Read the scoped findings and commands. This archive retains two failed Windows runs and their exact source inputs. baseline-1: 7 pass / 4 fail (three preservation defects plus one invalid fixture). baseline-2: 8 pass / 3 fail (corrected controls; the same preservation defects). No production correction, whole-suite pass, Linux or GUI acceptance is implied.

Each gate directory contains the raw stdout/stderr, start/final receipts, 605-input JSON manifest and ZIP. gate.py is the exact original capture runner; its absolute paths describe this host and need adaptation for a new capture. Use the Cargo command in the findings report to reproduce the test directly. The input ZIP identifies uncommitted-at-execution test bytes, not just its HEAD. The independent reviews retain their scope and distinguish setup errors.

From this directory, run python -B verify.py. It checks archive membership and SHA-256/length pins, both source ZIPs, receipt/log consistency, compiler artifact records, expected named failures and counts, and that only the test changed between captures. The verifier is read-only and requires only Python 3. files.json pins every other file, including this README and the verifier. Git identifies the manifest itself. This verifies record integrity rather than rerunning tests. Test executables are not included; their hashes are historical runner receipts. The default ignored regressions must be explicitly run with --include-ignored to observe the failures; remove ignores with the actual fix.