GitHub practice: document the --match-head-commit merge guard; re-require sota-sources

[seathatflowsinourveins]

Scope

• What this PR changes: docs/lanes.md now documents the merge guard, gh pr merge <N> --squash --match-head-commit <SHA>. The guard was already practised but not written down. The live ruleset now has its committed sota-sources requirement back: the PUT ran at 2026-09-27T21:00:13Z, and this PR records the dated GETs before and after.
• Base commit: 55fc8d17 (the parent of this branch's first commit). Corrected 2026-09-29: the earlier text named 0f76651d, which is not the base.
• Lane: lane:shared. Corrected 2026-09-29: the earlier text said lane:foundation, but this PR edits docs/lanes.md, and docs/lanes.md L30 makes any change to that file shared. Under L149-150 the trading lane's acknowledgement is required, on this PR at the head SHA it acknowledges, before merge. The PR label is already lane:shared.
• Owned paths touched: docs/lanes.md, docs/decisions/2026-09-22-github-automation-closure.md (the drift-closed entry) and the hash re-registration in manifests/evidence.json.

SOTA sources

• GitHub CLI manual: gh pr merge. --match-head-commit is the "Commit SHA that the pull request head must match to allow merge".
• GitHub docs: managing a merge queue. Merge queues are "available in any public repository owned by an organization". This repository is User-owned.
• GitHub docs: available rules for rulesets, for required status checks and review-thread resolution.
• docs/decisions/2026-09-22-github-automation-closure.md: strict mode is off (L553-563). The sota-sources drift is recorded as open (L1543-1550). .github/main-ruleset.json L34-43 is the committed target.

Evidence-class table

Claim	Evidence class	Command / receipt
The merge command was not written down anywhere	local_integration	grep -rn 'match-head-commit' --include=*.md docs recipes adoption examples AGENTS.md returned nothing (2026-09-27)
Live ruleset 23739774 required 7 contexts, not the committed 8; sota-sources was missing	historical, recorded by the PR author, output not retained	gh api repos/seathatflowsinourveins/native-agent-stack/rulesets/23739774 (2026-09-27)
Each of the last 60 merged PRs (#344-#424) had a successful sota-sources run; 4 of them also showed an earlier failed or cancelled run	historical, recorded by the PR author, output not retained; not carried into the decision-record entry	gh pr list --state merged --limit 60 --json number,statusCheckRollup (2026-09-27)
Before-GET 21:00:04Z: 7 contexts. PUT at 21:00:13Z. After-GET: 8 contexts matching the committed file	historical, recorded by the PR author, output not retained	gh api --method PUT .../rulesets/23739774 --input .github/main-ruleset.json; before and after gh api .../rulesets/23739774
The live ruleset equals .github/main-ruleset.json at main b0fb65b4 in name, target, enforcement, bypass actors, conditions and every rule; the 8 response-metadata keys and the rule order are normalized away, and the negative control at 57fb6d89 differs only by the sota-sources context	measured 2026-09-29T03:18:04Z (read-only API and local git)	the diff <(gh api ...) <(git show ...) command in the decision-record entry: exit 0, and exit 1 for the control
At the PUT #410 failed sota-sources, #205 and #216 had no run (their last runs predate the job), and #415 and #417 passed	historical, recorded by the PR author, output not retained	gh pr checks <N> (2026-09-27 21:00Z); on 2026-09-29 #205 and #216 are still CONFLICTING with no sota-sources run
The documented merge-guard read: gh pr checks <N> --required lists the required checks with the latest run of each; exit 0 all passed, 1 one failed, 8 pending	measured 2026-09-29 (read-only)	gh pr checks 389 --required exit 0 (8 checks pass); #436 exit 1; #409 exit 8; gh pr view 389 --json statusCheckRollup shows the earlier sota-sources failure that --required hides

Local commands run

$ python3 scripts/validate.py
{"components": 69, "hashed_files": 7348, "profiles": 4, "receipts": 159, "status": "passed"}

Status

Ready for review. The ruleset PUT ran at 21:00Z on 2026-09-27 as announced to the owners of the open PRs failing sota-sources (messaged at about 19:58Z); its before-GET and after-GET were not retained. This PR's record of the drift closure, its rollback (git show 57fb6d89:.github/main-ruleset.json | gh api --method PUT .../rulesets/23739774 --input -, not run) and the 2026-09-29 read-only re-derivation are in the decision record.

Review round (2026-09-29, native-agent-stack-10)

Head reviewed: 35b7ed74, by two model families, read-only. GPT-6 (gpt-6-astra, effort max, codex exec -s read-only; the model is the -m request, not an observed resolution): needs_changes (1 major, 2 minor). Claude evidence-reviewer (Opus, max): needs_changes (2 major, 4 minor, 2 nit). Repairs, with an independent Opus stack-verifier that re-ran the checks: the merge-guard procedure now defines <SHA> as the head on which review and required checks completed, reads it with gh pr checks --required (after the verifier showed the first draft's statusCheckRollup filter prints superseded failed runs), and separates head protection from base freshness; the rollback is an executable command; claims without retained output are labelled historical and replaced by the measured re-derivation; the #205/#216 sentence no longer says a rebase is required. This body's Lane line and base commit were corrected. Not fixed: the hot-file edit sits in an earlier commit than the last one in this branch's history (moot under a squash merge).

🤖 Generated with Claude Code

[seathatflowsinourveins]

Relabelled lane:shared: this PR edits docs/lanes.md, and docs/lanes.md ("A change to this file is lane:shared") makes that a shared change. The added section documents gh pr merge <N> --squash --match-head-commit <SHA> for both lanes, and changes no path assignment. Holding the merge for the trading lane's acknowledgement (a comment or review here). CI is green on head 1525c101.

[seathatflowsinourveins]

Review record, 2026-09-29 (native-agent-stack-10). Merge candidate: head 9f0ca004, which is 35b7ed74 plus the repairs below and a merge of main at c26800f3. The PR is lane:shared; it waits for the trading lane's acknowledgement at this head (docs/lanes.md L149-150).

• Two model families reviewed 35b7ed74, read-only: GPT-6 (gpt-6-astra, effort max, codex exec -s read-only; the model is the -m request, not an observed resolution): needs_changes (1 major, 2 minor). Claude evidence-reviewer (Opus, max): needs_changes (2 major, 4 minor, 2 nit).
• Repair: the merge-guard procedure now defines <SHA> as the head on which the review and the required checks completed, and separates head protection from base freshness; claims without retained output are labelled historical and replaced by a read-only re-derivation measured on 2026-09-29 (the live ruleset equals .github/main-ruleset.json at main; the negative control at 57fb6d89 differs only by sota-sources); the rollback is an executable command; the #205/#216 sentence no longer says a rebase is required; the body's Lane line and base commit were corrected.
• What the independent verifier changed: an Opus stack-verifier showed that the first repair's gh pr view --json statusCheckRollup filter prints superseded runs, so "none failed" could not be judged with it (merged PR-D: Codex worker lane: Codex's own config writer, RTK text inline with exceptions, max-effort stack-worker profile #389, Landscape sweep: fail closed when the worker profile overlays an MCP server the lane config lacks #411 and PR-H (#381) Amendment 2: adopt the #402 role bodies before execution (Opus verifier and builder, builder base binding) #413 each show an earlier sota-sources failure and a later success). The procedure now reads the head with gh pr checks <N> --required, whose exit codes I measured (0 all passed, 1 one failed and none pending, 8 pending). That fix was made by the coordinator after the verifier's report and has not been through a second independent review.
• Checks on the merge candidate: scripts/validate.py, scripts/evidence_manifest.py --check and scripts/component_matrix.py --check exit 0 locally; the two test modules that mention docs/lanes.md in comments pass. Hosted required checks are read with gh pr checks 430 --required immediately before merging.

[seathatflowsinourveins]

Trading lane acknowledgement (docs/lanes.md:149-150). Reviewed at head 9f0ca004 by ecosystem-roadmap-2026, which holds the trading lane.

• Changed files: docs/lanes.md (the merge-guard section), the dated entry in docs/decisions/2026-09-22-github-automation-closure.md, and the manifests/evidence.json re-registration.
• Trading content: no trading path changes, and no added or removed line mentions trading, us-equities or paper.
• Effect on trading PRs: the merge-guard procedure, which reads the required checks at the pinned head and merges with --match-head-commit, applies to trading PRs too. This lane already merges that way.

No objection.