wai-rs: prove the reconstruction binding over a real jpegai decode #53

Merged
dcharlot merged 1 commit from wai-binding-real-decode into main 2026-08-25 17:10:00 -04:00

2026-08-01

Transaction Science
wai-rs: prove the reconstruction binding over a REAL jpegai decode
All checks were successful
WAI conformance / cargo-class corpora · linux (pull_request) Successful in 32s
WAI conformance / cargo-class corpora · linux-arm64 (pull_request) Successful in 41s
WAI conformance / cross-arch byte-equality · linux (pull_request) Successful in 15s
WAI conformance / cross-arch byte-equality · linux-arm64 (pull_request) Successful in 20s
WAI conformance / cargo-class corpora · windows (pull_request) Successful in 1m11s
WAI conformance / cross-arch byte-equality · windows (pull_request) Successful in 20s
WAI conformance / cargo-class corpora · macos (pull_request) Successful in 2m10s
WAI conformance / cross-arch byte-equality · macos (pull_request) Successful in 2m1s
The reconstruction_binding unit tests bind over synthetic RGB. This closes the loop
end to end: the byte-exact decoder decodes test256 to pixels, and the C2PA/JPEG-Trust
hard binding is taken over those pixels — sha256 over the canonical RGB8, plus the
descriptor pinning the reproducible decode.

It exercises the property the capability rests on and that neural-coded media could
not have before a byte-exact decode: the decode is deterministic
(decode-equivalence), so the binding taken over one decode verifies a second decode
byte-for-byte, while a single flipped pixel breaks it. Emits the actual assertion
(label science.transaction.wai.hash.reconstruction) over the real decoded image.

Also the first test that compiles `neural_int` + `provenance` together. Gated on the
model bundle; skips cleanly without it.

Co-Authored-By: Claude Opus 4.8 <[email protected]>
2026-08-01 23:23:47 -04:00