§6 (dated 2026-09-27) said there was no Windows leg. Two are back: the golden-byte suite in wai-byte-equality.yml's windows-x86_64 leg since5f1158154(2026-11-05), and the cargo-class job of wai-conformance.yml since6ca202b8a(2026-11-10). The section now names a passing run of each, read from the forge on 2026-11-10: run 28636 (job 115713), where every golden of the suite passed on Windows, and run 28810 (job 116278), the cargo-class leg. It says what does not run there: the conformance workflow's own byte-equality job keeps its unix legs, its other jobs have no Windows leg, and within the cargo-class job the steps that need a unix shell or tool (the JPEG AI math-library symbol check, each integer feature alone, the wasm32 builds, the audio provenance check, the OER C kernel) are skipped there. No macOS x86_64 or Windows arm64 leg. The rows that said Windows had not decoded something now name the run where it has: the integer video and keyframe goldens, the int_codec audio and splat goldens and int_mlicpp's golden (run 28636), and float_repro's pinned-bits tests and tests/int_mlicpp.rs in both profiles (run 28810, where the host-exp comparison reports itself ignored, as its gate says it does on Windows). Co-Authored-By: Claude Opus 5.5 <[email protected]>
626 B
626 B
What the determinism tiers' evidence runs on Windows
extensions/determinism-tiers.md§6 no longer says there is no Windows leg. Windows x86_64 runs the golden-byte suite inwai-byte-equality.yml, back since 2026-11-05, and the cargo-class job ofwai-conformance.yml, back since 2026-11-10; the section names a passing run of each, and what does not run there: the conformance workflow's own byte-equality job and its other jobs, and the cargo-class steps that need a unix shell or tool. The rows on the golden-byte suite,float_repro,int_mlicppand the integer codecs name their Windows runs.