register what other bodies ratified: prior carriage, and derived-track operations #90

Merged

dcharlot merged 1 commit from wai/prior-carriage-and-derivations into main 2026-09-10 12:23:27 -04:00

Conversation 0 Commits 1 Files changed 2 +130

2026-09-10

Transaction Science

register what other bodies ratified: prior carriage, and derived-track operations ...

Two ingests of the same shape — name someone else's ratified work rather than
restate it, and add only the thing they left unspecified.

PRIOR CARRIAGE. WAI pins a prior's IDENTITY and says nothing about its FORM. The
digest fixes which bytes; nothing fixes how those bytes are packaged, so a prior
is in practice an informal directory — a layout convention between
implementations that happen to agree. Two conforming implementations can hold the
same digest over differently-arranged artifacts and neither is wrong, because the
specification never said what the artifact is.

That bites precisely where WAI is strongest: a determinism tier is a claim about
reproducing a computation, and a computation whose operand has no wire form is
reproducible only among parties already sharing an unwritten convention.

ISO/IEC 15938-17 is that form — the first international standard for neural
network compression, second edition 2024, covering trained parameters and their
incremental updates, with an open reference implementation. Where a prior is
carried in it the digest MUST cover the coded bitstream rather than an unpacked
directory (a digest over an unpacking is a digest over one implementation's
choices), and two priors MUST NOT be treated as interchangeable because their
unpacked parameters compare equal.

Scope discipline is explicit, because the survey's one unresolved flag sits here:
this is CARRIAGE, NOT AUTHENTICITY. Work on signing individual coded data units
within that standard is at amendment stage and could not be corroborated from
public sources, so nothing here depends on it and an implementation MUST NOT cite
this section as authority for a signing claim. Incremental update is likewise not
new — it landed in the published second edition.

DERIVED TRACKS. Stated plainly at the top because the tree needs it said: a track
carrying a process the receiver evaluates is NOT NEW. ISO/IEC 23001-16:2021
ratified derived visual tracks five years ago inside the container the industry
already ships. Any WAI text claiming instruction-carriage as its invention is
wrong and should be corrected on sight.

What that specification leaves unspecified is what WAI adds: a determinism
contract per operation, and a signed receipt binding the executed graph to
measured joules. So each operation MAY register as a capability and MUST state
its tier; an operation involving interpolation, blending or resampling MUST NOT
register at decode-equivalence unless the registration also pins the arithmetic —
naming an operation does not make it deterministic — and a derived sample's tier
is the WEAKEST among the operations it invokes, because a graph is as
reproducible as its least reproducible step.

Two qualifications kept in the text rather than dropped: the base operation set
includes blending operations with no bit-exactness requirement, so the tier work
is substantive; and derived visual tracks have negligible deployed player
support, so the collision is standards-level, not installed-base.

Neither extension makes WAI a renderer, and §3 of derived-tracks says so.

Ingest queue items 9 and 12 (#80 §6.4, §3.1).

Co-Authored-By: Claude Opus 5 <[email protected]>

93c5542b7a

• View at this point in history

2026-09-10 10:24:58 -04:00