[edbfi]

Finishes S12 (core rollout, Native tier): homebrew-taps, the last repo in the tier, is rolled out. EasyHDR and github-stats were done in #36.

What changed

• STATE.md:
  • S12 done;
  • S12a (the docs(state): finish S7, the obzorarr core pilot #28 rebuild) moves from blocked: S12 to pending;
  • the homebrew-taps row is filled in;
  • "Allow GitHub Actions to create and approve pull requests is on in homebrew-taps" is deleted, because it's off now.
• DECISIONS.md: B9 removed (resolved).
• design/repos.md: the stale-CI-reference list is removed with its Verification check, its "Why" bullet and the clause in rule 6. Both remaining entries are resolved:
  • homebrew-taps' cask headers, scripts/lib/common.sh, scripts/discover.sh and .gitignore now describe the tap pipeline (docs(state): finish S19, the homebrew-taps updater and the #3 reconciliation #40);
  • github-stats' README says no workflow regenerates the SVGs yet (github-stats#31). S17 changes that line again.
  • No repo still to roll out (dox, repo-patches) had an entry.

Evidence (homebrew-taps, 2026-10-09)

• B9 fixed in homebrew-taps#40 7c8d232: weston composites with the GL renderer (llvmpipe), not pixman.
  • The diagnosis reproduced the crash only in the full harness: 1 of 3 runs, with the backtrace in pixman_image_composite32 called from libweston-13 during repaint.
  • Validation 37946940536: x86 GL 30/30 and pixman control 9/10; ARM GL 10/10. The renderer was confirmed in weston's logs every time.
  • Details and the rejected alternatives are in homebrew-taps#40's body.
• docs(state): finish S19, the homebrew-taps updater and the #3 reconciliation #40 merged as d92563f, by hand with --match-head-commit (auto-merge.md rule 12): before the merge, the repo had no required checks and auto-merge was off.
  • PR run: both architectures built all four formulae, and all 10 X11 and Wayland checks passed.
  • Push CI 37960596468: green.
  • Pullfrog: no new issues; its two earlier nits were answered (REJECTED with design-rule reasons).
• Settings, matching EasyHDR (gh api read-back):
  • SHA pinning required (every uses: SHA-pinned first);
  • GITHUB_TOKEN read, and "Actions can approve PRs" off;
  • squash only, with title PR_TITLE, message COMMIT_MESSAGES, delete-on-merge and sign-off on;
  • Dependabot alerts and security updates, secret scanning and push protection on;
  • auto-merge on.
• Ruleset main (24803834): active; diff against EasyHDR's is identical: ci-ok and pr-policy required, not strict, admin role the only bypass. The D16 deploy-key actor waits for the owner's key (settings-A1, S19), as the deploy-key line in GitHub state says.
• Repo auto-merge was turned on before homebrew-taps#42 (dependabot-auto-merge.yml, byte-identical to the template), so docs(settings): drop homebrew-taps' DeployKey bypass actor and tighten the related rules #42 auto-merged through the ruleset: bdab08b, rule suite pass. Push CI green.
• First Dependabot merge: docs(state): record the updater's live retry check and correct the deploy-key fact #41 (pullfrog 0.1.95 → 0.1.98) was opened before the workflow existed. After @dependabot rebase, the workflow enabled auto-merge and docs(state): record the updater's live retry check and correct the deploy-key fact #41 merged as 8de8a8a (rule suite pass). Push CI green.
• Immortality: a dispatched run passed (37965637958).
• ci-A2: ubuntu-latest is still Ubuntu 24.04 (runner image log), so ubuntu-24.04-arm stays; it remains open for S19 and the 26.04 move.
• Guards: prek run --all-files --hook-stage manual: all pass.

Deviations

• None. The homebrew-taps ruleset is admin-only until D16, as recorded.

Owner actions

• Create homebrew-taps' write deploy key (D16) for S19.
• Grouped security updates in homebrew-taps (Settings → Advanced Security; there's no API), as in every rolled-out repo.

Review

• Pullfrog on 9032848: one nit, that ci-A2 still listed S12 as a re-check point. APPROVED in 5180919: it now names S19 and the 26.04 move.

[pullfrog]

ℹ️ No critical issues — one stale-wording nit outside the diff.

Reviewed changes

Reviewed the full single-commit diff (STATE.md, DECISIONS.md, design/repos.md) against main; python3 tools/doc_guard/doc_guard.py passes.

• S12 done, S12a unblocked — STATE.md flips S12 to done with the homebrew-taps PRs added, and S12a to pending; Last updated matches its ISO stamp.
• homebrew-taps row filled in — ruleset column records the admin-only bypass until D16, which matches the deploy-key line in GitHub state and design/settings.md rule 4.
• Drift line removed — the "Actions can approve PRs is on in homebrew-taps" line is gone; the "not yet rolled out" line still carries the right remaining exceptions (dox).
• B9 retired — removed from DECISIONS.md; no other reference to B9 remains in the repo.
• Stale-CI-reference list removed — design/repos.md drops the list, its Verification check, its Why bullet and the rule 6 clause together; no dangling references.

ℹ️ design/ci.md ci-A2 still names S12 as an upcoming re-check

design/ci.md:67 says "re-check the runner-images README when homebrew-taps rolls out (S12, S19)". S12 is now done and the PR body records that re-check, so only S19 is left. Not in this PR's diff, so there's no inline anchor.

Technical details

# ci-A2 names a finished step

## Affected sites
- design/ci.md:67 — "(S12, S19)" lists S12 as a pending re-check point.

## Required outcome
- ci-A2 names only the remaining closing points (S19 and the 26.04 move).

  ^{｜ Fix it ➔ ｜ View workflow run ｜ Using claude-sonnet-5-5 ｜ 𝕏}

[edbfi]

Pullfrog review 1 (ci-A2 names a finished step): APPROVED in 5180919. ci-A2 now says "re-check the runner-images README at S19 and at the 26.04 move".

[pullfrog]

✅ No new issues found.

Reviewed changes

Reviewed the incremental delta since the prior pullfrog review (9032848 → 5180919) and re-read the full diff; python3 tools/doc_guard/doc_guard.py passes.

• Dropped the finished S12 re-check from ci-A2 — design/ci.md:67 now closes at S19 and the 26.04 move, which resolves the nit from the prior review. The [V, 2026-10-09] tag and the 2026-10-19 to 2026-11-19 window are unchanged.
• Bumped Last updated — STATE.md line 1 reads 1791567322 (2026-10-09T17:35:22Z), and the two values agree.

  ^{｜ Fix it ➔ ｜ View workflow run ｜ Using claude-sonnet-5-5 ｜ 𝕏}