ARCADIA-Studio/harness/reviews/prerelease37-final04-tests-20260924.md

Final-04 workspace-test evidence review

2026-09-24. Read-only admission review of 28b8b5aad7d58910d09d0e183a3832fd27fe1c44:harness/prerelease37-windows-gates-20260924/final04-tests, source 715036e132677dd2bc2fb37ed316d93dc9987aec.

Result

No material discrepancy found. Independent local reconciliation of the captured inventory and literal stdout confirms 61 native harnesses, 2,208 names, 2,200 passed, eight ignored, zero failed, with no additional unlisted result rows. All 29 original required names are present once in their stated targets, nonignored, and individually recorded ok in actual stdout. Their cited source SHA-256/Git-blob identities match the captured source map. This includes the corrected State picker case that failed final-03; the earlier failure remains historical failure evidence.

The locally reviewed reconcile_native_tests.py ran unchanged, with assertions enabled, against this ZIP and exact source. Its success checks passed, including command exit0, no instrumentation failure, unchanged inputs, literal workspace-test command, complete name multiplicities, expected ignores and successful raw summaries. Output is final04-native-reconciliation.stdout.json; stderr is empty. This reconciler establishes aggregate native-name/status equality, not independent per-binary execution authentication; the separately published target-by-target audit is additional evidence, not a substitute for this check.

The archive has 64 actual result summaries: 61 native harnesses plus three empty doctest harnesses. Ignored cases remain ignored, not passed. They include four desktop timing/scaling probes, two source-annotation anchor cases, annotation performance and inheritance-chain scaling; their rationale is outside this result-count check and remains recorded in the source-bound audit.

Identity and raw gate checks

Read raw rustc version, Cargo version, workspace no-run compilation and literal workspace-test receipts. All use the exact715036e source, exit0, failure:null and unchanged inputs. Actual commands include:

• cargo test --locked --offline --workspace --no-run --message-format=json-render-diagnostics
• cargo test --locked --offline --workspace --no-fail-fast -- --test-threads=1

All phase source maps are byte-identical to the top-level before map, including the listing after map; each contains 3,709 entries. SHA-256: 38e4f7d37b67b06b7991d48560c5bd56ec5893909f3cffd224442d15bdc35f88. The listing phase records 122 invocations, passed:true and unchanged source/environment. Independently compared all 61 admitted, listed and post-test artifact records: path/length/hash identities agree. This compares captured binary receipts, not current remote executables.

The archived and receipt-bound runner matches reviewed v2: run.py SHA-256 698ed7182726c77fa4a1c85fc6dd08187ed20689941f57e749e3f0bc845f3252; final.py ce73a1680f8cd8bb5112cffa81929458dafaaefc282b8ff5391c7571cd34e666. Its phase-boundary ownership limitations remain as documented in RUNNER-V2-FAILURE-REVIEW.md.

Transport and review evidence

Read manifest/README/directory using Gitea MCP before downloading the exact named ZIP with a 3,500,000-byte cap. MCP manifest Git blob is a60b2d860d4bc08d3f86972227615da399e58370. final04-tests-715036e.zip received 3,415,179 bytes, SHA-256 04ed3c594884844b1f0fed2ba9ccd6842065f3c2a3d4ffca3cfb37971b78b180. Independently verified all 415 archive member names/counts, lengths and SHA-256 pins, rejecting duplicate names.

Read required-inventory and test-reconciliation audit data and compared the 29 required names against the original required map, final native inventory, source-map identities and literal stdout. Local reproducible inspector is inspect_final04_tests.py; it hashes/reads data and invokes only the local reconciler. No archived code or executable was run.

Scope limits

This checkpoint contains completed compiler/listing/test phases only. It deliberately excludes strict workspace Clippy, desktop/server product builds, package/runtime/notices verification, native GUI and server persistence acceptance. Later comments reporting full-gate success and product hashes were not admitted from this ZIP; the portable full-gate evidence must independently bind those claims and bytes. Do not infer full attempt acceptance from this report.

Host clock jump is explicitly retained in the manifest. No elapsed duration is inferred from receipt timestamps. Integrity/internal consistency is not an authenticated build or independent rerun; full Git-blob source comparison was not repeated here. No candidate writes, Cargo execution, Gitea mutation or product launch occurred.