| capture-diff-coverage.txt | ||
| foundation-diff-coverage.txt | ||
| r3_t5_evidence_sanitized.zip | ||
| README.md | ||
T5 round-three review evidence
Sanitized independent Devin evidence for foundation #49190 and capture #49133
Archive SHA-256 eeddf7eaa98b01205355e9b32d0b8ef439dc76d6e6e64dbf34695de31e675799
Baseline ccc56d27adeeece2a1069bde587c76486ba9ab39, foundation b368c432d03de80d5b6efc2dd6d6389c4d9c9b11, combined capture a2fbb6d260a5a25076542e836d9eb398648e51c5
The archive includes the acceptance report, provenance, live measurements/logs, reproduction scripts/configuration, and complete per-PR coverage JSON/XML. Start with README.txt and artifacts/R3_T5_ACCEPTANCE_REPORT.md inside it. Historical round-two carry-forward evidence retains its tested SHA attribution
Changed-line coverage with real PostgreSQL integration is 204/206 (99.03%) for foundation and 491/508 (96.65%) for capture. The report explains remaining lines and baseline-reproduced unrelated test failures
Live Vertex/GCS lifecycle was not run due to unavailable bucket access. The author accepted this explicit verification limitation for T5. Existing Vertex unit coverage and the local OpenAI route reproduction do not prove live GCS behavior
Ephemeral test credentials and database credentials are redacted, with a redaction manifest. Process-ID files and compiled Python caches are excluded. No measurement or coverage result is changed. No application code or workflow lives on this evidence-only branch. No scripts were executed during packaging