[Vilin97]

Two policy changes the maintainer asked for (2026-11-14).

Reviews always decide. There is no third verdict any more.

• Blocking rubrics (faithfulness, novelty, significance, sources) return pass or block. An unreadable or retired verdict fails closed as block.
• Code quality stays advisory: pass or flag (⚠️), never a veto.
• Overall verdict: any block, or an elided diff, gives request_changes; otherwise approve. needs_discussion is gone, from refactor reviews too.
• Each rubric's borderline guidance now says which way to decide: for example, suspected prior art without a named declaration is not a block, and an arguable disclosure is a block naming the honest card wording.
• Unresolved portion-review obligations become block/request_changes instead of discuss.

Stale PRs by age, not inactivity. actions/stale never fired on PRs because the review bots comment on every one. New python/lean_pool/stale_pull_requests.py (run every 6 h by stale.yml):

• labels a PR stale once it has been open 7 days;
• closes it once it has been open 21 days and has carried the label for at least 7 days, so PRs that are already old get a week's warning;
• pinned exempts a PR. Issues keep the existing inactivity policy.

A live dry run labels 17 open PRs now. #506/#507/#512/#639 already carry stale and would close 7 days after their label date.

The permanent reviewer's prompt gains matching rules: never defer, merge ready PRs, oldest first. That ships in the automation repo, along with this PR's merge authorization.

Flaky-on-26.04 test fix. test_watch_stops_a_lean_process_past_the_limit faked lean as a symlink to sleep. The ubuntu-26.04 runners ship multicall uutils coreutils, so that binary invoked as lean exits immediately. The test failed Python CI on every non-content PR, and it is what dequeued #758. It now fakes lean with Python, like the neighbouring tests.

Testing: ruff check clean. Updated review/portion tests and new stale-policy tests pass. The full suite passes except test_repository_notice_is_current (stale NOTICE, fixed by #758) and three test_proof_profile cases that fail identically on main.

🤖 Generated with Claude Code

[github-actions]

Proof profile (new / modified Lean files)

No added or modified Lean files under LeanPool/ in this PR. No compile-cost measurements are needed.

Advisory only — never blocks merge.

[greptile-apps]

Advisory cross-file review against Lean Pool's supervisory rubric. This is not by itself a significance, source-verification, or merge verdict.

[Medium impact] The PR appears safe to merge; no new actionable issues were found.

Summary

This PR replaces deferred review verdicts with clear decisions and introduces age-based PR closure.

• Reviews now decide whether a change passes or needs changes.
• Open pull requests get a stale warning before age-based closure.
• The timeout test keeps its fake Lean process running on multicall runners.
• The project guide links readers to source profiles and refreshes its pool details.

_{Reviews (3) · Last reviewed commit: "Preserve stale PR warning periods and is..." · Reviewed by Greptile}

[Vilin97]

Approve — reviewed exact head d8a014094a3808a65bdb23a4bc45e3eb348797b8.

The review-tool changes preserve fail-closed core verdicts, advisory quality flags and blocking unresolved integration obligations. The age policy retains the intended 7-day warning, 21-day age threshold, minimum 7 days labelled, and pinned exemption.

I fixed the remaining policy bugs in the three-file commit:

• Missing label history no longer counts as a completed warning period; a regression covers it.
• The issue-only stale action explicitly leaves PR stale labels intact. Its pinned implementation otherwise removes labels on activity before checking the disabled PR-close setting, resetting the new age policy's warning clock.
• The checkout job receives explicit read permission while retaining the needed issue/PR permissions. The issue job's 90/30-day policy is unchanged.

The author's newer reopening advice is preserved. The fake-Lean watchdog test retains its timeout assertions and passed against the unchanged current-main watchdog.

Validation: six policy tests, pagination/dry-run boundary checks, the actual watchdog fixture, Ruff lint/format, and twelve independently assessed review/portion cases passed. This is focused validation; the new head's required hosted checks must still pass before protected queue admission.