Transaction-Science/open-standards
Watch
1
0
Fork
You've already forked open-standards
 0
Code Issues Pull requests Projects Releases 15 Packages Wiki Activity Actions

wai.audio.opus.dred.int: DRED of draft-ietf-mlcodec-opus-dred-07 decoded exactly, an integer profile on wai-ops/1, a stream under loss by digest, and its receipts (B3a) #434

Closed
dcharlot wants to merge 0 commits from wai/dred-exact into main
pull from: wai/dred-exact
merge into: Transaction-Science:main
Conversation 0 Commits - Files changed -
dcharlot commented 2026-10-22 05:00:48 -04:00
Owner
Copy link

What this is

The deep redundancy (DRED) of draft-ietf-mlcodec-opus-dred-07 decoded exactly. DRED carries, in an Opus packet's padding, a compact neural latent of the recent speech, from which a receiver rebuilds the acoustic features of lost packets. The draft specifies the decoder network in floating point and admits any decoder within tolerances, so two receivers rebuild different features from the same packets. This adds an integer profile, dred-int/1: the draft's own network and its own weights file (pinned by SHA-256, never committed), lowered to wai-ops/1 graphs whose features are the same integers on every machine and on a GPU, with a stream-under-loss reconstruction named by a digest a verifier recomputes, and a player's statement that signs it.

  • Extension wai/extensions/opus-dred.md: registration (§2), finding DRED and its integers (§3), the profile (§4), a stream under loss (§5), the measurement (§6), receipts and synthesis's lattice (§7), refusals (§8), conformance (§9).
  • Registered (SPEC §5, §7, REGISTRY, export): wai.audio.opus.dred (the draft's decoding, StandardTolerance) and wai.audio.opus.dred.int (the profile, NeuralIntegerExact, one-file parameter set, ops lowering dred-int/1), both companion-only enhancement layers over wai.audio.opus.packet, both patent-pool.
  • wai-rs dred_exact module (feature dred_exact): RFC 6716 range decoder (and encoder, for the generator); Opus framing and the extension draft's iterator with its repeat mechanism; the payload's integers; the weights file parsed and expanded; the lowering; the driver; a deterministic f64 reference; the draft's feature comparison; the loss scenario; the corpus generator. Bin wai_dred_vectors.
  • Corpus wai/live-conformance/dred/ (vectors.json, README, fetch.sh) and its second implementation dred_verify.py (standard-library Python from the text, sharing no code with wai-rs).
  • Tests: tests/dred_exact.rs (corpus regenerates byte for byte; §8 codes; §4 digests; acc64-not-acc32), tests/dred_exact_measure.rs (§6's table on the draft's test vectors), tests/dred_exact_gpu.rs (the corpus on every GPU adapter, graph by graph against the interpreter).
  • CI: a step in wai-conformance.yml (independent-verifiers: fetch, unit, corpus, measurement, Python verifier) and in wai-ops-gpu.yml (build in checks, run on Metal in metal).
  • Proposal wai/proposals/dred-integer-profile.md: the profile as an optional integer conformance point for the working group. Not filed.
  • CHANGELOG, determinism-tiers.md §5/§6 rows, the research registry's class basis and licence note, the live-conformance index, a glossary exemption.

Decisions, and why

  • Companion (enhancement layer), not refinement. A refinement corrects what the base decoded; DRED corrects nothing the base decoded, it is a second, coarser encoding read only where the base is missing, and its offsets index one encode's frames. That is SPEC §5's enhancement-layer model (bound to exactly one rendition of a listed base, the base of no second layer), so both strings reuse AppliedForm::Enhancement and LCEVC's binding rule. Carried embedded (the draft's own carriage, no component) or separately (a WAI2 companion whose payload is one DRED payload for frame 0 of the packet in the same envelope). One base, wai.audio.opus.packet, because there the envelope aligns DRED with its packet; no alignment with an Ogg stream is pinned, as for LCEVC.
  • Two strings. A capability's class bounds the tier a pin may claim (SPEC §3.1), and the draft's decoding is admitted within tolerances, so it cannot back replicate. The exact profile needs its own NeuralIntegerExact string.
  • patent-pool, conservatively. The IETF's IPR disclosure 6063 states a royalty-free intent conditional on the draft becoming a standard; REGISTRY.md's rule classes an unestablished status patent-pool.
  • Weights pinned and fetched, never committed. The standalone weights file carries no licence text of its own; the same arrays ship in the draft's reference implementation under a BSD-style licence whose notice names its copyright holders, which the public-text rule keeps out of the tree. So the corpus names the file by the draft's SHA-256, and fetch.sh fetches it from the draft's location and checks the digest. Nothing derived from the weights is committed either (the corpus holds only digests, the drawn packets and the decoded rows' digests and one row per payload).
  • The arithmetic. Weights at 2^24, rounded half to even from the float32's exact bits (no float in any implementation); biases at the accumulator's scale; activations Q15; tanh read at Q12 and the sigmoid at Q11 from lut16 tables whose entries are a stated integer computation from one 62-bit constant (⌊e^(−1/2048)·2^62⌋, squared repeatedly), so a second implementation computes the same tables rather than trusting a float library; every rq half up; the GRU update written hc + z·(h − hc) so it rounds once; inputs Q16, x·256/s rounded half up. Every operator is in wai-ops/1-i32; the graphs meet acc64 (not acc32: a 16-bit activation times a 32-bit weight over up to 576 inputs), so the GPU runs them at acc64.
  • A magnitude cap (2^16, dred_latent_range): the draft's escape rule bounds nothing, and past a payload's end the range decoder reads zeros, so a crafted payload could keep a decoder reading. Where the draft's prose leaves a detail open (the magnitude is 1 + the escape symbols; the quantizer's rounding), the profile follows the draft's reference decoder, and §3.3 says so.
  • The vocoder is out of scope. The draft makes it non-normative and tests it by "sounds similar"; an exact profile of it would be WAI-invented. The profile's output is the feature sequence; §7.2's statement signs that, and says a player's statement of its vocoder's samples is its own.
  • Synthesis classes (§7.1): a reconstruction from DRED is convert (a decoder's reconstruction of what the packets carried); concealment beyond DRED's coverage is generate/synthesized (content the packets never carried, by a trained model; not assist, which edits an input that exists); speech coding enhancement and blind bandwidth extension are assist (as synthesis classes an upscale), tier attest.
  • The other mlcodec documents (§2.3): the extension format and the speech coding enhancement register nothing (no payload); the scalable quality extension rides embedded in the base; the test battery is informative.

Measured (§6), on the draft's eight test vectors

compared frames total max pitch largest difference band LSD mean (dB) band LSD max (dB)
integer, reference 16000 0.1314 0.0767 0.00492 0.4139 0.2945 1.0034
float, reference 16000 0.1314 0.0767 0.00492 0.4151 0.2944 1.0060
8-bit emulation, reference 16000 0.0831 0.0458 0.00255 0.2542 0.1844 0.6929
integer, float 16000 0.0021 0.0013 0.00011 0.0084 0.0048 0.0210
integer, float (packets) 80452 0.0022 0.0013 0.00004 0.0085 0.0049 0.0208

The profile passes the draft's feature test (0.5 / 0.15 / 0.02) on every vector. Informative, measured once by hand on the build host (arm64): the draft's reference decoder built in its default (8-bit) configuration lies 0.090–0.096 from its own test vectors, built with float weights 0.131–0.136; the two builds differ by about 0.13 from each other; the profile and the f64 network lie 0.015–0.018 from the float build. So the test vectors carry the arithmetic of the build that wrote them, and two builds of the reference decoder on one machine disagree; under the profile there is one answer.

Checks

check command result
unit tests cargo test --locked --no-default-features --features dred_exact --lib dred_exact:: (rb) 18 passed
corpus, §8, §4 digests, acc64/acc32 WAI_DRED_WEIGHTS=… cargo test --locked --release --no-default-features --features dred_exact --test dred_exact (rb) 4 passed
measurement WAI_DRED_WEIGHTS=… WAI_DRED_VECTORS=… cargo test --locked --release … --test dred_exact_measure (rb) 1 passed; table above
GPU (Metal, build host) WAI_OPS_GPU_REQUIRE=1 WAI_DRED_WEIGHTS=… cargo test --locked --release --no-default-features --features dred_exact,ops_gpu --test dred_exact_gpu (rb) dred-gpu: key=gpu-metal-0 payloads=3 streams=2 same=true graphs_on_gpu=29 (acc32 0, acc64 29) on_cpu=0 compared=29 differed=0 errors=[]
second implementation python3 -I live-conformance/dred_verify.py live-conformance/dred <weights> 9 payload vectors and 9 cases reach their outcomes; every §8 reason refuses one; all 16 mutated copies fail
registry and container cargo test --locked --no-default-features --lib -- codecs:: container::; cargo test --locked --lib (default) 103 passed; 193 passed
integration tests cargo test --locked --no-default-features --tests; cargo test --locked --tests (rb) all binaries pass (carriage.rs updated: DRED binds to an Opus packet)
registry export cargo run -q --no-default-features --bin wai -- registry export then --check (rb, --pull) up to date
clippy cargo clippy --locked --no-default-features --features dred_exact,ops_gpu --all-targets; cargo clippy --locked --all-targets (rb) no notes in touched files (three pre-existing collapsible_if notes in glossary_check.rs lines this change does not touch)
dependents cargo check --locked --target wasm32-unknown-unknown (wai-web); cargo check --locked (wai-pack) (rb) build
shared-text verifiers synthesis_verify.py, staged_verify.py, http_verify.py, pin_verify.py, pack_verify.py all pass
repo checks check-tests-in-ci.py, check-workflow-concurrency.sh, check-spec-refs.py, check-wai-section-refs.py, check-no-vanity-metrics.sh, test-public-text-checks.sh, check-comparative-prose.sh, check-orphan-sources.py all clean
public names check-public-names.sh on the 35 changed files; commit messages clean
lockfiles no dependency added; every cargo run above used --locked unchanged

Falsifications (committed first, each restored after)

  • Python verifier: dropping the output-digest check → exit 1 (float_decode_claimed_exact accepted); dropping the magnitude cap → exit 1 (magnitude_past_cap decodes); tanh entries floored instead of rounded → exit 1; the GRU update rounded twice → exit 1 (a payload's rows differ); dropping the loss-pattern check → exit 1.
  • Rust: one shift of the lowering changed (output 23 → 22) → the_corpus_regenerates… fails ("vectors.json differs"); one digit of a row digest in vectors.json → fails; an extra code in §8's table → the_refusals_table_is_the_modules fails (and the Python verifier exits 1); one figure of §6's table → the measurement test fails.
  • The Python verifier and wai-rs agree bit for bit on every row of every payload and stream (the corpus's row digests are wai-rs's; the verifier recomputes them in its own integer arithmetic).

Not done

  • The vocoder stays outside the profile (draft §2.2.4, §4.2); no segmental SNR, which needs samples. The band log-spectral distance is computed from the features.
  • The separate carriage (a WAI2 companion holding a DRED payload) is specified and registered; the corpus exercises the embedded carriage only.
  • Fabrics: one GPU adapter (Metal on the build host) and macOS arm64 CPUs by hand; CI will add the Linux arm64 runner; no x86_64, Windows or wasm run yet. No forge run recorded.
  • CI fetches the weights and test vectors from the draft's location; if that location moves, the step fails at the fetch, by design (digest-checked).
  • The reference-build comparison in §6 is informative and was run once by hand; CI does not build the reference decoder.
  • No glossary entry for DRED (exempted in GLOSSARY_NO_ENTRY).
  • Nothing was filed or posted: the working-group proposal is a draft in the repository.

🤖 Generated with Claude Code

Closed: landed on main through merge train C (#461, e0ac15f5); this PR's head is an ancestor of main.

## What this is The deep redundancy (DRED) of **draft-ietf-mlcodec-opus-dred-07** decoded exactly. DRED carries, in an Opus packet's padding, a compact neural latent of the recent speech, from which a receiver rebuilds the acoustic features of lost packets. The draft specifies the decoder network in floating point and admits any decoder within tolerances, so two receivers rebuild different features from the same packets. This adds an **integer profile, `dred-int/1`**: the draft's own network and its own weights file (pinned by SHA-256, never committed), lowered to `wai-ops/1` graphs whose features are the same integers on every machine and on a GPU, with a stream-under-loss reconstruction named by a digest a verifier recomputes, and a player's statement that signs it. - **Extension** `wai/extensions/opus-dred.md`: registration (§2), finding DRED and its integers (§3), the profile (§4), a stream under loss (§5), the measurement (§6), receipts and synthesis's lattice (§7), refusals (§8), conformance (§9). - **Registered** (SPEC §5, §7, REGISTRY, export): `wai.audio.opus.dred` (the draft's decoding, `StandardTolerance`) and `wai.audio.opus.dred.int` (the profile, `NeuralIntegerExact`, one-file parameter set, `ops` lowering `dred-int/1`), both companion-only **enhancement layers** over `wai.audio.opus.packet`, both `patent-pool`. - **wai-rs** `dred_exact` module (feature `dred_exact`): RFC 6716 range decoder (and encoder, for the generator); Opus framing and the extension draft's iterator with its repeat mechanism; the payload's integers; the weights file parsed and expanded; the lowering; the driver; a deterministic `f64` reference; the draft's feature comparison; the loss scenario; the corpus generator. Bin `wai_dred_vectors`. - **Corpus** `wai/live-conformance/dred/` (`vectors.json`, README, `fetch.sh`) and its **second implementation** `dred_verify.py` (standard-library Python from the text, sharing no code with wai-rs). - **Tests**: `tests/dred_exact.rs` (corpus regenerates byte for byte; §8 codes; §4 digests; acc64-not-acc32), `tests/dred_exact_measure.rs` (§6's table on the draft's test vectors), `tests/dred_exact_gpu.rs` (the corpus on every GPU adapter, graph by graph against the interpreter). - **CI**: a step in `wai-conformance.yml` (independent-verifiers: fetch, unit, corpus, measurement, Python verifier) and in `wai-ops-gpu.yml` (build in `checks`, run on Metal in `metal`). - **Proposal** `wai/proposals/dred-integer-profile.md`: the profile as an optional integer conformance point for the working group. **Not filed.** - CHANGELOG, `determinism-tiers.md` §5/§6 rows, the research registry's class basis and licence note, the live-conformance index, a glossary exemption. ## Decisions, and why - **Companion (enhancement layer), not refinement.** A refinement corrects what the base decoded; DRED corrects nothing the base decoded, it is a second, coarser encoding read only where the base is missing, and its offsets index one encode's frames. That is SPEC §5's enhancement-layer model (bound to exactly one rendition of a listed base, the base of no second layer), so both strings reuse `AppliedForm::Enhancement` and LCEVC's binding rule. Carried embedded (the draft's own carriage, no component) or separately (a `WAI2` companion whose payload is one DRED payload for frame 0 of the packet in the same envelope). One base, `wai.audio.opus.packet`, because there the envelope aligns DRED with its packet; no alignment with an Ogg stream is pinned, as for LCEVC. - **Two strings.** A capability's class bounds the tier a pin may claim (SPEC §3.1), and the draft's decoding is admitted within tolerances, so it cannot back `replicate`. The exact profile needs its own `NeuralIntegerExact` string. - **`patent-pool`, conservatively.** The IETF's IPR disclosure 6063 states a royalty-free intent conditional on the draft becoming a standard; REGISTRY.md's rule classes an unestablished status `patent-pool`. - **Weights pinned and fetched, never committed.** The standalone weights file carries no licence text of its own; the same arrays ship in the draft's reference implementation under a BSD-style licence whose notice names its copyright holders, which the public-text rule keeps out of the tree. So the corpus names the file by the draft's SHA-256, and `fetch.sh` fetches it from the draft's location and checks the digest. Nothing derived from the weights is committed either (the corpus holds only digests, the drawn packets and the decoded rows' digests and one row per payload). - **The arithmetic.** Weights at `2^24`, rounded half to even from the float32's exact bits (no float in any implementation); biases at the accumulator's scale; activations Q15; `tanh` read at Q12 and the sigmoid at Q11 from `lut16` tables whose entries are a stated integer computation from one 62-bit constant (`⌊e^(−1/2048)·2^62⌋`, squared repeatedly), so a second implementation computes the same tables rather than trusting a float library; every `rq` half up; the GRU update written `hc + z·(h − hc)` so it rounds once; inputs Q16, `x·256/s` rounded half up. Every operator is in `wai-ops/1-i32`; the graphs meet `acc64` (not `acc32`: a 16-bit activation times a 32-bit weight over up to 576 inputs), so the GPU runs them at `acc64`. - **A magnitude cap** (`2^16`, `dred_latent_range`): the draft's escape rule bounds nothing, and past a payload's end the range decoder reads zeros, so a crafted payload could keep a decoder reading. Where the draft's prose leaves a detail open (the magnitude is `1 +` the escape symbols; the quantizer's rounding), the profile follows the draft's reference decoder, and §3.3 says so. - **The vocoder is out of scope.** The draft makes it non-normative and tests it by "sounds similar"; an exact profile of it would be WAI-invented. The profile's output is the feature sequence; §7.2's statement signs that, and says a player's statement of its vocoder's samples is its own. - **Synthesis classes** (§7.1): a reconstruction from DRED is `convert` (a decoder's reconstruction of what the packets carried); concealment beyond DRED's coverage is `generate`/`synthesized` (content the packets never carried, by a trained model; not `assist`, which edits an input that exists); speech coding enhancement and blind bandwidth extension are `assist` (as synthesis classes an upscale), tier `attest`. - **The other mlcodec documents** (§2.3): the extension format and the speech coding enhancement register nothing (no payload); the scalable quality extension rides embedded in the base; the test battery is informative. ## Measured (§6), on the draft's eight test vectors | compared | frames | total | max | pitch | largest difference | band LSD mean (dB) | band LSD max (dB) | |---|---|---|---|---|---|---|---| | integer, reference | 16000 | 0.1314 | 0.0767 | 0.00492 | 0.4139 | 0.2945 | 1.0034 | | float, reference | 16000 | 0.1314 | 0.0767 | 0.00492 | 0.4151 | 0.2944 | 1.0060 | | 8-bit emulation, reference | 16000 | 0.0831 | 0.0458 | 0.00255 | 0.2542 | 0.1844 | 0.6929 | | integer, float | 16000 | 0.0021 | 0.0013 | 0.00011 | 0.0084 | 0.0048 | 0.0210 | | integer, float (packets) | 80452 | 0.0022 | 0.0013 | 0.00004 | 0.0085 | 0.0049 | 0.0208 | The profile passes the draft's feature test (0.5 / 0.15 / 0.02) on every vector. Informative, measured once by hand on the build host (arm64): the draft's reference decoder built in its default (8-bit) configuration lies 0.090–0.096 from its own test vectors, built with float weights 0.131–0.136; the two builds differ by about 0.13 from each other; the profile and the `f64` network lie 0.015–0.018 from the float build. So the test vectors carry the arithmetic of the build that wrote them, and two builds of the reference decoder on one machine disagree; under the profile there is one answer. ## Checks | check | command | result | |---|---|---| | unit tests | `cargo test --locked --no-default-features --features dred_exact --lib dred_exact::` (rb) | 18 passed | | corpus, §8, §4 digests, acc64/acc32 | `WAI_DRED_WEIGHTS=… cargo test --locked --release --no-default-features --features dred_exact --test dred_exact` (rb) | 4 passed | | measurement | `WAI_DRED_WEIGHTS=… WAI_DRED_VECTORS=… cargo test --locked --release … --test dred_exact_measure` (rb) | 1 passed; table above | | GPU (Metal, build host) | `WAI_OPS_GPU_REQUIRE=1 WAI_DRED_WEIGHTS=… cargo test --locked --release --no-default-features --features dred_exact,ops_gpu --test dred_exact_gpu` (rb) | `dred-gpu: key=gpu-metal-0 payloads=3 streams=2 same=true graphs_on_gpu=29 (acc32 0, acc64 29) on_cpu=0 compared=29 differed=0 errors=[]` | | second implementation | `python3 -I live-conformance/dred_verify.py live-conformance/dred <weights>` | 9 payload vectors and 9 cases reach their outcomes; every §8 reason refuses one; all 16 mutated copies fail | | registry and container | `cargo test --locked --no-default-features --lib -- codecs:: container::`; `cargo test --locked --lib` (default) | 103 passed; 193 passed | | integration tests | `cargo test --locked --no-default-features --tests`; `cargo test --locked --tests` (rb) | all binaries pass (`carriage.rs` updated: DRED binds to an Opus packet) | | registry export | `cargo run -q --no-default-features --bin wai -- registry export` then `--check` (rb, `--pull`) | up to date | | clippy | `cargo clippy --locked --no-default-features --features dred_exact,ops_gpu --all-targets`; `cargo clippy --locked --all-targets` (rb) | no notes in touched files (three pre-existing `collapsible_if` notes in `glossary_check.rs` lines this change does not touch) | | dependents | `cargo check --locked --target wasm32-unknown-unknown` (wai-web); `cargo check --locked` (wai-pack) (rb) | build | | shared-text verifiers | `synthesis_verify.py`, `staged_verify.py`, `http_verify.py`, `pin_verify.py`, `pack_verify.py` | all pass | | repo checks | `check-tests-in-ci.py`, `check-workflow-concurrency.sh`, `check-spec-refs.py`, `check-wai-section-refs.py`, `check-no-vanity-metrics.sh`, `test-public-text-checks.sh`, `check-comparative-prose.sh`, `check-orphan-sources.py` | all clean | | public names | `check-public-names.sh` on the 35 changed files; commit messages | clean | | lockfiles | no dependency added; every cargo run above used `--locked` | unchanged | ## Falsifications (committed first, each restored after) - Python verifier: dropping the output-digest check → exit 1 (`float_decode_claimed_exact` accepted); dropping the magnitude cap → exit 1 (`magnitude_past_cap` decodes); `tanh` entries floored instead of rounded → exit 1; the GRU update rounded twice → exit 1 (a payload's rows differ); dropping the loss-pattern check → exit 1. - Rust: one shift of the lowering changed (output `23 → 22`) → `the_corpus_regenerates…` fails ("vectors.json differs"); one digit of a row digest in `vectors.json` → fails; an extra code in §8's table → `the_refusals_table_is_the_modules` fails (and the Python verifier exits 1); one figure of §6's table → the measurement test fails. - The Python verifier and wai-rs agree bit for bit on every row of every payload and stream (the corpus's row digests are wai-rs's; the verifier recomputes them in its own integer arithmetic). ## Not done - **The vocoder** stays outside the profile (draft §2.2.4, §4.2); no segmental SNR, which needs samples. The band log-spectral distance is computed from the features. - **The separate carriage** (a `WAI2` companion holding a DRED payload) is specified and registered; the corpus exercises the embedded carriage only. - **Fabrics**: one GPU adapter (Metal on the build host) and macOS arm64 CPUs by hand; CI will add the Linux arm64 runner; no x86_64, Windows or wasm run yet. No forge run recorded. - **CI fetches** the weights and test vectors from the draft's location; if that location moves, the step fails at the fetch, by design (digest-checked). - **The reference-build comparison** in §6 is informative and was run once by hand; CI does not build the reference decoder. - **No glossary entry** for DRED (exempted in `GLOSSARY_NO_ENTRY`). - Nothing was filed or posted: the working-group proposal is a draft in the repository. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --- **Closed:** landed on main through merge train C (#461, e0ac15f5); this PR's head is an ancestor of main. 
The DRED extension found in an Opus packet's padding (RFC 6716 framing, the
extension draft's iterator with its repeat mechanism), its payload parsed to
integers over RFC 6716's range decoder, the decoder weights file parsed and
expanded from its storage layouts, and the lowering dred-int/1: weights
quantized from their float32 bits, tanh and the sigmoid as lut16 tables of a
fixed integer computation, the decoder as two WHG1 graphs loaded under acc64.
A floating-point reference with a deterministic exponential, the draft's
feature comparison, a stream under a seeded loss pattern, and the corpus
generator. Behind the feature dred_exact; the weights are pinned by digest
and fetched, never committed.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The deep redundancy of draft-ietf-mlcodec-opus-dred-07 registered as two
companion-only enhancement layers over wai.audio.opus.packet: the draft's
decoding within its tolerances, and the integer profile dred-int/1. The
extension text (registration, finding DRED, the profile, a stream under
loss, the measurement, receipts and the synthesis lattice, refusals), SPEC
§5 and §7, the registry export, the research registry's basis and licence
note, the corpus live-conformance/dred/ with its fetch script, its second
implementation dred_verify.py, CI steps, the CHANGELOG, and the proposal
to the working group, drafted and not filed.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
The carriage test bound every companion to a video base; the two DRED
enhancement layers bind to their one base, wai.audio.opus.packet. The GPU
test now decodes the corpus's streams as well as its payloads.
determinism-tiers §5 and §6 record the profile, and the live-conformance
index lists its corpus.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
wai-rs: tests/dred_exact.rs holds §4.8: both graphs load under acc64, are refused under acc32, and use wai-ops/1-i32 only
Some checks failed
wai-pack / wai-pack · dependency licences (pull_request) Has been cancelled
Details
wai-ops-gpu integer operators on GPUs / does this change touch it (pull_request) Has been cancelled
Details
wai-ops-gpu integer operators on GPUs / builds · no adapter is not a pass (linux) (pull_request) Has been cancelled
Details
wai-ops-gpu integer operators on GPUs / Metal GPU and browser WebGPU (macos) (pull_request) Has been cancelled
Details
WAI neural sink gates / does this change touch it (pull_request) Has been cancelled
Details
WAI neural sink gates / model-free refusals of the neural sink (pull_request) Has been cancelled
Details
WAI energy receipts / does this change touch it (pull_request) Has been cancelled
Details
WAI energy receipts / energy receipts · linux (pull_request) Has been cancelled
Details
WAI energy receipts / energy receipts · linux-arm64 (pull_request) Has been cancelled
Details
WAI energy receipts / energy receipts · macos (pull_request) Has been cancelled
Details
WAI conformance / does this change touch it (pull_request) Has been cancelled
Details
WAI conformance / cargo-class corpora · linux (pull_request) Has been cancelled
Details
WAI conformance / cargo-class corpora · linux-arm64 (pull_request) Has been cancelled
Details
WAI conformance / cargo-class corpora · macos (pull_request) Has been cancelled
Details
WAI conformance / quantum core and corpus under wasm32-wasip2 (wasmtime) (pull_request) Has been cancelled
Details
WAI conformance / v2 vocabulary · manifests validate as stated (pull_request) Has been cancelled
Details
WAI conformance / independent verifiers · claims, JWP, key discovery and integer payloads (Python) (pull_request) Has been cancelled
Details
WAI conformance / delivery telemetry · CMCD and CMSD against an independent library (pull_request) Has been cancelled
Details
WAI conformance / cross-arch byte-equality · linux (pull_request) Has been cancelled
Details
WAI conformance / cross-arch byte-equality · linux-arm64 (pull_request) Has been cancelled
Details
WAI conformance / cross-arch byte-equality · macos (pull_request) Has been cancelled
Details
WAI conformance / codec wrappers · fuzz smoke, text decode limits, C entry points, provenance, the semantic path (pull_request) Has been cancelled
Details
WAI conformance / C2PA manifest emission · signed, read back, validated (pull_request) Has been cancelled
Details
standards workspaces build + test / which workspaces this change touches (pull_request) Has been cancelled
Details
standards workspaces build + test / macOS energy backends (pull_request) Has been cancelled
Details
repository checks / lockfiles, tests in CI, workflow hygiene, spec and section references, public text (pull_request) Has been cancelled
Details
joule-moq-web browser client / joule-moq-web · does this change touch it (pull_request) Has been cancelled
Details
joule-moq-web browser client / joule-moq-web · browser end-to-end (macos) (pull_request) Has been cancelled
Details
joule-moq-room room service / joule-moq-room · does this change touch it (pull_request) Has been cancelled
Details
joule-moq-room room service / joule-moq-room, every target (pull_request) Has been cancelled
Details
3ad5e8b0cf 
Co-Authored-By: Claude Opus 5.5 <[email protected]>
dcharlot closed this pull request 2026-10-25 23:51:07 -04:00
Some checks failed
wai-web wasm32 / does this change touch it (pull_request) Has been cancelled
Details
wai-web wasm32 / wai-web wasm32 build · corpus, menu and writers (pull_request) Has been cancelled
Details
wai-web reach builds / wai-web reach · does this change touch it (pull_request) Has been cancelled
Details
wai-web reach builds / wai-web reach builds · the package of record, every output as recorded (pull_request) Has been cancelled
Details
wai-web reach builds / wai-web reach builds · this tree's wai-web, every rung on its engine line (pull_request) Has been cancelled
Details
wai-quantum studio and desktop / does this change touch it (pull_request) Has been cancelled
Details
wai-quantum studio and desktop / studio · browser (pull_request) Has been cancelled
Details
wai-quantum studio and desktop / desktop app · macos (pull_request) Has been cancelled
Details
wai-pack / wai-pack · does this change touch it (pull_request) Has been cancelled
Details
wai-pack / wai-pack · linux (pull_request) Has been cancelled
Details
wai-pack / wai-pack · linux-arm64 (pull_request) Has been cancelled
Details
wai-pack / wai-pack · macos (pull_request) Has been cancelled
Details
wai-pack / wai-pack · independent verifier (Python) (pull_request) Has been cancelled
Details
wai-pack / wai-pack · the floor through a media prober (pull_request) Has been cancelled
Details
wai-pack / wai-pack · dependency licences (pull_request) Has been cancelled
Details
wai-ops-gpu integer operators on GPUs / does this change touch it (pull_request) Has been cancelled
Details
wai-ops-gpu integer operators on GPUs / builds · no adapter is not a pass (linux) (pull_request) Has been cancelled
Details
wai-ops-gpu integer operators on GPUs / Metal GPU and browser WebGPU (macos) (pull_request) Has been cancelled
Details
WAI neural sink gates / does this change touch it (pull_request) Has been cancelled
Details
WAI neural sink gates / model-free refusals of the neural sink (pull_request) Has been cancelled
Details
WAI energy receipts / does this change touch it (pull_request) Has been cancelled
Details
WAI energy receipts / energy receipts · linux (pull_request) Has been cancelled
Details
WAI energy receipts / energy receipts · linux-arm64 (pull_request) Has been cancelled
Details
WAI energy receipts / energy receipts · macos (pull_request) Has been cancelled
Details
WAI conformance / does this change touch it (pull_request) Has been cancelled
Details
WAI conformance / cargo-class corpora · linux (pull_request) Has been cancelled
Details
WAI conformance / cargo-class corpora · linux-arm64 (pull_request) Has been cancelled
Details
WAI conformance / cargo-class corpora · macos (pull_request) Has been cancelled
Details
WAI conformance / quantum core and corpus under wasm32-wasip2 (wasmtime) (pull_request) Has been cancelled
Details
WAI conformance / v2 vocabulary · manifests validate as stated (pull_request) Has been cancelled
Details
WAI conformance / independent verifiers · claims, JWP, key discovery and integer payloads (Python) (pull_request) Has been cancelled
Details
WAI conformance / delivery telemetry · CMCD and CMSD against an independent library (pull_request) Has been cancelled
Details
WAI conformance / cross-arch byte-equality · linux (pull_request) Has been cancelled
Details
WAI conformance / cross-arch byte-equality · linux-arm64 (pull_request) Has been cancelled
Details
WAI conformance / cross-arch byte-equality · macos (pull_request) Has been cancelled
Details
WAI conformance / codec wrappers · fuzz smoke, text decode limits, C entry points, provenance, the semantic path (pull_request) Has been cancelled
Details
WAI conformance / C2PA manifest emission · signed, read back, validated (pull_request) Has been cancelled
Details
standards workspaces build + test / which workspaces this change touches (pull_request) Has been cancelled
Details
standards workspaces build + test / macOS energy backends (pull_request) Has been cancelled
Details
repository checks / lockfiles, tests in CI, workflow hygiene, spec and section references, public text (pull_request) Has been cancelled
Details
joule-moq-web browser client / joule-moq-web · does this change touch it (pull_request) Has been cancelled
Details
joule-moq-web browser client / joule-moq-web · browser end-to-end (macos) (pull_request) Has been cancelled
Details
joule-moq-room room service / joule-moq-room · does this change touch it (pull_request) Has been cancelled
Details
joule-moq-room room service / joule-moq-room, every target (pull_request) Has been cancelled
Details

Pull request closed

Please reopen this pull request to perform a merge.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
Transaction-Science/open-standards!434
No description provided.