[JacobStephens2]

Summary

Releases wait for an Experiment against the last release tag on touched Eval suites.

release content commit (pinned across interruption)
  → Experiment --release → touched suites vs last tag
      pass                → continue
      pending rechecks    → list Scores and poll the same Experiment
      fail / inconclusive → show intervals and ask to release anyway
      no suite / old tag without Trial flag → note and continue

A pushed bump branch is checked again before opening its PR. Update the manual and hand-written site pages. Wait for fonts before browser layout assertions to avoid measuring a fallback typeface.

Closes #1022

Evidence

• Before: release mode was rejected as an unknown flag. The Spec review's exact interrupted_release_before_push_keeps_its_experiment_when_main_moves regression failed with two different Candidates.
  After: the release CLI tests pass, including real SIGINT/resume. The exact review regression passes with the original Candidate retained.
• Fake-cargo release tests cover pass, fail/inconclusive overrides, pending rechecks, resume, and skip notes. The compact-help regression and both focused 320px layout checks pass. Clippy with warnings denied and all 14 Python tooling tests pass.
• /usr/bin/time -p scripts/affected-tests 42bad19bb424fd31111add2ed36eb05034a5eb74 passed: 2,563 Rust tests (58 skipped), 44 browser tests, and the selected release-generator Python check. Total wall time: 1,388.57 seconds, including 11.84 seconds of Rust compilation; Nextest execution: 1,318.191 seconds.
• CI passed on the exact pushed head cb172f4c.

Test seams

• Experiment CLI: release invocation and public saved Experiment listing/views, using real Git repositories and cached Arm binaries.
• Release script: end-to-end release fixture with fake cargo at the external command seam.
• Existing browser layout assertions against the rendered pages.

Review

Independent Standards and Spec reviews against issue-977 at 42bad19b: both Standards findings and the Spec finding are addressed. The Spec finding's supplied test was run verbatim, failed before the fix, and passes afterward.

Both axes read all 16 implementation/review-fix files, including generated site/docs/index.html; no generated output was excluded. A subsequent fixture change, site/tests/pages.test.cjs, was left unread by both axes and is disclosed in both saved reports. It waits for fonts before the existing layout assertions; the focused tests pass.

Unaddressed findings

• Standards: None.
• Spec: None.
• Security: None (guidance-mode review of the release Experiment change against 42bad19b; no introduced finding confirmed; the 6 fake-cargo release-gate tests pass).

Security review outcome

No unaddressed introduced findings.

Merge Danger

Door: two-way

Blast Radius: Releases

Adds a local release-candidate Git ref until the branch push. Fail or inconclusive Verdicts require explicit permission; EOF stops before publishing more release artifacts.

Built with codex · gpt-6.1-sol · xhigh