[renmengye]

On a freshly seeded repository (the quickstart, for one) every row of BENCHMARKS.md said "provenance unknown" in the main-commit column, which reads like a fault on the first page a new user sees. A row with no main commit whose best is still its baseline now shows —, matching its progress column. Imported and self-reported rows are unchanged.

Seen in a live local-forge run of the quickstart on 0.5.0. One test line added; full gate passes.

Written by Claude (Opus 5.5); reviewed by codex (no findings).

[github-actions]

Round 1 — reviewed head fc213351 — reviewer hermes/gpt-5.6-terra.

terra
Advisory findings from outerloop — the code owner decides. Reply to disagree; the outerloop:no-review label opts this PR out.

Verdict: no defects found.

[github-actions]

Round 2 (re-run on the same head) — reviewed head fc213351 — reviewer summarizer:hermes/gpt-5.6-terra over coverage+credentials+deployment+general+lifecycle+prose.

terra
Advisory findings from outerloop — the code owner decides. Reply to disagree; the outerloop:no-review label opts this PR out.

Verdict: 1 blocking, 2 advisory.

3 findings attached to the lines below.

Two substantive display regressions: imported baseline rows lose the required provenance-unknown label (blocking), and self-reported baseline rows are changed despite their stated exception. Rejected: none; the general and lifecycle imported-row findings were merged as the same claim, while the prose rewrite is retained separately as required.

[github-actions]

Round 3 — reviewed head bbb01b0c — reviewer hermes/gpt-5.6-terra.

terra
Advisory findings from outerloop — the code owner decides. Reply to disagree; the outerloop:no-review label opts this PR out.

Verdict: no defects found.