A wai.neural.int_synth payload arrived: int8 latents + a model spec. This tab decodes it
to RGB with the same pure-i64 Rust that decodes natively, compiled to wasm32 — no
onnxruntime-web, no WebGPU, no float anywhere. Because the decode is integer all the way down
(the integer-entropy contract), the reconstruction is byte-identical on every machine, so the page
re-checks its BLAKE3 against the hash the native reference froze into
neural-conformance/expected_hashes.json — a learned-codec image verified client-side,
no server round-trip.
Why integer synthesis: a float learned decode can round differently on another platform, and a
drift of ~1e-5 in the entropy scale desyncs it, so its decode is reproducible only on the same build —
its reconstruction cannot go in a receipt. int_synth is integer all the way down, so
the hash above is portable: this browser, a native sink, and the conformance corpus all produce the
same bytes. The wire carried latents + weights; your device produced the exact image, and proved it.