else { unsafe { std::slice::from_raw_parts(b.data, b.len) }.to_vec() };
+        unsafe { wai_buffer_free(b) };
+        v
+    }
+
+    fn empty() -> WaiBuffer {
+        WaiBuffer { data: std::ptr::null_mut(), len: 0, cap: 0 }
+    }
+
+    /// The reason of the last refusal: the code before the first colon.
+    fn refused() -> Value {
+        let s = unsafe { CStr::from_ptr(wai_last_error()) }.to_string_lossy().into_owned();
+        json!({"error": s.split(':').next().unwrap_or_default()})
+    }
+
+    fn c(s: &str) -> Result<CString, String> {
+        CString::new(s).map_err(|e| e.to_string())
+    }
+
+    fn parsed(b: WaiBuffer) -> Result<Value, String> {
+        serde_json::from_slice(&take(b)).map_err(|e| e.to_string())
+    }
+
+    pub fn answer_case(case: &Value) -> Result<Vec<Value>, String> {
+        let obj = |n: &Value| -> Result<Option<Vec<u8>>, String> {
+            match n.as_str() {
+                None => Ok(None),
+                Some(n) => case["objects"][n].as_str().and_then(wai::stream_audit::unhex).map(Some).ok_or_else(|| format!("no object {n}")),
+            }
+        };
+        let cfg = c(&case["config"].to_string())?;
+        let seed = case["seed"].as_str().and_then(wai::stream_audit::unhex).filter(|s| s.len() == 32).ok_or("seed: 32 bytes, hexadecimal")?;
+        let start = case["start_us"].as_u64().ok_or("start_us: an integer")?;
+        let mut s: *mut WaiSink = std::ptr::null_mut();
+        if unsafe { wai_sink_open(cfg.as_ptr(), seed.as_ptr(), 32, start, &mut s) } != WAI_OK {
+            return Ok(vec![refused()]);
+        }
+        // The session is freed however the case ends.
+        struct Open(*mut WaiSink);
+        impl Drop for Open {
+            fn drop(&mut self) {
+                unsafe { wai_sink_free(self.0) };
+            }
+        }
+        let open = Open(s);
+        let s = open.0;
+        let mut out = Vec::new();
+        for call in case["calls"].as_array().ok_or("no calls")? {
+            let d = &call["do"];
+            let text = |k: &str| d[k].as_str().ok_or_else(|| format!("{k}: a string")).and_then(c);
+            let mut buf = empty();
+            let got = match d["call"].as_str().unwrap_or("") {
+                "playlist" => {
+                    let (url, b, r) = (text("url")?, obj(&d["body"])?.unwrap_or_default(), obj(&d["receipt"])?);
+                    let rp = r.as_ref().map_or(std::ptr::null(), |r| r.as_ptr());
+                    let st = unsafe { wai_sink_playlist(s, url.as_ptr(), b.as_ptr(), b.len(), rp, r.as_ref().map_or(0, Vec::len), &mut buf) };
+                    if st == WAI_OK { parsed(buf)? } else { refused() }
+                }
+                "parent" => {
+                    let (url, name, b) = (text("url")?, text("name")?, obj(&d["body"])?);
+                    let bp = b.as_ref().map_or(std::ptr::null(), |b| b.as_ptr());
+                    let st = unsafe { wai_sink_playlist_parent(s, url.as_ptr(), name.as_ptr(), bp, b.as_ref().map_or(0, Vec::len)) };
+                    if st == WAI_OK { json!({"ok": true}) } else { refused() }
+                }
+                "object" => {
+                    let (meta, b) = (c(&d["meta"].to_string())?, obj(&d["body"])?.unwrap_or_default());
+                    let st = unsafe { wai_sink_object(s, meta.as_ptr(), b.as_ptr(), b.len(), &mut buf) };
+                    if st == WAI_OK { parsed(buf)? } else { refused() }
+                }
+                "event" => {
+                    let n = text("name")?;
+                    let st = unsafe { wai_sink_event(s, n.as_ptr(), d["t_us"].as_u64().unwrap_or(0)) };
+                    if st == WAI_OK { json!({"ok": true}) } else { refused() }
+                }
+                "observe" => {
+                    let o = c(&d["obs"].to_string())?;
+                    let st = unsafe { wai_sink_observe(s, o.as_ptr(), &mut buf) };
+                    if st == WAI_OK { parsed(buf)? } else { refused() }
+                }
+                "energy" => {
+                    let e = c(&d["entry"].to_string())?;
+                    let st = unsafe { wai_sink_energy(s, e.as_ptr()) };
+                    if st == WAI_OK { json!({"ok": true}) } else { refused() }
+                }
+                "attach" => {
+                    let (n, b) = (text("name")?, obj(&d["body"])?.unwrap_or_default());
+                    let st = unsafe { wai_sink_attach(s, n.as_ptr(), b.as_ptr(), b.len()) };
+                    if st == WAI_OK { json!({"ok": true}) } else { refused() }
+                }
+                "cmcd" => {
+                    let st = unsafe { wai_sink_take_cmcd(s, &mut buf) };
+                    if st == WAI_OK { json!({"cmcd": String::from_utf8(take(buf)).map_err(|e| e.to_string())?}) } else { refused() }
+                }
+                "end" => {
+                    let n = text("ended")?;
+                    let st = unsafe { wai_sink_end(s, n.as_ptr(), d["t_us"].as_u64().unwrap_or(0), &mut buf) };
+                    if st == WAI_OK { json!({"bundle": String::from_utf8(take(buf)).map_err(|e| e.to_string())?}) } else { refused() }
+                }
+                other => return Err(format!("an unknown call {other}")),
+            };
+            out.push(got);
+        }
+        drop(open);
+        Ok(out)
+    }
+}
diff --git a/wai/wai-rs/src/catalogue.rs b/wai/wai-rs/src/catalogue.rs
index 2b82c9396..b692b68ed 100644
--- a/wai/wai-rs/src/catalogue.rs
+++ b/wai/wai-rs/src/catalogue.rs
@@ -408,9 +408,14 @@ pub const OPS: &[Op] = &[
         id: "receipt.verify",
         summary: "Every receipt index (*.wai.receipts) under each path, its group receipts and claims checked offline, and every conformance result, fleet report bundle and envelope there",
         cli: Some("verify"),
-        usage: "verify <path>...   the receipts (receipt indexes *.wai.receipts), conformance results (*.wai-conform.json),
-    fleet report bundles (*.wai-fleet.json) and envelopes' pins (*.wai) under each path, checked offline",
-        input: &[pos("paths", Ty::Strings, true, "Files or directories holding receipt indexes, conformance results, fleet report bundles or envelopes.")],
+        usage: "verify <path>... [--conformant [--root <wai-dir>]]   the receipts (receipt indexes *.wai.receipts), conformance
+    results (*.wai-conform.json), fleet report bundles (*.wai-fleet.json) and envelopes' pins (*.wai) under each path,
+    checked offline; --conformant admits a result only for a run that passed against this checkout's corpus",
+        input: &[
+            pos("paths", Ty::Strings, true, "Files or directories holding receipt indexes, conformance results, fleet report bundles or envelopes."),
+            switch("conformant", "--conformant", "Admit each conformance result only when its run passed, against the corpus of a checkout."),
+            flag("root", "--root", Ty::Path, false, "With --conformant, the wai/ directory of the checkout whose corpus a result names."),
+        ],
         needs: Needs::All(&["quickstart"]),
         mcp: true,
         ..OP
@@ -941,9 +946,9 @@ http vectors diff|fuzz <dir> --seed S --count N --out F",
     the kinds and suites; a record that exists is replaced only with --yes (--confirm, deprecated, is the same)",
         effect: Effect::WritesProject,
         input: &[
-            pos("kind", Ty::String, true, "The kind of implementation (wai conformance list): a decoder, a verifier, an origin, a packager."),
+            pos("kind", Ty::String, true, "The kind of implementation (wai conformance list): a decoder, a verifier, an origin, a packager, a sink."),
             flag("suite", "--suite", Ty::String, false, "The suite (wai conformance list)."),
-            flag("impl", "--impl", Ty::String, true, "The implementation: a command with {in}/{out}/{corpus} placeholders, or a URL."),
+            flag("impl", "--impl", Ty::String, true, "The implementation: a command with {case}/{in}/{out}/{corpus} placeholders, or a URL."),
             flag("key", "--key", Ty::Path, false, "The seed the record is signed with."),
             flag("out", "--out", Ty::Path, false, "The record (<suite>.wai-conform.json)."),
             flag("root", "--root", Ty::Path, false, "The wai/ directory whose corpora are run."),
diff --git a/wai/wai-rs/src/native_sink/corpus.rs b/wai/wai-rs/src/native_sink/corpus.rs
index bb68a64cd..9bed547bf 100644
--- a/wai/wai-rs/src/native_sink/corpus.rs
+++ b/wai/wai-rs/src/native_sink/corpus.rs
@@ -618,6 +618,30 @@ fn run(case: &Value, objects: &Objects) -> (Vec<Value>, Option<Value>) {
     (answers, outcome)
 }
 
+/// A case as the conformance runner gives it to a sink, its answers withheld (`wai conform sink
+/// --suite native-sink`): `objects`, the bodies its calls name, hexadecimal; `config`, `seed`
+/// and `start_us`; and `calls`, each call's `do`. A configuration the corpus states a session
+/// refuses is a case of no call. What this module answers: a session opened from `config` under
+/// the key whose seed is `seed`, and each call's answer in order; or, where the configuration is
+/// refused, that refusal alone, `{"error": <reason>}`.
+pub fn answer_case(case: &Value) -> Result<Vec<Value>, String> {
+    let objects: Objects = case["objects"]
+        .as_object()
+        .ok_or("no objects")?
+        .iter()
+        .map(|(k, v)| v.as_str().and_then(audit::unhex).map(|b| (k.clone(), b)).ok_or_else(|| format!("object {k} is not hexadecimal")))
+        .collect::<Result<_, String>>()?;
+    let cfg = match Config::from_json(&case["config"]) {
+        Ok(c) => c,
+        Err(r) => return Ok(vec![json!({"error": r.code()})]),
+    };
+    let seed = case["seed"].as_str().and_then(unhex32).ok_or("seed: 32 bytes, hexadecimal")?;
+    let start = case["start_us"].as_u64().ok_or("start_us: an integer")?;
+    let mut sink = Sink::open(cfg, SigningKey::from_bytes(&seed), start);
+    let calls = case["calls"].as_array().ok_or("no calls")?;
+    Ok(calls.iter().map(|c| run_call(&mut sink, &objects, &c["do"])).collect())
+}
+
 /// The corpus document.
 pub fn generate() -> Value {
     let mut objects: Objects = BTreeMap::new();
diff --git a/wai/wai-rs/tests/cli_conform_sinks.rs b/wai/wai-rs/tests/cli_conform_sinks.rs
new file mode 100644
index 000000000..bbcfd3329
--- /dev/null
+++ b/wai/wai-rs/tests/cli_conform_sinks.rs
@@ -0,0 +1,163 @@
+//! `conform-result/1` run against WAI's native sink (conformance-results §4.1), through its Rust
+//! module and through the C interface its platform integrations link, and against the score
+//! decoder, each run through `wai conform` as another implementation would be. Each run
+//! is conformant, `wai verify` accepts each record, and `wai verify --conformant` admits each
+//! against the corpus of this checkout. A planted sink that presents an object whose digest does
+//! not match is refused (`not_conformant`); the same sink run against a copy of the corpus changed
+//! to suit it is conformant there, and refused against this checkout's (`corpus_mismatch`); a
+//! record of a suite this build does not run is refused (`suite_unknown`).
+#![cfg(all(feature = "quickstart", feature = "score", feature = "native_sink"))]
+
+use serde_json::Value;
+use std::path::{Path, PathBuf};
+use std::process::{Command, Output};
+
+fn tmp(name: &str) -> PathBuf {
+    let d = std::env::temp_dir().join(format!("wai-conform-sinks-{name}-{}", std::process::id()));
+    let _ = std::fs::remove_dir_all(&d);
+    std::fs::create_dir_all(&d).unwrap();
+    d
+}
+
+fn wai_dir() -> PathBuf {
+    Path::new(env!("CARGO_MANIFEST_DIR")).parent().unwrap().to_path_buf()
+}
+
+fn wai(dir: &Path, args: &[&str]) -> Output {
+    Command::new(env!("CARGO_BIN_EXE_wai")).current_dir(dir).args(args).output().expect("wai runs")
+}
+
+fn text(o: &Output) -> String {
+    format!("{}{}", String::from_utf8_lossy(&o.stdout), String::from_utf8_lossy(&o.stderr))
+}
+
+fn q(p: &str) -> String {
+    format!("'{}'", p.replace('\'', "'\\''"))
+}
+
+/// `wai conform <kind> --suite <suite> --impl <imp> --root <root> --json`, its record written
+/// to `out`: the exit status and the JSON envelope.
+fn conform(dir: &Path, root: &Path, kind: &str, suite: &str, imp: &str, out: &Path) -> (i32, Value) {
+    let o = wai(dir, &["conform", kind, "--suite", suite, "--impl", imp, "--root", root.to_str().unwrap(), "--out", out.to_str().unwrap(), "--yes", "--json"]);
+    let v: Value = serde_json::from_slice(&o.stdout).unwrap_or_else(|_| panic!("JSON: {}", text(&o)));
+    (o.status.code().unwrap_or(-1), v)
+}
+
+/// The reference native sink, through its module or its C interface.
+fn native(c_abi: bool) -> String {
+    format!("{} answers {}{{case}}", q(env!("CARGO_BIN_EXE_wai_native_sink_vectors")), if c_abi { "--c-abi " } else { "" })
+}
+
+/// A sink that presents an object whose digest does not match the SHA-256 its name states: the
+/// reference sink's answers with that refusal's `present` turned true.
+fn planted() -> String {
+    format!("{} | sed '/object_digest_mismatch/s/\"present\":false/\"present\":true/'", native(false))
+}
+
+#[test]
+fn the_native_sink_and_the_score_decoder_conform_and_each_record_is_admitted() {
+    let d = tmp("own");
+    let root = wai_dir();
+    let w = q(env!("CARGO_BIN_EXE_wai"));
+    let runs = [
+        ("sink", "native-sink", native(false), "native-sink"),
+        ("sink", "native-sink", native(true), "native-sink-c-abi"),
+        ("decoder", "score", format!("{w} score render {{in}} --pcm {{out}}"), "score"),
+    ];
+    for (kind, suite, imp, name) in &runs {
+        let (code, v) = conform(&d, &root, kind, suite, imp, &d.join(format!("{name}.wai-conform.json")));
+        assert_eq!(code, 0, "{name}: {v:#}");
+        assert_eq!(v["data"]["conformant"], true, "{name}");
+        assert_eq!(v["data"]["failed"], 0, "{name}");
+    }
+    let rec: Value = serde_json::from_slice(&std::fs::read(d.join("native-sink.wai-conform.json")).unwrap()).unwrap();
+    assert_eq!((rec["kind"].as_str(), rec["suite"].as_str()), (Some("sink"), Some("native-sink")));
+    // Every session of the corpus and every configuration it refuses is a case.
+    let corpus: Value = serde_json::from_slice(&std::fs::read(root.join("live-conformance/native-sink/vectors.json")).unwrap()).unwrap();
+    let n = corpus["cases"].as_array().unwrap().len() + corpus["configs"].as_array().unwrap().len();
+    assert_eq!(rec["summary"]["cases"].as_u64(), Some(n as u64), "{}", rec["summary"]);
+
+    let o = wai(&d, &["verify", d.to_str().unwrap()]);
+    assert!(o.status.success() && text(&o).contains("3 conformance results verified"), "{}", text(&o));
+    let o = wai(&d, &["verify", "--conformant", "--root", root.to_str().unwrap(), d.to_str().unwrap()]);
+    assert!(o.status.success(), "{}", text(&o));
+    assert_eq!(text(&o).matches("ok   admitted").count(), 3, "{}", text(&o));
+    assert!(text(&o).contains("every conformance result admits its implementation"), "{}", text(&o));
+    let _ = std::fs::remove_dir_all(&d);
+}
+
+#[test]
+fn a_planted_failing_sink_is_refused() {
+    let d = tmp("planted");
+    let root = wai_dir();
+    let out = d.join("planted.wai-conform.json");
+    let (code, v) = conform(&d, &root, "sink", "native-sink", &planted(), &out);
+    assert_eq!(code, 1, "{v:#}");
+    assert_eq!(v["data"]["conformant"], false);
+    let failed: Vec<&str> = v["data"]["failures"].as_array().unwrap().iter().map(|f| f["name"].as_str().unwrap()).collect();
+    assert!(failed.contains(&"tampered-fragment"), "{failed:?}");
+    assert!(!failed.contains(&"signed-vod"), "a session with no refused object still passes: {failed:?}");
+    // The record is a genuine record of a run that did not pass: it verifies, and admits nothing.
+    let o = wai(&d, &["verify", out.to_str().unwrap()]);
+    assert!(o.status.success() && text(&o).contains("not conformant"), "{}", text(&o));
+    let o = wai(&d, &["verify", "--conformant", "--root", root.to_str().unwrap(), out.to_str().unwrap()]);
+    assert_eq!(o.status.code(), Some(1), "{}", text(&o));
+    assert!(text(&o).contains("FAIL admission rule 2 not_conformant"), "{}", text(&o));
+    let _ = std::fs::remove_dir_all(&d);
+}
+
+#[test]
+fn a_sink_run_against_a_corpus_changed_to_suit_it_is_refused() {
+    let d = tmp("changed");
+    let root = wai_dir();
+    // A copy of what a run reads, its answers changed to the planted sink's.
+    let copy = d.join("wai");
+    std::fs::create_dir_all(copy.join("registry")).unwrap();
+    std::fs::create_dir_all(copy.join("live-conformance/native-sink")).unwrap();
+    for f in ["SPEC.md", "CHANGELOG.md", "registry/capabilities.json"] {
+        std::fs::copy(root.join(f), copy.join(f)).unwrap();
+    }
+    let mut corpus: Value = serde_json::from_slice(&std::fs::read(root.join("live-conformance/native-sink/vectors.json")).unwrap()).unwrap();
+    for c in corpus["cases"].as_array_mut().unwrap() {
+        for call in c["calls"].as_array_mut().unwrap() {
+            if call["answer"]["refused"] == "object_digest_mismatch" {
+                call["answer"]["present"] = Value::Bool(true);
+            }
+        }
+    }
+    std::fs::write(copy.join("live-conformance/native-sink/vectors.json"), serde_json::to_vec_pretty(&corpus).unwrap()).unwrap();
+    let out = d.join("suited.wai-conform.json");
+    let (code, v) = conform(&d, &copy, "sink", "native-sink", &planted(), &out);
+    assert_eq!(code, 0, "against the changed corpus the planted sink passes: {v:#}");
+    let o = wai(&d, &["verify", "--conformant", "--root", copy.to_str().unwrap(), out.to_str().unwrap()]);
+    assert!(o.status.success(), "a relying party holding the changed corpus admits it: {}", text(&o));
+    let o = wai(&d, &["verify", "--conformant", "--root", root.to_str().unwrap(), out.to_str().unwrap()]);
+    assert_eq!(o.status.code(), Some(1), "{}", text(&o));
+    assert!(text(&o).contains("FAIL admission rule 4 corpus_mismatch"), "{}", text(&o));
+    let _ = std::fs::remove_dir_all(&d);
+}
+
+#[test]
+fn a_record_of_a_suite_this_build_does_not_run_is_refused() {
+    let d = tmp("unknown");
+    let root = wai_dir();
+    let out = d.join("native-sink.wai-conform.json");
+    let (code, _) = conform(&d, &root, "sink", "native-sink", &native(false), &out);
+    assert_eq!(code, 0);
+    let rec: Value = serde_json::from_slice(&std::fs::read(&out).unwrap()).unwrap();
+    let mut r = rec.as_object().unwrap().clone();
+    r.remove("sig");
+    r.remove("signer_pubkey");
+    r.insert("suite".into(), Value::from("nonesuch"));
+    let key = ed25519_dalek::SigningKey::from_bytes(&[0x45; 32]);
+    let r = wai::conform_result::sign(r, &key).unwrap();
+    let other = d.join("nonesuch.wai-conform.json");
+    std::fs::write(&other, serde_json::to_vec_pretty(&Value::Object(r)).unwrap()).unwrap();
+    assert!(wai(&d, &["verify", other.to_str().unwrap()]).status.success(), "signed again, it verifies");
+    let o = wai(&d, &["verify", "--conformant", "--root", root.to_str().unwrap(), other.to_str().unwrap()]);
+    assert_eq!(o.status.code(), Some(1), "{}", text(&o));
+    assert!(text(&o).contains("FAIL admission rule 3 suite_unknown"), "{}", text(&o));
+    // --root alone is a usage error: it names the corpus --conformant compares with.
+    assert_eq!(wai(&d, &["verify", "--root", root.to_str().unwrap(), out.to_str().unwrap()]).status.code(), Some(2));
+    let _ = std::fs::remove_dir_all(&d);
+}
diff --git a/wai/wai-studio/src/actions.generated.js b/wai/wai-studio/src/actions.generated.js
index 6df3bac33..1631dc11f 100644
--- a/wai/wai-studio/src/actions.generated.js
+++ b/wai/wai-studio/src/actions.generated.js
@@ -1,12 +1,12 @@
 // GENERATED by scripts/actions.mjs from the WAI operation catalogue. Do not edit: change the
 // catalogue (wai-rs/src/catalogue.rs, exported by `wai ops export`), then run
 // `node scripts/actions.mjs`. `node scripts/actions.mjs --check` fails when this file differs.
-export const CATALOGUE = {"format":"wai.ops/1","toolchain":"0.4.0","sha256":"7b58de8b18b78ac1baaff4a1f2e74e387da3a5d243080b0bf0e0bd1237756312"};
+export const CATALOGUE = {"format":"wai.ops/1","toolchain":"0.4.0","sha256":"bbb0d3f85cfb9d2c1e6a0a2a69fff6f9b8fc6762a1a9560834d8136a0b7c47bd"};
 export const ACTIONS = [
   {"id":"toolchain.test","noun":"toolchain","verb":"test","menu":"Toolchain","item":"Test","palette":"toolchain test","summary":"WAI's four steps (name, dispatch, rebuild, receipt) on a picture made here, its receipt checked offline","placement":"local","effect":"writes-project","stability":"stable","toolset":"core","cli":"wai hello","mcp":null,"route":null,"call":"toolchain_test"},
   {"id":"project.create","noun":"project","verb":"create","menu":"Project","item":"Create","palette":"project create","summary":"A new project from one of the examples, written into a new or empty directory; --list names the templates","placement":"local","effect":"writes-project","stability":"stable","toolset":"core","cli":"wai create","mcp":null,"route":null,"call":"project_create"},
   {"id":"dev.start","noun":"dev","verb":"start","menu":"Dev","item":"Start","palette":"dev start","summary":"A relay, a room service and the room page on one origin on this machine, for development; --check starts, checks every piece and stops","placement":"local","effect":"runs","stability":"stable","toolset":"dev","cli":"wai dev","mcp":"dev_start","route":null,"call":"dev_start"},
-  {"id":"receipt.verify","noun":"receipt","verb":"verify","menu":"Receipt","item":"Verify","palette":"receipt verify","summary":"Every receipt index (*.wai.receipts) under each path, its group receipts and claims checked offline, and every conformance result and fleet report bundle there","placement":"local","effect":"reads","stability":"stable","toolset":"core","cli":"wai verify","mcp":"receipt_verify","route":null,"call":"receipt_verify"},
+  {"id":"receipt.verify","noun":"receipt","verb":"verify","menu":"Receipt","item":"Verify","palette":"receipt verify","summary":"Every receipt index (*.wai.receipts) under each path, its group receipts and claims checked offline, and every conformance result, fleet report bundle and envelope there","placement":"local","effect":"reads","stability":"stable","toolset":"core","cli":"wai verify","mcp":"receipt_verify","route":null,"call":"receipt_verify"},
   {"id":"envelope.inspect","noun":"envelope","verb":"inspect","menu":"Envelope","item":"Inspect","palette":"envelope inspect","summary":"What an envelope holds and whether this build decodes it: media, intent, pin, capability, fallback, the dispatch verdict and the conformance checks","placement":"local","effect":"reads","stability":"stable","toolset":"core","cli":"wai inspect","mcp":"envelope_inspect","route":null,"call":"envelope_inspect"},
   {"id":"envelope.wrap","noun":"envelope","verb":"wrap","menu":"Envelope","item":"Wrap","palette":"envelope wrap","summary":"Carry a registered capability's bytes in an envelope, byte for byte, under SPEC's intent, role and candidate rules","placement":"local","effect":"writes-project","stability":"stable","toolset":"core","cli":"wai wrap","mcp":"envelope_wrap","route":null,"call":"envelope_wrap"},
   {"id":"envelope.unwrap","noun":"envelope","verb":"unwrap","menu":"Envelope","item":"Unwrap","palette":"envelope unwrap","summary":"An envelope's capability and its checks, and with -o its payload or one component, exactly as wrapped","placement":"local","effect":"writes-project","stability":"stable","toolset":"core","cli":"wai unwrap","mcp":null,"route":null,"call":"envelope_unwrap"},
-- 
2.54.0 (Apple Git-157)


From b37b85aa00587ffb2144ac7adf1a970f88f3284d Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 09:58:59 -0500
Subject: [PATCH 23/29] wai: an honest session's base entry names the pin it
 verified, so its stage hash seals
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

every_corpus_case_seals_an_honest_session replays each staged case as a sink
and records its session, but its decoder entries named no pin and no tier
for any base. Since determinism-tiers §3 rule 3, a stage hash over a
NeuralIntegerExact base is refused unless every entry naming the base is at
decode-equivalence, so the honest sessions of the pinned int_hyper,
int_codec and splat cases were refused with session_stage_hash_not_exact.

A base's entry now names what the sink's decode states (jwp-receipts
§2.4.12): the pin it verified, with the tier that pin declares; for an
unpinned decode that read a set, the digest a pin over the bytes read
names, with the tier none (SPEC §3.1, Pin by default, as
IntParams::read_pin names it); nothing for a decode that read no set.
Layer entries name nothing, as before.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/wai-rs/src/staged/claim_tests.rs | 51 ++++++++++++++++++++++++----
 1 file changed, 45 insertions(+), 6 deletions(-)

diff --git a/wai/wai-rs/src/staged/claim_tests.rs b/wai/wai-rs/src/staged/claim_tests.rs
index 88ec6a810..421e05a4c 100644
--- a/wai/wai-rs/src/staged/claim_tests.rs
+++ b/wai/wai-rs/src/staged/claim_tests.rs
@@ -492,10 +492,48 @@ fn replay_session(dir: &std::path::Path) -> Option<Replayed> {
     use crate::session_recorder::{DecoderKey, SessionRecorder};
     let key = ed25519_dalek::SigningKey::from_bytes(&[0x44; 32]);
     let case: Value = serde_json::from_slice(&std::fs::read(dir.join("case.json")).expect("case.json")).expect("JSON");
+    // What the sink's decoder entry for a base names as its parameter set (jwp-receipts
+    // §2.4.12): the pin its decode verified, with the tier that pin declares; for an
+    // unpinned decode that read a set, the digest a pin over the bytes it read names, with
+    // the tier `none` (SPEC §3.1, *Pin by default*, as `int_payload::IntParams::read_pin`
+    // names it); nothing for a decode that read no set. A stage hash over a
+    // `NeuralIntegerExact` base needs the entry at `decode-equivalence` (determinism-tiers
+    // §3, rule 3), so a sink that names no pin seals no session over its stage hash.
+    let wai_dir = PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("..");
+    let named_set = |b: &Value, env: &[u8]| -> (Option<[u8; 32]>, Option<String>) {
+        use crate::codecs::DeclaredTier;
+        if b["pin_resolved"] == true {
+            let w = Wai::from_bytes(env).expect("a base envelope");
+            let p = w.prior().expect("a pin that resolved is the envelope's");
+            let mut pin = [0u8; 32];
+            for (i, x) in pin.iter_mut().enumerate() {
+                *x = u8::from_str_radix(&p.sha256[2 * i..2 * i + 2], 16).expect("a SHA-256 in hexadecimal");
+            }
+            let tier = p.tier().and_then(DeclaredTier::from_token).unwrap_or(DeclaredTier::None);
+            return (Some(pin), Some(tier.token().to_owned()));
+        }
+        let read: BTreeMap<String, Vec<u8>> = b["decode"]["set"]
+            .as_object()
+            .map(|s| s.iter().map(|(m, p)| (m.clone(), std::fs::read(wai_dir.join(p.as_str().expect("a path"))).expect("a set file"))).collect())
+            .unwrap_or_default();
+        if read.is_empty() {
+            return (None, None);
+        }
+        let cap = b["decoded_via"].as_str().expect("decoded_via");
+        let digest = if crate::container::pin_shape(cap) == crate::codecs::ParameterSetShape::OneFile {
+            crate::pin::sha256(read.values().next().expect("one file"))
+        } else {
+            let files: BTreeMap<String, &[u8]> = read.iter().map(|(m, x)| (m.clone(), x.as_slice())).collect();
+            crate::pin::sha256(&crate::pin::listing_from(&files).expect("a listing of the set"))
+        };
+        (Some(digest), Some(DeclaredTier::None.token().to_owned()))
+    };
     let mut bases = Vec::new();
+    let mut base_sets = Vec::new();
     for b in case["bases"].as_array().expect("bases") {
         let env = std::fs::read(dir.join(b["file"].as_str().expect("file"))).expect("base");
         let stage0 = std::fs::read(dir.join(b["stage0"].as_str().expect("stage0"))).expect("stage0");
+        base_sets.push(named_set(b, &env));
         let info = BaseInfo {
             content_hash: chash(&env),
             capability: b["decoded_via"].as_str().expect("decoded_via").into(),
@@ -619,14 +657,15 @@ fn replay_session(dir: &std::path::Path) -> Option<Replayed> {
     let mut rec = SessionRecorder::new([7; 32], "video", host, None);
     let mut handles = BTreeMap::new();
     for (&(unit, layers), source) in &presented {
-        // The capability, and the title's unit that is unit 0 of the object's base.
-        let (capability, first_unit) = match bases.iter().find(|(info, _, _)| info.content_hash == *source) {
-            Some((info, _, _)) => (info.capability.clone(), info.first_unit),
+        // The capability, the title's unit that is unit 0 of the object's base, and, for a
+        // base, the parameter set its entry names.
+        let (capability, first_unit, (pin, tier)) = match bases.iter().position(|(info, _, _)| info.content_hash == *source) {
+            Some(i) => (bases[i].0.capability.clone(), bases[i].0.first_unit, base_sets[i].clone()),
             None => {
                 let w = files.values().find(|f| chash(f) == *source).and_then(|f| Wai::from_bytes(f).ok()).expect("a layer");
                 let base = w.refines().and_then(Result::ok).map(|r| r.base).expect("a layer refines a base");
                 let first = bases.iter().find(|(info, _, _)| hex(&info.content_hash) == base).map_or(0, |(info, _, _)| info.first_unit);
-                (w.capability().to_owned(), first)
+                (w.capability().to_owned(), first, (None, None))
             }
         };
         let h = *handles.entry((capability.clone(), *source)).or_insert_with(|| {
@@ -638,8 +677,8 @@ fn replay_session(dir: &std::path::Path) -> Option<Replayed> {
                 path: SessionPath::Primary,
                 decoder: "builtin:wai@0.4.0".into(),
                 variant: None,
-                pin: None,
-                tier: None,
+                pin,
+                tier,
                 module_status: None,
             })
         });
-- 
2.54.0 (Apple Git-157)


From 3dabd97e6c6c76a8521806dabdb4e82e74fa0f35 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 09:58:59 -0500
Subject: [PATCH 24/29] CHANGELOG: the pin-by-default entry acknowledges the
 staged-measure goldens' break

wai breaking reports staged-measure-conformance/goldens.json as a wire break:
the unpinned fixture's evaluation was added, and the corpus-wide dedup
record, taken over every fixture's objects, counts its objects too, so its
totals moved. The entry named no break for it. It now acknowledges it, and
says that no other fixture's evaluation changes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index e2ffa2385..d66559572 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -53,6 +53,9 @@ wire format may change and such changes are recorded here.
 
 - **Breaking:** `api:rust:wai::staged::BaseDecode::set`: a new member of a public struct, the set the decode read; a
   struct literal or a pattern that lists every member adds it.
+- **Breaking:** `wire:corpus:wai/staged-measure-conformance/goldens.json`: it gains the `unpinned` fixture's
+  evaluation, and its `dedup` record, taken over the objects of every fixture, now counts that fixture's too, so its
+  totals move; no other fixture's evaluation changes. A verifier reads the goldens of the corpus it checks.
 
 ### A decode reports the path that ran, and no tier lies between the two
 
-- 
2.54.0 (Apple Git-157)


From 45eae2dd1643e498777b4bcf3475ae0ac51ba9f3 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 11:15:33 -0500
Subject: [PATCH 25/29] =?UTF-8?q?determinism-tiers=20=C2=A71=20and=20?=
 =?UTF-8?q?=C2=A77.1:=20each=20published=20route=20described=20on=20its=20?=
 =?UTF-8?q?own?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Two passages followed a published route's limitation with a sentence about
WAI's integer networks: in §1, an integerised video codec's accumulators,
whose overflow was observed not to happen rather than bounded; in §7.1, a
requantisation that was not bit-exact across neural processors. Each
passage now describes the route alone. integer-operators §8 and §5.1
specify WAI's bounds and rounding.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/extensions/determinism-tiers.md | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/wai/extensions/determinism-tiers.md b/wai/extensions/determinism-tiers.md
index eb02575e5..9c00b6bf0 100644
--- a/wai/extensions/determinism-tiers.md
+++ b/wai/extensions/determinism-tiers.md
@@ -74,8 +74,7 @@ integer arithmetic:
   (arXiv 2502.20762) runs its float model in a 16-bit integer mode without
   retraining: each value is mapped to round(512 · v), and its accumulators are
   32-bit, of which it reports that "no overflow issues have been observed"
-  (§4.5). That is an observation, not a bound. WAI's integer networks state their
-  bounds before they decode ([integer-operators](integer-operators.md) §8).
+  (§4.5). That is an observation, not a bound.
 - **Integer training.** A published learned video codec is trained as an
   integer-only model from the start, and claims coding consistency across
   platforms from integer decoding in the encoder's reconstruction loop
@@ -1035,9 +1034,7 @@ this record. arXiv 2608.00483 found its streams to decode only within the groups
 its tests established (Table IV). arXiv 2606.28027 found its float reconstruction to diverge
 across devices unless it is resynchronised (§3.2), and its 8-bit integer decode not to be
 bit-exact across neural processors, for "non-standard implementations of rounding and
-multiplication in quantize/dequantize and re-quantization" (Supplement D.1). The
-requantisation of WAI's integer networks states its rounding
-([integer-operators](integer-operators.md) §5.1).
+multiplication in quantize/dequantize and re-quantization" (Supplement D.1).
 
 What the record does not cover is listed in its §1: `wai.image.jpegai`'s exact decoder
 (one configuration, so one rate point and no curve; and the reference set was not coded
-- 
2.54.0 (Apple Git-157)


From 081b71e9b41a05ee1f3d0501cf04006bd2d8c7d6 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 11:26:58 -0500
Subject: [PATCH 26/29] claim-conformance: an unpinned decode's entry names the
 set it read, and one vector keeps the earlier text's null
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Two decode-path vectors stated an unpinned NeuralIntegerExact decode with
pin and tier null, the form of the earlier text. Under SPEC §3.1, Pin by
default, rule 4 and jwp-receipts §2.4.12, a sink names the set it read, by
the digest a pin over it names, with the tier none, and null means a decode
that read no set or an entry from a sink of the earlier text.
session_variant_not_exact_unpinned and session_stage_hash_over_an_unpinned_base
now write the entry so, and each is still refused for its own reason: rule
13, and a stage hash over a base whose tier is not decode-equivalence. A
vector appended after them, session_stage_hash_over_an_unpinned_base_earlier_text,
keeps the null form, which a verifier cannot tell from a decode that read
no set, and is refused all the same. The vectors before the decode-path
rules are unchanged, as the test that pins them by digest requires.

differential.py reaches the tier none too: in its decode-path sweep, in the
mutation that puts a path under a pin, and where its stage-bounds sweep
unpins the base's entry, which now writes either form.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 wai/CHANGELOG.md                              |  5 +++-
 wai/claim-conformance/README.md               |  8 +++++--
 wai/claim-conformance/differential.py         | 19 +++++++++------
 wai/claim-conformance/vectors.json            | 22 ++++++++++++++---
 wai/wai-rs/src/claim_receipt.rs               |  2 +-
 .../src/claim_receipt/corpus/session.rs       | 24 +++++++++++++++----
 6 files changed, 62 insertions(+), 18 deletions(-)

diff --git a/wai/CHANGELOG.md b/wai/CHANGELOG.md
index d66559572..816fb90de 100644
--- a/wai/CHANGELOG.md
+++ b/wai/CHANGELOG.md
@@ -48,7 +48,10 @@ wire format may change and such changes are recorded here.
   `staged-measure-conformance/fixtures/unpinned`, over the staged corpus's unpinned `wai.neural.int_hyper` base, holds
   both recorders and both verifiers to it; every other committed session and golden pins its bases, so none of them
   changes. A decoder entry sealed by a sink of the earlier text may carry `null` for an unpinned decode that read a
-  set, and a verifier cannot tell it from a decode that read none (jwp-receipts §2.4.12).
+  set, and a verifier cannot tell it from a decode that read none (jwp-receipts §2.4.12). `claim-conformance/`'s
+  vectors of an unpinned `wai.neural.int_hyper` decode name the set it read, and one more, appended, writes the base
+  in the earlier text's form: a stage hash over either is refused (`session_stage_hash_over_an_unpinned_base`,
+  `session_stage_hash_over_an_unpinned_base_earlier_text`).
 - **SPEC §3.1 rule 2** says that an unpinned envelope conforms unless §7 requires the pin (`replicate_unpinned`).
 
 - **Breaking:** `api:rust:wai::staged::BaseDecode::set`: a new member of a public struct, the set the decode read; a
diff --git a/wai/claim-conformance/README.md b/wai/claim-conformance/README.md
index f3ce95f79..df8bd2cf6 100644
--- a/wai/claim-conformance/README.md
+++ b/wai/claim-conformance/README.md
@@ -362,7 +362,8 @@ vector, which a test pins by digest:
 1. Rule 13, `session_variant_not_exact`: a `NeuralIntegerExact` capability named with a
    path that is not exact. `session_variant_not_exact` names `wai.neural.int_hyper`
    with `float` under a pin of `decode-equivalence`, `session_variant_not_exact_unpinned`
-   with an unregistered term and no pin, and
+   with an unregistered term and no pin verified, its entry naming the set it read with the
+   tier `none`, and
    `session_variant_lowering_of_another_capability` with `dred-int/1`, the lowering of
    another capability. Accepted: `session_variant_exact_integer_accept`, the integer path
    stated as `exact`, and `session_variant_lowering_accept`, `wai.audio.opus.dred.int`'s
@@ -391,7 +392,10 @@ vector, which a test pins by digest:
    check being per object (`session_stage_hash_not_exact_outside_the_set`); while one of
    its layers is stated as decoded on a `float` path
    (`session_stage_hash_over_a_float_layer`); or while its base is stated as decoded with
-   no pin (`session_stage_hash_over_an_unpinned_base`). Accepted:
+   no pin verified, its entry naming the set it read with the tier `none`
+   (`session_stage_hash_over_an_unpinned_base`), or with `pin` and `tier` `null`, as a sink
+   of an earlier text of jwp-receipts writes an unpinned decode
+   (`session_stage_hash_over_an_unpinned_base_earlier_text`). Accepted:
    `session_stage_hash_exact_accept`.
 
 ## Steps 1 and 2: types, and the order a body is read in
diff --git a/wai/claim-conformance/differential.py b/wai/claim-conformance/differential.py
index c2b120f90..509475551 100644
--- a/wai/claim-conformance/differential.py
+++ b/wai/claim-conformance/differential.py
@@ -852,12 +852,14 @@ def sessions(ins, rng, n):
                 # (whm1/1 is the model session's integer capabilities' own), or none.
                 rng.choice(ds)["variant"] = rng.choice([None, "exact", "float", "fp16", "whm1/1", "dred-int/1"])
             elif k == 32 and ds:
-                # A float path under a pin of either tier, or none, through a capability the
-                # registry classifies or one it does not.
+                # A float path under a pin of either tier, naming the set read with no pin
+                # verified, or naming no set, through a capability the registry classifies or
+                # one it does not.
                 d = rng.choice(ds)
                 d["variant"] = rng.choice(["float", "exact", "fp16", "dred-int/1"])
                 d["pin"], d["tier"] = rng.choice([(CV.blake3(b"differential: a pin").hex(), "decode-equivalence"),
-                                                  (CV.blake3(b"differential: a pin").hex(), "entropy-consistency"), (None, None)])
+                                                  (CV.blake3(b"differential: a pin").hex(), "entropy-consistency"),
+                                                  (CV.blake3(b"differential: a pin").hex(), "none"), (None, None)])
                 if rng.random() < 0.6:
                     d["capability"] = rng.choice(["org.example.picture", "org.example.picture", "wai.image.jpegai",
                                                   "wai.audio.opus.dred.int"])
@@ -927,12 +929,14 @@ def session_stage_bounds(ins, rng, n):
             rng.choice(decoders)["first_unit"] += 1
         # At decode-equivalence the claim states a stage hash (item 5), which holds while the
         # base's entry decodes it at decode-equivalence and each entry runs an exact path;
-        # now and then the base's entry is unpinned, or an entry names another path.
+        # now and then the base's entry is unpinned (naming the set it read with the tier
+        # none, or with pin and tier null as a sink of an earlier text writes it), or an
+        # entry names another path.
         tier = rng.choice(["none", "none", "decode-equivalence"])
         if tier == "decode-equivalence" and rng.random() < 0.5:
             d = rng.choice(decoders)
             if d is decoders[0] and rng.random() < 0.5:
-                d["pin"], d["tier"] = None, None
+                d["pin"], d["tier"] = rng.choice([(CV.blake3(b"differential: a pin").hex(), "none"), (None, None)])
             else:
                 d["variant"] = rng.choice(["float", "fp16", "exact", "whm1/1"])
         w = session_stage_claim(rng, n_set, rng.choice([at1, at1, max(0, at1 - 1), min(n_set, at1 + 1)]), VIEWER, base_hash,
@@ -1621,12 +1625,13 @@ def session_decode_paths(ins):
     for every capability below (classified at each tier, or not at all), decoded by a
     builtin or a module, on a path named by every term below (exact, float, a lowering a
     registration names, one it names for another capability, one no revision registers, or
-    none), under a pin of either tier or none."""
+    none), under a pin of either tier, naming the set read with no pin verified (the tier
+    `none`), or naming no set."""
     base = ins.receipt("session_variant_float_accept")
     pin = CV.blake3(b"differential: a pin").hex()
     for cap in ["org.example.picture", "wai.image.jpegai", "wai.neural.int_hyper", "wai.audio.opus.dred.int", "wai.neural.dcvc_rt"]:
         for variant in [None, "exact", "float", "fp16", "dred-int/1", "whm1/1", "jpegai-int/1"]:
-            for tier in [None, "entropy-consistency", "decode-equivalence"]:
+            for tier in [None, "none", "entropy-consistency", "decode-equivalence"]:
                 for decoder, status in [("builtin:wai@0.4.0", None), ("module:sha256:" + "5e" * 32, "pinned")]:
                     d = dict(base["body"]["decoders"][0], capability=cap, variant=variant, pin=pin if tier else None, tier=tier,
                              decoder=decoder, module_status=status)
diff --git a/wai/claim-conformance/vectors.json b/wai/claim-conformance/vectors.json
index 7c15c5c4e..5694237c6 100644
--- a/wai/claim-conformance/vectors.json
+++ b/wai/claim-conformance/vectors.json
@@ -24695,9 +24695,9 @@
     },
     {
       "name": "session_variant_not_exact_unpinned",
-      "about": "the same with no pin, and a path named by a term this revision does not register: only null, exact or the capability's registered lowering names the integer path",
+      "about": "the same with no pin verified, its entry naming the set it read with the tier none, and a path named by a term this revision does not register: only null, exact or the capability's registered lowering names the integer path",
       "procedure": "claim",
-      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":\"fp16\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"95f4b49237eed8b2c76bfb0c9a00f9f8ffc5a8e2e6f47454e4a1c021c373a33fb16ddccfb3f69e7b98fe991028fff9e5d37d832b6ac50e35e28e3b28da468d06\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
+      "receipt": "{\"body\":{\"classical_declared\":null,\"classical_source\":null,\"counted_discarded\":0,\"counted_fallback\":0,\"counted_primary\":0,\"deadline_us\":null,\"decoders\":[{\"capability\":\"wai.neural.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"none\",\"variant\":\"fp16\"}],\"energy\":null,\"fallback_units\":0,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":1,\"media\":\"image\",\"missed\":0,\"missed_units\":[],\"output_root\":\"9bbb2c453e90eab8c5e9ca19a9b9f355103fa5d112ba9b50f23f277b3849e7e7\",\"primary_units\":1,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[],\"switches\":[],\"switches_total\":0,\"transferred\":367,\"uncounted\":1,\"units\":1},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0f53b6de5fe4a90a9fe8a74292c13e6f7e01267ac334512ff47f9f3cdf77aa31d25cbac82963626f9cdb92b3a0fa7bd01cd3a4b16f87f8fef643294ac1625400\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"666d5383cd03ed68010ca0f5d2749addc0145bc4b14f864eae26d876b7f247c9\",\"subject_type\":\"object\"}",
       "expected": {
         "outcome": "reject",
         "step": 4,
@@ -24834,7 +24834,23 @@
     },
     {
       "name": "session_stage_hash_over_an_unpinned_base",
-      "about": "the same stage claim, while the base is stated as decoded with no pin: its NeuralIntegerExact capability holds decode-equivalence only under a pin that declares it (staged-delivery 5)",
+      "about": "the same stage claim, while the base is stated as decoded with no pin verified, its entry naming the set it read with the tier none: its NeuralIntegerExact capability holds decode-equivalence only under a pin that declares it (staged-delivery 5)",
+      "procedure": "claim",
+      "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":4,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":null,\"tier\":\"none\",\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":5,\"leaves\":1,\"module_status\":null,\"object\":\"13b04249eca2b1a9061dba8ca2846169d3f5fa3548d9e138a0cf2cb553798cf4\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"b62a8425b7fe3a085a14f2f2156ea22ae3b4349884d799b5afe66dccc7d4dd7f\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":1,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"c98f973df33707192c3a999563f40bf5c907124879ccc3e6291ff3ebb06602e8839f81f1ac22a24b5e885c0e999c1aa64ef78f3bb5585a5cbf5ce1cfb7696b0d\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
+      "context": {
+        "stage_claims": [
+          "{\"body\":{\"bytes_envelopes\":10607,\"conflicts\":0,\"deadline_misses\":2,\"objects\":[{\"bytes\":900,\"evidence\":\"none\",\"layer\":1,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":901,\"evidence\":\"none\",\"layer\":1,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":1}},{\"bytes\":902,\"evidence\":\"none\",\"layer\":1,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":2}},{\"bytes\":903,\"evidence\":\"none\",\"layer\":1,\"object\":\"792a12990c5207ca83a86b7420815f1f2d34918553a0130a4f9570a11354f2ca\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":3}},{\"bytes\":1000,\"evidence\":\"none\",\"layer\":2,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"reason\":null,\"status\":\"applied\",\"units\":{\"count\":1,\"from\":0}},{\"bytes\":1001,\"evidence\":\"none\",\"layer\":2,\"object\":\"1c6eae0ed41ed7fc1b2f00b46ea1de98f21a6d6cbf62bf6b22cf76a664a49376\",\"reason\":null,\"status\":\"late\",\"units\":{\"count\":1,\"from\":1}}],\"op_committed\":null,\"op_delivered\":null,\"op_requested\":null,\"presented\":{\"wai-stage:layers\":0},\"stage_hash\":\"082bc00210e6d32f18f39ee723d863fd7b99f206a415ac2f44a8687793b2e068\",\"tier\":\"decode-equivalence\",\"units_by_stage\":[1,2,1],\"vocabulary\":\"urn:wai:staged-delivery:1\"},\"class\":\"stage\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"0c29c7a2e9cfdb393273506b5acc9672b225192b28b9114f3b814d19c91bf4f994a74f2eecc077abbe6840c84f98f09ebbd2bd4f08d84592cca9a944918f590d\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}"
+        ]
+      },
+      "expected": {
+        "outcome": "reject",
+        "step": 6,
+        "reason": "session_stage_hash_not_exact"
+      }
+    },
+    {
+      "name": "session_stage_hash_over_an_unpinned_base_earlier_text",
+      "about": "the same, with the base's entry as a sink of an earlier text of this extension writes an unpinned decode, pin and tier null, which a verifier cannot tell from a decode that read no set: refused all the same, since no pin declares decode-equivalence",
       "procedure": "claim",
       "receipt": "{\"body\":{\"classical_declared\":700,\"classical_source\":\"wai2-table\",\"counted_discarded\":1904,\"counted_fallback\":0,\"counted_primary\":5935,\"deadline_us\":40000,\"decoders\":[{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":4,\"module_status\":null,\"object\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_hyper\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":4,\"leaves\":1,\"module_status\":null,\"object\":\"f18f33c4eebadf2caf00fb22a38cf394e19d31a62ebe68f3043c31c84ecbcf44\",\"path\":\"primary\",\"pin\":\"7b4371854df0aa1645ac6525b373387c0617b873d5a57748932ee0f386ea164b\",\"rendition\":0,\"tier\":\"decode-equivalence\",\"variant\":null},{\"capability\":\"wai.video.av1\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":5,\"leaves\":1,\"module_status\":null,\"object\":\"13b04249eca2b1a9061dba8ca2846169d3f5fa3548d9e138a0cf2cb553798cf4\",\"path\":\"fallback\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"47a2c23727d2eb9d2beef0931c3e9b2675408d372660420cce427dc27fd21fe1\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"6b74be11e64461c7d972fc5cb6ba15e98427e9d6e60622b0a753222138de28cf\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"5e7a38ca5c04442cbe00b040d58e4494f5aa36fb81e8bbe0d304fd2d1ce722b2\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null},{\"capability\":\"wai.video.int_refine\",\"decoder\":\"builtin:wai@0.4.0\",\"first_unit\":0,\"leaves\":1,\"module_status\":null,\"object\":\"be5055af8e32359265852c512667b10ab6a07455e5ab06603e278763f6633a85\",\"path\":\"primary\",\"pin\":null,\"rendition\":null,\"tier\":null,\"variant\":null}],\"energy\":null,\"fallback_units\":1,\"framing\":\"registered\",\"host\":{\"build\":{\"profile\":\"release\",\"target\":\"aarch64-unknown-linux-gnu\",\"toolchain\":\"rustc 1.95.0\"},\"profile\":\"native-ffi\",\"target\":\"aarch64-unknown-linux-gnu\",\"threads\":4},\"leaf_count\":10,\"media\":\"video\",\"missed\":1,\"missed_units\":[{\"late_us\":1200,\"layers\":1,\"unit\":5}],\"output_root\":\"b62a8425b7fe3a085a14f2f2156ea22ae3b4349884d799b5afe66dccc7d4dd7f\",\"primary_units\":5,\"receipt_bytes\":640,\"session_nonce\":\"57c5462dfdabd1b7bdd204d34ba33649c057eb535349f29f45eb8f8c7d4a7e7f\",\"stage_sets\":[{\"claim\":\"edc1de4f22af2c57dca1a362e302b7fe6309bd2e7840ab05c957002bf527b3f0\",\"first_unit\":0}],\"switches\":[{\"from\":\"primary\",\"reason\":\"set-incomplete\",\"to\":\"fallback\",\"unit\":5}],\"switches_total\":1,\"transferred\":9995,\"uncounted\":2,\"units\":6},\"class\":\"reconstruct-session\",\"issued_at\":\"2026-09-27T12:00:00Z\",\"kind\":\"claim\",\"parent_receipt_hash\":null,\"seq\":0,\"sig\":\"d8f9aa9a0222d8a800e880d7fe6f6295ca39d842f727244b6d6458cbaafe4e3e7acb0d26015d64c6cd843cb2ca27aa0eaa775d172487a68ed32def05fa0edc0a\",\"signer_id\":\"did:example:sink\",\"signer_pubkey\":\"d759793bbc13a2819a827c76adb6fba8a49aee007f49f2d0992d99b825ad2c48\",\"subject\":\"d09526cec3961c21e36cc08f0b393323eb82dcc4203cb08df57fee36784b1113\",\"subject_type\":\"object\"}",
       "context": {
diff --git a/wai/wai-rs/src/claim_receipt.rs b/wai/wai-rs/src/claim_receipt.rs
index 736b534e4..6a5cbdf36 100644
--- a/wai/wai-rs/src/claim_receipt.rs
+++ b/wai/wai-rs/src/claim_receipt.rs
@@ -4261,7 +4261,7 @@ pub(crate) fn rfc3339(s: &str) -> bool {
 /// fails instead of passing on nothing. (A function rather than a public constant so
 /// cbindgen does not export it into the C header.)
 pub fn corpus_vector_count() -> usize {
-    652
+    653
 }
 
 /// The conformance corpus (`claim-conformance/vectors.json`), regenerated. Ed25519
diff --git a/wai/wai-rs/src/claim_receipt/corpus/session.rs b/wai/wai-rs/src/claim_receipt/corpus/session.rs
index 94fc8d9cb..804fa6cad 100644
--- a/wai/wai-rs/src/claim_receipt/corpus/session.rs
+++ b/wai/wai-rs/src/claim_receipt/corpus/session.rs
@@ -2729,7 +2729,9 @@ fn picture() -> [u8; 32] {
 }
 
 /// A session of one picture through one decoder entry: `capability` decoded by `decoder`
-/// on the path `variant` names, under a pin of `tier` or none.
+/// on the path `variant` names. `tier` is the declared tier of the pin the sink verified,
+/// or `none` for a decode that verified no pin, the entry naming the set it read either
+/// way; with no tier the entry names no set.
 fn one_picture(capability: &str, decoder: &str, module_status: Option<&str>, variant: Option<&str>, tier: Option<&str>) -> SessionClaim {
     let ps = [pres(0, 1, 0)];
     SessionClaim {
@@ -2812,8 +2814,8 @@ pub(super) fn decode_path_vectors() -> Vec<V> {
         ),
         reject(
             "session_variant_not_exact_unpinned",
-            "the same with no pin, and a path named by a term this revision does not register: only null, exact or the capability's registered lowering names the integer path",
-            picture_bad(one_picture("wai.neural.int_hyper", BUILTIN, None, Some("fp16"), None)),
+            "the same with no pin verified, its entry naming the set it read with the tier none, and a path named by a term this revision does not register: only null, exact or the capability's registered lowering names the integer path",
+            picture_bad(one_picture("wai.neural.int_hyper", BUILTIN, None, Some("fp16"), Some("none"))),
             4,
             Reason::SessionVariantNotExact,
         ),
@@ -2931,7 +2933,21 @@ pub(super) fn decode_path_vectors() -> Vec<V> {
     v.push(held(
         reject(
             "session_stage_hash_over_an_unpinned_base",
-            "the same stage claim, while the base is stated as decoded with no pin: its NeuralIntegerExact capability holds decode-equivalence only under a pin that declares it (staged-delivery 5)",
+            "the same stage claim, while the base is stated as decoded with no pin verified, its entry naming the set it read with the tier none: its NeuralIntegerExact capability holds decode-equivalence only under a pin that declares it (staged-delivery 5)",
+            sealed_or_bad({
+                let mut s = named(&de_stage);
+                s.decoders[0].tier = Some("none".into());
+                s
+            }),
+            6,
+            Reason::SessionStageHashNotExact,
+        ),
+        &de_stage,
+    ));
+    v.push(held(
+        reject(
+            "session_stage_hash_over_an_unpinned_base_earlier_text",
+            "the same, with the base's entry as a sink of an earlier text of this extension writes an unpinned decode, pin and tier null, which a verifier cannot tell from a decode that read no set: refused all the same, since no pin declares decode-equivalence",
             sealed_or_bad({
                 let mut s = named(&de_stage);
                 s.decoders[0].pin = None;
-- 
2.54.0 (Apple Git-157)


From 2a48fffdb7a528348086ce5c9cfdab8e17f4f6f1 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 14:19:56 -0500
Subject: [PATCH 27/29] wai-web-wasm: the native unit test count includes the
 decoder entry's set test

Pin by default added measure::tests::a_decoder_entry_names_the_set_its_decode_read,
so the native unit tests step's count check must expect it; the step otherwise
fails after every test has passed.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .github/workflows/wai-web-wasm.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/wai-web-wasm.yml b/.github/workflows/wai-web-wasm.yml
index 454539028..1191d8322 100644
--- a/.github/workflows/wai-web-wasm.yml
+++ b/.github/workflows/wai-web-wasm.yml
@@ -290,7 +290,7 @@ jobs:
         run: |
           set -euo pipefail
           cargo test --locked --lib 2>&1 | tee "$RUNNER_TEMP/wai-web-native.log"
-          grep -q "test result: ok. 36 passed; 0 failed" "$RUNNER_TEMP/wai-web-native.log"
+          grep -q "test result: ok. 37 passed; 0 failed" "$RUNNER_TEMP/wai-web-native.log"
       - name: wasm32 int_synth decode equals the frozen native corpus
         shell: bash
         working-directory: wai/wai-web
-- 
2.54.0 (Apple Git-157)


From 981974a93707fd9bec6289d1d56c8bf1cc2a2dcc Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 16:42:48 -0500
Subject: [PATCH 28/29] joule-moq-relay DVR: count an eviction under the index
 lock; the by-age test counts late groups

Under load the relay can take a group whole only after the age window has
passed its number, because its streams finish out of order. Such a group is
late: it is never stored, and it is counted in groups_late. The by-age test
expected six evictions whatever the order, and failed 3 against 6 when three
groups came late. It now checks that every group is evicted or late, and that
each group stored is counted as it goes. A second run makes the out-of-order
case happen on purpose (group 0 and group 2's receipts, then, once the window
has passed group 2, group 1 and group 2's objects) and checks three stored,
three late.

The store also counted an eviction after releasing the index lock, so a
reader that found the group gone could read the count one short (5 against 6
with a delay injected there). The count now moves under the same lock as the
removal.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .../crates/joule-moq-relay/src/dvr.rs         |   8 +-
 .../crates/joule-moq-relay/tests/dvr.rs       | 143 ++++++++++++------
 2 files changed, 99 insertions(+), 52 deletions(-)

diff --git a/joule-code/joule-code-rs/crates/joule-moq-relay/src/dvr.rs b/joule-code/joule-code-rs/crates/joule-moq-relay/src/dvr.rs
index beb37743a..89d53da40 100644
--- a/joule-code/joule-code-rs/crates/joule-moq-relay/src/dvr.rs
+++ b/joule-code/joule-code-rs/crates/joule-moq-relay/src/dvr.rs
@@ -382,9 +382,11 @@ impl Index {
 pub struct DvrStats {
     pub groups_written: AtomicU64,
     pub objects_written: AtomicU64,
+    /// Groups stored that left the window, by age or by its byte budget: each media group
+    /// and each receipts group, counted as it goes.
     pub groups_evicted: AtomicU64,
     /// Groups that became whole after the window had moved past them (their receipts or
-    /// media group already evicted): not stored.
+    /// media group already evicted): not stored, so never counted as evicted.
     pub groups_late: AtomicU64,
     /// Objects dropped because their check or their receipt did not verify.
     pub objects_dropped: AtomicU64,
@@ -2050,9 +2052,11 @@ impl Dvr {
                     if let Some(r) = room {
                         ix.series.entry(r).or_default().change(now_us(), -(grp.bytes as i128), 0, 0, 0);
                     }
+                    // Counted under the lock it leaves the index under: whoever finds the
+                    // group gone (`groups`, `counts`) finds it counted.
+                    self.stats.groups_evicted.fetch_add(1, Ordering::Relaxed);
                     grp.seg
                 };
-                self.stats.groups_evicted.fetch_add(1, Ordering::Relaxed);
                 let open = w.open.get(&k).map(|s| s.seq);
                 self.retire(&k, seg, open);
             }
diff --git a/joule-code/joule-code-rs/crates/joule-moq-relay/tests/dvr.rs b/joule-code/joule-code-rs/crates/joule-moq-relay/tests/dvr.rs
index 0fa91d83e..429c79aaa 100644
--- a/joule-code/joule-code-rs/crates/joule-moq-relay/tests/dvr.rs
+++ b/joule-code/joule-code-rs/crates/joule-moq-relay/tests/dvr.rs
@@ -3,7 +3,9 @@
 //!
 //! - **The window.** A long publish crosses the window's byte budget: the oldest groups go,
 //!   whole, each media group with its receipts group, and the window holds what it says. A
-//!   window by age empties as its groups pass it.
+//!   window by age empties as its groups pass it, each group counted once: evicted, or late
+//!   when the relay took it whole only after the window had passed it (out of order, as a
+//!   loaded host takes streams).
 //! - **FETCH inside, across and before the window**, ascending and descending: inside, every
 //!   object, each verifying against the fetched receipts; across, an End of Unknown Range
 //!   for what was evicted and every object after it, from the store and the cache together;
@@ -110,13 +112,23 @@ fn bundle_of(key: &SigningKey, receipt_alias: u64, g: u64, payloads: &[Vec<u8>])
 
 /// Publish group `g` of `t`: its receipts, then its objects on one subgroup stream.
 async fn publish_multi(c: &Client, key: &SigningKey, t: &Tracks, g: u64, tag: &str) {
+    publish_parts(c, key, t, g, tag, true, true).await;
+}
+
+/// Publish group `g` of `t` in part: its receipts (with `receipts`), then its objects on one
+/// subgroup stream (with `media`).
+async fn publish_parts(c: &Client, key: &SigningKey, t: &Tracks, g: u64, tag: &str, receipts: bool, media: bool) {
     let payloads: Vec<Vec<u8>> = (0..OBJECTS).map(|o| object_payload(tag, g, o)).collect();
-    let bundle = bundle_of(key, t.receipt_alias, g, &payloads);
-    let rh = SubgroupHeader { track_alias: t.receipts_alias, group_id: g, subgroup_id: SubgroupId::Zero, publisher_priority: Some(t.priority), properties: false, end_of_group: true, first_object: true };
-    c.send_subgroup(rh, &[SubgroupObject::normal(0, bundle.to_payload().unwrap())]).await.expect("the receipts");
-    let mh = SubgroupHeader { track_alias: t.media_alias, group_id: g, subgroup_id: SubgroupId::Zero, publisher_priority: Some(t.priority), properties: false, end_of_group: true, first_object: true };
-    let objects: Vec<SubgroupObject> = payloads.into_iter().enumerate().map(|(o, p)| SubgroupObject::normal(o as u64, p)).collect();
-    c.send_subgroup(mh, &objects).await.expect("the objects");
+    if receipts {
+        let bundle = bundle_of(key, t.receipt_alias, g, &payloads);
+        let rh = SubgroupHeader { track_alias: t.receipts_alias, group_id: g, subgroup_id: SubgroupId::Zero, publisher_priority: Some(t.priority), properties: false, end_of_group: true, first_object: true };
+        c.send_subgroup(rh, &[SubgroupObject::normal(0, bundle.to_payload().unwrap())]).await.expect("the receipts");
+    }
+    if media {
+        let mh = SubgroupHeader { track_alias: t.media_alias, group_id: g, subgroup_id: SubgroupId::Zero, publisher_priority: Some(t.priority), properties: false, end_of_group: true, first_object: true };
+        let objects: Vec<SubgroupObject> = payloads.into_iter().enumerate().map(|(o, p)| SubgroupObject::normal(o as u64, p)).collect();
+        c.send_subgroup(mh, &objects).await.expect("the objects");
+    }
 }
 
 fn room_tracks(room: [u8; 32], publisher: &SigningKey) -> Tracks {
@@ -276,52 +288,83 @@ async fn window_run() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
 async fn a_window_by_age_empties_as_its_groups_pass_it() {
-    timeout(Duration::from_secs(60), async {
-        let dir = store_dir("age");
-        let room = [0x53; 32];
-        let publisher = SigningKey::from_bytes(&[0x54; 32]);
-        let t = room_tracks(room, &publisher);
-        let cfg = dvr_config(&dir, 2, 1 << 30, vec![]);
-        let server = relay(|c| {
-            // About two groups: what aged out of the store is out of the cache too.
-            c.cache.max_bytes = 8 * 1024;
-            c.dvr = Some(cfg.clone());
-        })
-        .await;
-        let (p, _) = connect(&server, Via::WebSocket, ClientOptions::default()).await;
-        let _pub = publish_tracks(&p, &t, None).await;
+    timeout(Duration::from_secs(60), age_run(false)).await.expect("in time");
+}
+
+/// The same window when the relay takes the groups whole out of order, as a loaded host does
+/// (each stream is read by a task of its own): group 0 and group 2's receipts first, then,
+/// once the window has passed group 2, group 1 and group 2's objects.
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn a_window_by_age_passes_groups_taken_out_of_order() {
+    timeout(Duration::from_secs(60), age_run(true)).await.expect("in time");
+}
+
+async fn age_run(out_of_order: bool) {
+    let dir = store_dir(if out_of_order { "age-order" } else { "age" });
+    let room = [0x53; 32];
+    let publisher = SigningKey::from_bytes(&[0x54; 32]);
+    let t = room_tracks(room, &publisher);
+    let cfg = dvr_config(&dir, 2, 1 << 30, vec![]);
+    let server = relay(|c| {
+        // About two groups: what aged out of the store is out of the cache too.
+        c.cache.max_bytes = 8 * 1024;
+        c.dvr = Some(cfg.clone());
+    })
+    .await;
+    let (p, _) = connect(&server, Via::WebSocket, ClientOptions::default()).await;
+    let _pub = publish_tracks(&p, &t, None).await;
+    let d = server.relay().dvr().unwrap().clone();
+    let (key, rkey) = ((t.ns.clone(), t.media.clone()), (t.ns.clone(), t.receipts.clone()));
+    let empty = || d.groups(&key).is_empty() && d.groups(&rkey).is_empty();
+    if out_of_order {
+        // Group 0, then group 2's receipts, each taken before the next is sent: all three
+        // are stored.
+        publish_multi(&p, &publisher, &t, 0, "a").await;
+        written(&server, 2).await;
+        publish_parts(&p, &publisher, &t, 2, "a", true, false).await;
+        written(&server, 3).await;
+        assert!(until(Duration::from_secs(10), empty).await, "past max_age_s groups 0 and 2 go");
+        publish_multi(&p, &publisher, &t, 1, "a").await;
+        publish_parts(&p, &publisher, &t, 2, "a", false, true).await;
+    } else {
         for g in 0..3 {
             publish_multi(&p, &publisher, &t, g, "a").await;
         }
-        written(&server, 6).await;
-        let d = server.relay().dvr().unwrap().clone();
-        let (key, rkey) = ((t.ns.clone(), t.media.clone()), (t.ns.clone(), t.receipts.clone()));
-        assert!(until(Duration::from_secs(10), || d.groups(&key).is_empty() && d.groups(&rkey).is_empty()).await, "past max_age_s every group goes");
-        assert_eq!(d.counts().groups_evicted, 6);
-        // The next group, fetched at once: what aged out is unknown, and the group is served.
-        publish_multi(&p, &publisher, &t, 3, "a").await;
-        // The relay has taken group 3 into both tracks before the fetches: taking it may
-        // evict a pair from the cache, and the media and receipts FETCHes are two requests,
-        // so a pair evicted between them would leave media without receipts.
-        let relay = server.relay().clone();
-        assert!(
-            until(STEP, || relay.largest(&key) == Some(Location::new(3, OBJECTS - 1)) && relay.largest(&rkey) == Some(Location::new(3, 0))).await,
-            "the relay took group 3"
-        );
-        let (v, _) = connect(&server, Via::WebSocket, ClientOptions::default()).await;
-        let f = fetch_verified(&v, &t, &publisher, range((0, 0), 3)).await;
-        // What aged out is unknown through the last group neither the store nor the cache
-        // holds; every group after it comes whole.
-        assert_eq!(f.unknown.len(), 1);
-        let k = f.unknown[0];
-        assert!(k.object == u64::MAX && k.group < 3, "{k:?}");
-        let want: Vec<(u64, u64)> = (k.group + 1..=3).flat_map(|g| (0..OBJECTS).map(move |o| (g, o))).collect();
-        assert_eq!(f.objects.keys().copied().collect::<Vec<_>>(), want);
-        server.shutdown().await;
-        let _ = std::fs::remove_dir_all(&dir);
-    })
-    .await
-    .expect("in time");
+    }
+    written(&server, 6).await;
+    assert!(until(Duration::from_secs(10), empty).await, "past max_age_s every group goes");
+    // Each group was stored and then evicted, or was late: whole only once the window had
+    // passed its number, and never stored. Each group stored is counted as it goes, the
+    // receipts group evicted with its media group too.
+    let c = d.counts();
+    assert_eq!(c.groups_evicted + c.groups_late, 6, "every group evicted or late: {c:?}");
+    assert_eq!(c.groups_evicted, c.groups_written, "each group stored is counted as it goes: {c:?}");
+    if out_of_order {
+        // Group 0 and group 2's receipts were stored; group 1 and group 2's objects came after
+        // the window had passed group 2.
+        assert_eq!((c.groups_written, c.groups_late), (3, 3), "{c:?}");
+    }
+    // The next group, fetched at once: what aged out is unknown, and the group is served.
+    publish_multi(&p, &publisher, &t, 3, "a").await;
+    // The relay has taken group 3 into both tracks before the fetches: taking it may
+    // evict a pair from the cache, and the media and receipts FETCHes are two requests,
+    // so a pair evicted between them would leave media without receipts.
+    let relay = server.relay().clone();
+    assert!(
+        until(STEP, || relay.largest(&key) == Some(Location::new(3, OBJECTS - 1)) && relay.largest(&rkey) == Some(Location::new(3, 0))).await,
+        "the relay took group 3"
+    );
+    let (v, _) = connect(&server, Via::WebSocket, ClientOptions::default()).await;
+    let f = fetch_verified(&v, &t, &publisher, range((0, 0), 3)).await;
+    // What aged out is unknown through the last group neither the store nor the cache
+    // holds; every group after it comes whole.
+    assert_eq!(f.unknown.len(), 1);
+    let k = f.unknown[0];
+    assert!(k.object == u64::MAX && k.group < 3, "{k:?}");
+    let want: Vec<(u64, u64)> = (k.group + 1..=3).flat_map(|g| (0..OBJECTS).map(move |o| (g, o))).collect();
+    assert_eq!(f.objects.keys().copied().collect::<Vec<_>>(), want);
+    server.shutdown().await;
+    let _ = std::fs::remove_dir_all(&dir);
 }
 
 /// Rewrite one object of the group record whose frame is at `frame_at` in a segment, in
-- 
2.54.0 (Apple Git-157)


From 4e8e4fb0981c083fbb28b73c5a5aa1bed4d8ac44 Mon Sep 17 00:00:00 2001
From: Transaction Science <hello@transaction.science>
Date: Wed, 4 Nov 2026 18:56:36 -0500
Subject: [PATCH 29/29] watch-together: the page makes its audio context before
 its join; a slot is set apart only when the machine ran nothing of the page
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

The watch-together example failed twice on the macOS runner for #616, on a loaded host (load
average about 50, then about 30). Each time the first page in the browser, Ada's, signed its
first presentation statement after its join, at 3.83 s and then 3.44 s on the room clock, and
then none in the slots from 4 s and 5 s (and 6 s the second time), while it was still entering
(begin()): it followed the log from 7.49 s, then 8.49 s, 3.7 s and 5.0 s after its first
statement, where Ben's page took 1.2 s and 1.0 s. Ada's page was silent at least 2.17 s, then
at least 3.56 s, with nothing refused, and was never taken for gone. What held it on the runner
was not measured.

What was measured:
- On the CI host, with CI jobs running (load average 30 to 39), in a browser launched as the
  example's test launches it (headless, the same flags, at nice 10), on a blank page with
  nothing else in it: the first audio context in a fresh browser held its page's thread 172 to
  192 ms (3 runs), for the constructor call alone; a second page's took 3.8 to 4.0 ms; neither
  page's thread had a gap over 60 ms in the 1.5 s after. That is a tenth or less of Ada's
  silences. Moving the context before the join takes about 0.2 s, as measured there, out of the
  stretch in which the page must state; what held Ada's page for the rest is unexplained. The
  rest of begin() (its audio worklet, its relay session, its WebAssembly, its first draws, the
  camera and microphone) was not measured there, nor an audio context made beside that work.
- On a laptop, in 25 runs of the example: the first page's audio context 94 to 128 ms, the
  second's 0.8 to 1.4 ms.
- On a laptop, on main, two injected faults each leave empty the slots just after the first
  page's first statement, as in CI: the audio-context call held 2.5 s (a busy wait in it; the
  slots from 1 s and 2 s, 1 run), and the first page's renderer process stopped 2.5 s from its
  join (SIGSTOP; the slot from 2 s, 2 of 2 runs). Neither was measured on the CI host.

This change is not shown to make the example pass on the CI host. If what held Ada's page there
was the page's own work on its thread, or the host running the page's other threads and not
that one, the check below still fails (its heartbeat ticking), and what it prints does not tell
those two apart.

1. The page makes the composition's audio context before it enters the co-watch log (app.js
   enter(), the new Compositor.openAudio()), since from that join it states in every slot
   (co-watch §7.3), so that hold is not in the stretch in which it must state. The rest of the
   entry (begin()) still runs after the join. begin() takes the compositor the entry made; in
   any other room it makes one, as before. The snapshot carries when the context was made, on
   the room clock, and how long making it held the thread (cowatch.audio_open).

2. A compositor made for an entry that failed is closed when the next entry makes its own
   (Compositor.close(): its mix timer, its worklet node, its drawing, its frames and its audio
   context). Before, it stayed, its audio context open and its mix timer running.

3. The statement loop records its stalls (cowatch.js run(), snapshot().stalls): each gap over
   half a slot between two of its passes, `{ from, to }` on the page's estimate of the room
   clock, with how long its wait and its signing took in it, on that clock (`timer_us`,
   `signing_us`); the rest of a stall is the loop's own work. It states every half slot while
   it runs, so it leaves a slot empty only inside such a stall. Its wait is a method (`wait`),
   and roomTime(ms) gives the room clock at an instant of the page's clock. The snapshot
   carries the loop's wait (`pass_ms`) and the page's clock beside the room clock (`wall_ms`).
   The co-watch rule and §7.3 are unchanged.

4. The example's slot check (co-watch §7.3: a statement in every slot while present) sets an
   empty slot apart only if the machine ran nothing of the page in the slot's second half: a
   stall of the loop spans that half; a heartbeat worker that the test starts in each page,
   ticking every 20 ms, did not tick in it either (to within a pass at each end); and the stall
   is the loop's wait and its signing but for at most one pass of its own work. Otherwise the
   check fails and says why: no stall of the loop (a statement lost); the heartbeat ticking (the
   page's thread did not run its loop while the worker ran: the page's own work holding the
   thread, or the host running the worker and not the page's thread, which the check does not
   tell apart); or more than a pass of the loop's own work (its own code, an await outside its
   measures, or the thread taken off the processor inside it). On a host that runs the page's
   threads unevenly, the check can fail in these ways with no defect in the page. Pass or fail,
   it prints each page's audio-context time, its loop's stalls and its heartbeat's silences over
   100 ms. They show whether the page's thread or the whole page was held, not what held it.

Unit tests (cowatch.test.mjs) run the loop on a room clock the test moves: no stall and every
slot held while it runs; a thread held while the loop waited, recorded once with all of the
stall its wait, and the only empty slots those it spans; a gap just over half a slot recorded
and one of half a slot not; a stall in a pass's own work recorded with a wait of one pass, the
rest neither wait nor signing.

Verified on a laptop with this change on main (152aab5c7), in a browser launched as the
example's test launches it and niced:
- From an archive build.sh made, with the WebAssembly build-web.sh made from crates this change
  does not touch, run-examples.sh browser passes video-room, watch-together, messaging,
  remote-desktop and package-and-deliver in each of 2 runs, and watch-together alone 2 more.
- With this room client served beside an earlier archive's binaries (the same Rust but for the
  relay's DVR change in the commit before), watch-together passes 3 of 3, and the faults below
  ran on that set. In each passing run the first page's audio context was made before its join
  and Ada's loop had no stall. Ben's deliberate hold, a synchronous request, shows as a stall of
  3.8 to 4.2 s with his heartbeat never silent.
- The room client's unit tests pass under node --test (77).
- Falsified, each fault put into a scratch copy of the room client, 2 runs each unless said:
  the audio context made in begin() again and that call held 2.5 s (a busy wait), the check
  fails: no statement in the slots from 1 s and 2 s (Ada) and 4 s and 5 s (Ben), each in a
  stall of the loop with its heartbeat ticking, the context printed as made after the join in
  2.5 to 2.6 s. A 2.5 s busy wait just after the join, or 4 s after it, fails it the same way:
  the slots it spans, with the heartbeat ticking. Every statement of the third whole slot after
  each join dropped fails it (no stall of its loop). One 1.5 s await in the loop's own work
  fails it (heartbeat ticking, 1.5 s neither its wait nor its signing); the same await inside
  the loop's wait fails it on the heartbeat alone. The page's renderer process stopped 2.5 s
  (SIGSTOP) at Ada's join, or 2.5 s after it, or Ben's 1 s after his (1 run): it passes, the
  slots inside the stop set apart; the heartbeat was silent and the loop stalled over the stop,
  to within 25 ms. The stop at Ada's join and a 2.5 s busy wait 4 s after it (1 run): the
  stop's slot is set apart and the busy wait's fail.
- In the unit tests, the wait measured on another clock than the stall, no stall recorded, a
  stall recorded only past a whole slot, and the wait left out of the record each fail one or
  two of the stall tests.
- A video room: one page publishes; the other presses Watch, its first relay connection is
  refused (a fault the probe injects), and it presses Watch again. Then that page holds one
  audio context open (the first closed) and one mix timer, and mixes 50 blocks of 20 ms a
  second. On main it holds two open and two timers; with close() doing nothing, two open. The
  same in a watch room: one open with this change, two on main.
- Not shown: that the example passes on the CI host.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---
 .../crates/joule-moq-web/web/room/app.js      | 34 ++++++--
 .../crates/joule-moq-web/web/room/compose.js  | 34 +++++++-
 .../crates/joule-moq-web/web/room/cowatch.js  | 50 +++++++++--
 .../joule-moq-web/web/room/cowatch.test.mjs   | 87 ++++++++++++++++++-
 .../test/watch-together.test.mjs              | 84 +++++++++++++++---
 5 files changed, 263 insertions(+), 26 deletions(-)

diff --git a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/app.js b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/app.js
index d618f01e3..7228a8a0d 100644
--- a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/app.js
+++ b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/app.js
@@ -521,11 +521,17 @@ async function enter(role, given = null) {
   window.roomTest.role = remoteRoom ? "viewer" : ROLE_OF[role];
   // A watch room's page enters its co-watch log first: the clock's `join` for it is then on the
   // clock's events track, which the page subscribes to next, and which a room of viewers alone
-  // has no other record on yet.
+  // has no other record on yet. From that join the page states what it presents in every slot
+  // (co-watch §7.3), so the composition's audio context is made before it: making one waits, on
+  // the page's thread, until the browser has its audio output (the first context in a browser
+  // starts it, about 0.2 s on a loaded test runner; a few ms for the next). The rest of its
+  // entry (begin()) still runs after the join.
+  const made = v.watch && fx.compose ? makeCompositor() : null;
+  if (made) made.openAudio();
   if (v.watch) await startCoWatch(v, r.body, seed, principal);
   // A remote desktop room's page publishes nothing but its events: it enters as a viewer does.
   try {
-    await begin(remoteRoom ? "viewer" : ROLE_OF[role], stream);
+    await begin(remoteRoom ? "viewer" : ROLE_OF[role], stream, made);
   } catch (e) {
     // A watch room's page that has no session follows no log and applies no control: it stops
     // stating, rather than state a presentation it does not keep, and the room takes it for gone.
@@ -552,8 +558,18 @@ function constraintsOf(d) {
   };
 }
 
-// The room, the composition and the session, once the page knows who it is.
-async function begin(role, stream) {
+// The room's composition, drawn on the page's canvas (./compose.js), made for one entry. One an
+// earlier entry made, which failed, is closed first: its mix, its drawing and its audio context.
+function makeCompositor() {
+  if (compositor) compositor.close();
+  compositor = new Compositor(wai, $("canvas"), unhex(fx.compose), { log, bindingOf: (vf) => SideTable.of(vf), onError: fail });
+  return compositor;
+}
+
+// The room, the composition and the session, once the page knows who it is. `made`: the
+// composition its entry made already (a watch room's, its audio context made before its join);
+// without one, begin() makes it.
+async function begin(role, stream, made = null) {
   roomId = unhex(fx.room);
   t0 = fx.t0_us;
   const who = $("who");
@@ -566,7 +582,7 @@ async function begin(role, stream) {
   // A remote desktop room (WAI remote-desktop) has no composition: its host's screen is drawn
   // on a canvas of its own.
   if (role !== "clock" && fx.compose) {
-    compositor = new Compositor(wai, $("canvas"), unhex(fx.compose), { log, bindingOf: (vf) => SideTable.of(vf), onError: fail });
+    compositor = made || makeCompositor();
     if (DEBUG_BINDING) setInterval(() => { try { renderStates($("binding-debug"), bindingRows()); } catch (e) { fail(e); } }, 500);
     await compositor.startAudio(new URL("./worklet.js", import.meta.url));
     await startTermination();
@@ -2126,7 +2142,13 @@ window.roomTest.snapshot = () => ({
   e2ee: e2ee && { ...state.e2ee, ...e2ee.stats, at: e2ee.at, stopped: e2ee.stopped, member: e2ee.state(), dom: $("e2ee") && !$("e2ee").hidden ? $("e2ee").textContent : null },
   age: state.age,
   co: co && co.snapshot(),
-  cowatch: cowatch && { ...cowatch.snapshot(), dom: $("cw-sync") ? $("cw-sync").textContent : null },
+  // `audio_open`: when the composition's audio context was made, on the room clock, and how
+  // long making it held the page's thread, beside the stalls of the page's statement loop.
+  cowatch: cowatch && {
+    ...cowatch.snapshot(),
+    audio_open: compositor && compositor.audioOpened ? { at: cowatch.roomTime(compositor.audioOpened.wall_ms), took_us: compositor.audioOpened.took_us } : null,
+    dom: $("cw-sync") ? $("cw-sync").textContent : null,
+  },
   termination: terminationSnapshot(),
 });
 // A shared control of a watch room (co-watch §5): `play`, `pause`, `seek` (to, µs) or `rate`
diff --git a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/compose.js b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/compose.js
index b4967e67a..04862454a 100644
--- a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/compose.js
+++ b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/compose.js
@@ -91,6 +91,7 @@ export class Compositor {
     this.waiting = true;
     requestAnimationFrame(() => {
       this.waiting = false;
+      if (!this.started) return; // closed while it waited
       try {
         this.draw();
       } catch (e) {
@@ -151,8 +152,23 @@ export class Compositor {
 
   // ------------------------------------------------------------------------------ audio
 
+  /** The audio context the mix plays into, made now if it is not yet. Making one holds the
+   * page's thread until the browser has its audio output, which the first context in a browser
+   * starts: a page that must not stall later (a watch room's, stating twice a slot) makes it
+   * first. `audioOpened`: when it was made (`wall_ms`, the page's `Date.now()`) and how long
+   * making it held the thread (`took_us`). */
+  openAudio() {
+    if (!this.ac) {
+      const wall = Date.now();
+      const t = performance.now();
+      this.ac = new AudioContext({ sampleRate: this.desc.params.sample_rate, latencyHint: "interactive" });
+      this.audioOpened = { wall_ms: wall, took_us: Math.round((performance.now() - t) * 1000) };
+    }
+    return this.ac;
+  }
+
   async startAudio(url) {
-    this.ac = new AudioContext({ sampleRate: this.desc.params.sample_rate, latencyHint: "interactive" });
+    this.openAudio();
     await this.ac.audioWorklet.addModule(url);
     this.worklet = new AudioWorkletNode(this.ac, "room-mix", { outputChannelCount: [2] });
     this.worklet.port.onmessage = (e) => { this.workletStats = e.data; };
@@ -161,6 +177,22 @@ export class Compositor {
     this.mixTimer = setInterval(() => this.mixOnce(), 20);
   }
 
+  /** Stop for good, as when the entry it was made for failed and another makes its own: no more
+   * mixing or drawing, its frames released and its audio context closed. */
+  close() {
+    this.started = false;
+    clearInterval(this.mixTimer);
+    this.mixTimer = null;
+    if (this.worklet) {
+      this.worklet.port.onmessage = null;
+      this.worklet.disconnect();
+      this.worklet = null;
+    }
+    for (const vf of this.video.values()) vf.close();
+    this.video.clear();
+    if (this.ac && this.ac.state !== "closed") this.ac.close().catch(() => {});
+  }
+
   addAudio(principal, audioData) {
     let v = this.audio.get(principal);
     if (!v) { v = { chunks: [], channels: audioData.numberOfChannels }; this.audio.set(principal, v); }
diff --git a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.js b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.js
index 5a25db31c..8480da929 100644
--- a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.js
+++ b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.js
@@ -277,6 +277,10 @@ const POST_EVENTS = 4;
 // Times a page taken for gone enters again with nothing of it carried since its last entry
 // (taken for gone again at once, as by a second stall) before it stops.
 const REENTRIES = 3;
+// The player's wait between two passes of its loop, in ms, and the stalls of the loop a page
+// keeps (`stalls`).
+const PASS_MS = 50;
+const STALLS_KEPT = 256;
 
 export class CoWatch {
   // o: { api (path, body) -> {status, body}, wai, room, t0, seed (Uint8Array), principal (hex),
@@ -304,14 +308,20 @@ export class CoWatch {
     this.others = {}; // principal -> its last statement: at, position, error
     // `gone`: posts refused `not_present`, the room having taken the page for gone; `rejoins`:
     // its entries again; `resigned`: controls signed again after one; `dropped`: statements
-    // and controls signed before one and not signed again.
-    this.stats = { log: 0, controls: 0, applied: 0, refused: {}, statements: 0, posted: 0, post_refused: {}, seeks: 0, nudges: 0, verified_source: false, gone: 0, rejoins: 0, resigned: 0, dropped: 0, followed_at: null, errors: [] };
+    // and controls signed before one and not signed again. `stalls`: the stretches in which
+    // the player's loop did not run (`run`), the last STALLS_KEPT of them.
+    this.stats = { log: 0, controls: 0, applied: 0, refused: {}, statements: 0, posted: 0, post_refused: {}, seeks: 0, nudges: 0, verified_source: false, gone: 0, rejoins: 0, resigned: 0, dropped: 0, followed_at: null, stalls: [], errors: [] };
     this.onChange = o.onChange || (() => {});
   }
 
   /** The room clock now, on this page's estimate (§7.3). */
   now() {
-    return Math.max(0, Math.round(Date.now() * 1000 + this.offset - this.t0));
+    return this.roomTime(Date.now());
+  }
+
+  /** The room clock at `ms` on this page's clock (`Date.now()`), on its estimate. */
+  roomTime(ms) {
+    return Math.max(0, Math.round(ms * 1000 + this.offset - this.t0));
   }
 
   async estimateClock() {
@@ -547,12 +557,30 @@ export class CoWatch {
     return false;
   }
 
-  /** Keep the media at the controlled position (§9) and state what it presents (§7.3). */
+  /** Keep the media at the controlled position (§9) and state what it presents (§7.3).
+   *
+   * Each pass of the loop states, if half a slot has passed since its last statement, then
+   * waits PASS_MS. Two statements are so at most half a slot and the gap between two passes
+   * apart, and while no such gap is over half a slot every slot holds one: a slot is left empty
+   * only if no pass began in its second half. The page records each gap between two passes
+   * over half a slot (`stalls`): `{ from, to }`, and how long the wait and the signing in it
+   * took (`timer_us`, `signing_us`), all on its estimate of the room clock. The rest of the gap
+   * is the loop's own work. A wait or a signing that took long is the page's thread held, by
+   * the page's own work or by its machine not running it; the record does not tell which. */
   async run(alive) {
     const per = (this.policy.period_ms * 1000) / STATES_PER_SLOT;
     let next = 0;
+    let last = null; // the room clock at the previous pass
+    let timer = 0; // how long its wait and its signing took, on the room clock (µs)
+    let signing = 0;
     while (alive() && !this.stopped) {
       const x = this.now();
+      if (last != null && x - last > per) {
+        this.stats.stalls.push({ from: last, to: x, timer_us: timer, signing_us: signing });
+        if (this.stats.stalls.length > STALLS_KEPT) this.stats.stalls.shift();
+      }
+      last = x;
+      signing = 0;
       const st = this.timeline.stateAt(x);
       const duration = this.timeline.durationAt(x);
       const controlled = this.timeline.positionAt(x);
@@ -576,7 +604,9 @@ export class CoWatch {
         const pos = Math.round(this.media.currentTime * 1e6);
         const body = { room: this.room, player: this.principal, source: this.policy.source, at: x, position: pos, state, rate: Math.max(1, Math.round(this.media.playbackRate * 1e6)) };
         const entries = this.stats.rejoins;
+        const s0 = this.now();
         const signed = await signPresentation(body, this.key.privateKey);
+        signing = this.now() - s0;
         if (this.stats.rejoins !== entries) {
           // The page entered again while it signed (`flush`): `x` is an instant from before its
           // present join, which it states nothing of. It states anew at once.
@@ -589,16 +619,24 @@ export class CoWatch {
         }
       }
       this.onChange();
-      await sleep(50);
+      const w = this.now();
+      await this.wait(PASS_MS);
+      timer = this.now() - w;
     }
   }
 
+  /** The loop's wait between two passes. */
+  wait(ms) {
+    return sleep(ms);
+  }
+
   /** What a test reads, and what the page shows. */
   snapshot() {
     const last = this.skews.length ? this.skews[this.skews.length - 1] : null;
     const lastWindow = last ? this.skews.filter((s) => s.from === last.from) : [];
+    const wall = Date.now();
     return {
-      principal: this.principal, joined: this.joinT, at: this.at, offset_us: this.offset, room_now: this.now(), controls: this.controls.length, outbox: this.outbox.length,
+      principal: this.principal, joined: this.joinT, at: this.at, offset_us: this.offset, room_now: this.roomTime(wall), wall_ms: wall, pass_ms: PASS_MS, controls: this.controls.length, outbox: this.outbox.length,
       overridden: this.controls.filter((c) => this.timeline.overridden.has(c.digest)).length,
       controlled: this.controlled ?? null, error_us: this.error ?? null,
       playing: !this.media.paused, rate: this.media.playbackRate, current_time: this.media.currentTime,
diff --git a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.test.mjs b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.test.mjs
index 3c6dd4c68..c1bc436a9 100644
--- a/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.test.mjs
+++ b/joule-code/joule-code-rs/crates/joule-moq-web/web/room/cowatch.test.mjs
@@ -3,8 +3,9 @@
 // its correction (co-watch §9); each accepted room's controlled position recomputed from its
 // policies and controls (§5.4, §6), as the corpus states it at every change and at every
 // statement; its controls written byte for byte (§5.1); its presentation statements
-// written and signed byte for byte (§7.1), each at the rate the catch-up rule gives (§9); and
-// the player taken for gone by the room re-entering with its chain whole (§7.3, §10).
+// written and signed byte for byte (§7.1), each at the rate the catch-up rule gives (§9); the
+// player taken for gone by the room re-entering with its chain whole (§7.3, §10); and the
+// player stating in every slot while its loop runs, recording each stretch it did not (§7.3).
 import test from "node:test";
 import assert from "node:assert/strict";
 import { createHash } from "node:crypto";
@@ -135,7 +136,7 @@ const fakeWai = {
 // sign (503). `leaveAfterJoin`: the next joins whose presence a `leave` the room had asked
 // before them, signed after them, ends at once (a service that ordered them so).
 function watchService(lead_ms, late) {
-  const s = { now: 0, present: false, joined: 0, lastT: 0, joins: 0, chain: [], controls: [], failJoins: 0, leaveAfterJoin: 0 };
+  const s = { now: 0, present: false, joined: 0, lastT: 0, joins: 0, chain: [], controls: [], presentations: [], failJoins: 0, leaveAfterJoin: 0 };
   s.api = async (path, body) => {
     if (path === "/watch/challenge") return { status: 200, body: { challenge: "00".repeat(32) } };
     if (path === "/watch/join") {
@@ -167,6 +168,7 @@ function watchService(lead_ms, late) {
       s.chain.push(createHash("sha256").update(cw.unhex(h)).digest("hex"));
       s.lastT = ev.t;
       if (st && st.kind === "control") s.controls.push({ action: st.action, at: st.at, t: ev.t });
+      if (st && st.kind === "presentation") s.presentations.push(st.at);
     }
     return { status: 200, body: { ok: true, carried: body.events.length } };
   };
@@ -328,3 +330,82 @@ test("a control pressed while the room had taken the page for gone is dropped wh
   assert.deepEqual(svc.controls.map((x) => x.action), [cw.ACTION.play, cw.ACTION.seek]);
   assert.ok(svc.controls[1].at > seekAt && svc.controls[1].at === svc.joined + lead * 1000 + 50_000);
 });
+
+/** The player's loop run for `passes` passes on a room clock the test moves, so the test, not
+ * the host, sets each pass's instant: by `step(i)` µs in the loop's `i`th wait between two
+ * passes, and by `work(i)` µs in the work of its `i`th pass (in `onChange`, which the loop
+ * calls in each pass, as work of its own or an await it does not measure would be); then
+ * until the page has posted all it holds. */
+async function runFor(c, svc, passes, step, work = () => 0) {
+  let i = 0;
+  c.media = { currentTime: 0, paused: true, ended: false, readyState: 4, playbackRate: 1, play: () => Promise.resolve(), pause: () => {} };
+  c.now = () => svc.now;
+  c.onChange = () => { svc.now += work(i); };
+  c.wait = async () => { svc.now += step(i++); await new Promise((r) => setTimeout(r, 0)); };
+  await c.run(() => i < passes);
+  await settled(c);
+}
+
+/** The slots from the page's first statement to its last: each held by a statement, else
+ * empty, and of those whose second half a stall of the page spans. */
+function slotsOf(ats, stalls, P) {
+  const out = { held: 0, empty: [], stalled: [] };
+  for (let s = Math.ceil(ats[0] / P) * P; s + P <= ats[ats.length - 1]; s += P) {
+    if (ats.some((x) => x >= s && x < s + P)) out.held++;
+    else (stalls.some((x) => x.from < s + P / 2 && x.to >= s + P) ? out.stalled : out.empty).push(s);
+  }
+  return out;
+}
+
+test("while its loop runs, the player states in every slot and records no stall", async () => {
+  const { svc, c, logs } = await entered();
+  const P = c.policy.period_ms * 1000;
+  await runFor(c, svc, 40, () => 50_000);
+  assert.equal(c.stopped, null, logs.join("\n"));
+  assert.deepEqual(c.stats.stalls, []);
+  const s = slotsOf(svc.presentations, c.stats.stalls, P);
+  assert.ok(s.held >= 1, JSON.stringify(s));
+  assert.deepEqual([s.empty, s.stalled], [[], []]);
+  for (let k = 1; k < svc.presentations.length; k++) assert.ok(svc.presentations[k] - svc.presentations[k - 1] <= P / 2 + 50_000);
+});
+
+test("the player records the stretch its thread was held, and only then leaves a slot empty", async () => {
+  const { svc, c, logs } = await entered();
+  const P = c.policy.period_ms * 1000;
+  // Its passes every 50 ms, but one: the page's thread held for three and a half slots while
+  // the loop waited, so its wait took that long.
+  const HELD = 15;
+  await runFor(c, svc, 50, (i) => (i === HELD ? 3.5 * P : 50_000));
+  assert.equal(c.stopped, null, logs.join("\n"));
+  assert.equal(c.stats.stalls.length, 1, JSON.stringify(c.stats.stalls));
+  const [st] = c.stats.stalls;
+  assert.equal(st.to - st.from, 3.5 * P);
+  // The whole stall is the wait: the loop's own work and its signing took no time on the clock.
+  assert.deepEqual([st.timer_us, st.signing_us], [3.5 * P, 0]);
+  // The slots the held thread left empty are those the stall spans; every other holds a
+  // statement, and none is empty outside it.
+  const s = slotsOf(svc.presentations, c.stats.stalls, P);
+  assert.ok(s.stalled.length >= 2, JSON.stringify(s));
+  assert.deepEqual(s.empty, []);
+  for (const x of s.stalled) assert.ok(st.from < x + P / 2 && st.to >= x + P);
+  // A pass more than half a slot after the one before is a stall: the loop states every half
+  // slot, so only such a stretch can leave a slot empty.
+  const { svc: svc2, c: c2 } = await entered();
+  await runFor(c2, svc2, 30, (i) => (i === HELD ? P / 2 + 1 : i === HELD + 5 ? P / 2 : 50_000));
+  assert.deepEqual(c2.stats.stalls.map((x) => x.to - x.from), [P / 2 + 1]);
+});
+
+test("a stall in the loop's own work is recorded with a wait that does not cover it", async () => {
+  const { svc, c, logs } = await entered();
+  const P = c.policy.period_ms * 1000;
+  // One pass's own work takes two slots (or an await the loop does not measure would): its
+  // wait after it took 50 ms, and the stall is the rest.
+  const HELD = 15;
+  await runFor(c, svc, 50, () => 50_000, (i) => (i === HELD ? 2 * P : 0));
+  assert.equal(c.stopped, null, logs.join("\n"));
+  assert.equal(c.stats.stalls.length, 1, JSON.stringify(c.stats.stalls));
+  const [st] = c.stats.stalls;
+  assert.equal(st.to - st.from, 2 * P + 50_000);
+  assert.deepEqual([st.timer_us, st.signing_us], [50_000, 0]);
+  assert.equal(st.to - st.from - st.timer_us - st.signing_us, 2 * P);
+});
diff --git a/wai/examples/watch-together/test/watch-together.test.mjs b/wai/examples/watch-together/test/watch-together.test.mjs
index 892f5bfc0..f3614a261 100644
--- a/wai/examples/watch-together/test/watch-together.test.mjs
+++ b/wai/examples/watch-together/test/watch-together.test.mjs
@@ -10,8 +10,10 @@
 // play from Ada's seek at the instant the room carried, within the bound, until Ada pauses
 // them at one position again. At the end the room's log shows what happened, and each page's
 // own counts agree with it: one `leave`, Ben's, in the hold; each page stating from its join,
-// before it had its relay session, and in every slot while it was present (co-watch §7.3);
-// Ada never taken for gone, Ben once; and no page stopped.
+// before it had its relay session, and in every slot while it was present (co-watch §7.3),
+// but for a slot in whose second half the machine ran nothing of the page, neither its
+// statement loop nor a worker of its own; Ada never taken for gone, Ben once; and no page
+// stopped.
 //
 //   node wai/examples/watch-together/test/watch-together.test.mjs
 import { check, create, dev, dump, finish, frame, launch, open, sleep, STEP, tmpdir, until } from "../../../quickstart/test/lib.mjs";
@@ -29,6 +31,18 @@ try {
     pages.push(p);
     const f = await frame(p, "/room/");
     await until(`${name}: the room page ready`, STEP, () => f.evaluate(() => window.roomTest && window.roomTest.state && window.roomTest.state.phase === "idle"), 200);
+    // The page's heartbeat: a worker of its own that does nothing but tick, every 20 ms, and
+    // records each stretch over 100 ms between two ticks, on the page's clock (Date.now(), ms).
+    // It runs on a thread of its own, in the page's process: the page's work on its own thread
+    // does not hold it, and the machine running nothing of the page does. Its ticking while the
+    // page's thread did not run does not say why: the page's own work, or the machine running
+    // this thread and not that one.
+    await f.evaluate(() => {
+      const src = "let last = Date.now(); setInterval(() => { const t = Date.now(); if (t - last > 100) postMessage([last, t]); last = t; }, 20);";
+      const h = { gaps: [], worker: new Worker(URL.createObjectURL(new Blob([src], { type: "text/javascript" }))) };
+      h.worker.onmessage = (e) => h.gaps.push({ from: e.data[0], to: e.data[1] });
+      window.heartbeat = h;
+    });
     await f.click("#watch");
     const s = await until(`${name}: in the watch room`, STEP, async () => {
       const x = await f.evaluate(() => window.roomTest.snapshot());
@@ -108,21 +122,71 @@ try {
   check("Ada entered once, Ben twice: again after his leave", ja.length === 1 && jb.length === 2 && leaves.length === 1 && jb[1].entry > leaves[0].entry, `${ja.length} and ${jb.length} joins`);
   // Every slot wholly within a stretch of a page's presence holds a statement of it (co-watch
   // §7.3), from its join to a slot before the last statement the log holds, and for Ben's
-  // first stretch to a slot before his hold began.
+  // first stretch to a slot before his hold began. The page's loop states every half slot while
+  // it runs (cowatch.js `run`), so a slot it left empty is one in whose second half no pass of
+  // it began: a stall of the loop the page records (`stalls`), with how long its wait and its
+  // signing took. Such a slot is counted apart only if the machine ran nothing of the page in
+  // that half slot: the page's heartbeat (its worker, which the page's own work does not hold)
+  // has no tick there either, and the stall is its wait and its signing, but for at most one
+  // pass of the loop's own work. A slot left empty while the heartbeat ticked fails, whether
+  // the page's own work held its thread or the machine ran the worker and not the page's thread:
+  // the check does not tell the two apart, and counts neither apart. So does one left empty by
+  // an await of the loop's own, or with no stall at all.
   const stated = (p) => L.entries.filter((e) => e.principal === p && e.kind === "presentation").map((e) => e.at);
   const last = Math.max(...stated(A.principal), ...stated(B.principal)) - P;
+  // Each page's heartbeat, and its clock (ms) as the room clock (µs): room = ms * 1000 + k.
+  const beat = await Promise.all([A, B].map((x) => x.f.evaluate(() => {
+    const c = window.roomTest.snapshot().cowatch;
+    return { gaps: window.heartbeat.gaps, k: c.room_now - c.wall_ms * 1000 };
+  })));
   let slots = 0;
   let widest = 0;
-  const empty = (name, p, from, to) => {
-    const ats = stated(p).filter((x) => x >= from && x < to);
-    ats.forEach((x, i) => { if (i) widest = Math.max(widest, x - ats[i - 1]); });
+  const apart = [];
+  // Why the empty slot from `s` of page `i` fails the check; null if the machine ran nothing
+  // of the page in its second half.
+  const unexcused = (i, s) => {
+    const c = end[i];
+    const pass = c.pass_ms * 1000;
+    const st = c.stalls.find((x) => x.from < s + P / 2 && x.to >= s + P);
+    if (!st) return "no stall of its loop";
+    const why = [];
+    const [w0, w1] = [s + P / 2, s + P].map((r) => (r - beat[i].k) / 1000);
+    if (!beat[i].gaps.some((g) => g.from <= w0 + c.pass_ms && g.to >= w1 - c.pass_ms)) why.push("its heartbeat ticking");
+    const own = st.to - st.from - st.timer_us - st.signing_us;
+    if (own > pass) why.push(`${ms(own)} of it neither its wait nor its signing`);
+    return why.length ? `its loop's stall ${st.from}..${st.to}, ${why.join(", ")}` : null;
+  };
+  const empty = (i, from, to) => {
+    const name = [A, B][i].name;
+    const ats = stated(end[i].principal).filter((x) => x >= from && x < to);
+    ats.forEach((x, k) => { if (k) widest = Math.max(widest, x - ats[k - 1]); });
     const out = [];
-    for (let s = Math.ceil(from / P) * P; s + P <= to; s += P, slots++) if (!ats.some((x) => x >= s && x < s + P)) out.push(`${name} ${s}`);
+    for (let s = Math.ceil(from / P) * P; s + P <= to; s += P, slots++) {
+      if (ats.some((x) => x >= s && x < s + P)) continue;
+      const why = unexcused(i, s);
+      if (why) out.push(`${name} ${s}: ${why}`);
+      else apart.push(`${name} ${s}`);
+    }
     return out;
   };
-  const gaps = ja.length === 1 && jb.length === 2 ? [...empty("ada", A.principal, ja[0].t, last), ...empty("ben", B.principal, jb[0].t, hold.began - P), ...empty("ben", B.principal, jb[1].t, last)] : ["no stretches to check"];
-  check("each page stated what it presented in every slot while it was present (co-watch §7.3), but in Ben's hold", gaps.length === 0,
-    gaps.length ? `no statement in the slot from ${gaps.slice(0, 6).join(", ")}` : `${slots} slots; at most ${ms(widest)} between two statements of a page`);
+  const gaps = ja.length === 1 && jb.length === 2 ? [...empty(0, ja[0].t, last), ...empty(1, jb[0].t, hold.began - P), ...empty(1, jb[1].t, last)] : ["no stretches to check"];
+  // What each page recorded, printed whether the check holds or not: when its audio context was
+  // made and how long that held the thread, the stalls of its loop with their wait and signing,
+  // and the stretches its heartbeat did not tick, all on the room clock. They say whether the
+  // page's thread or the whole page was held, not what held it.
+  const seen = end.map((c, i) => {
+    const o = c.audio_open;
+    const joined = [ja, jb][i][0].t;
+    const st = c.stalls.slice(0, 8).map((x) => `${x.from}..${x.to} (wait ${ms(x.timer_us)}, signing ${ms(x.signing_us)})`);
+    const longest = [...beat[i].gaps].sort((x, y) => (y.to - y.from) - (x.to - x.from)).slice(0, 8).sort((x, y) => x.from - y.from);
+    const hb = longest.map((g) => `${g.from * 1000 + beat[i].k}..${g.to * 1000 + beat[i].k}`);
+    return `${[A, B][i].name}: audio context made at ${o ? `${o.at}, ${o.at < joined ? "before" : "after"} its first join at ${joined}, in ${ms(o.took_us)}` : "none"}; `
+      + `${c.stalls.length} stall(s) of its loop${st.length ? `: ${st.join(", ")}` : ""}; `
+      + `its heartbeat silent over 100 ms ${beat[i].gaps.length} time(s)${hb.length ? `, the longest ${hb.join(", ")}` : ""}`;
+  }).join("; ");
+  check("each page stated what it presented in every slot while it was present (co-watch §7.3), but in Ben's hold and where the machine ran nothing of the page", gaps.length === 0,
+    (gaps.length ? `no statement in the slot from ${gaps.slice(0, 6).join("; ")}` : `${slots} slots; at most ${ms(widest)} between two statements of a page`)
+      + (apart.length ? `; ${apart.length} empty with the machine running nothing of the page: ${apart.slice(0, 6).join(", ")}` : "") + `. ${seen}`);
   check("each page stated from its join, before it had its relay session and followed the log",
     [a, b].every((c) => c.followed_at != null && stated(c.principal).length && stated(c.principal)[0] < c.followed_at), [a, b].map((c, i) => `${[A, B][i].name}: first at ${stated(c.principal)[0]}, following from ${c.followed_at}`).join("; "));
   check("Ada's page was never taken for gone, Ben's once, entering again once with his play signed again; the room refused no other event of either, and no page stopped",
-- 
2.54.0 (Apple Git-157)