wai: ingest the determinism-tier spectrum (entropy-consistency vs decode-equivalence) #52
2026-08-01
wai: ingest the determinism-tier spectrum for neural decode
All checks were successful
no-vanity-metrics / Scan committed Markdown for vanity metrics (pull_request) Successful in 4s
The SOTA sweep surfaced two facts that line up: JPEG AI's normative conformance point is the entropy-decoder output (not pixels), and the mainstream learned-codec cross-platform fix is explicit parameter transmission — both guarantee a reproducible ENTROPY stage while leaving reconstruction implementation-defined. That is a real, named guarantee the field converges on; WAI supports it AND a stronger one, but the standard didn't name the spectrum. New extensions/determinism-tiers.md names it: - entropy-consistency — the entropy decode reproduces across devices (integer entropy arithmetic or explicit parameter transmission); the base learned-codec floor and the minimum for a sound neural replicate. - decode-equivalence — the whole decode to the final medium reproduces, so the rendered result has a portable hash; the tier WAI adds on top, and the precondition for a hard reconstruction binding over decoded pixels. A capability satisfying decode-equivalence satisfies entropy-consistency by construction; the reverse doesn't hold. Referenced from SPEC §7 (Replicate- soundness) — the ReplicateClass values already span these tiers, this names them — and cross-linked from the reconstruction-binding extension (which requires decode-equivalence). Changelog entry added. Framed self-referentially: WAI spans the spectrum, entropy-consistency is the field's floor, decode-equivalence is the addition — no comparisons. Co-Authored-By: Claude Opus 4.8 <[email protected]>