Record the macOS full-suite coverage decision, the CI measurements and five anti-patterns

[seathatflowsinourveins]

Scope

• What this PR changes, in one or two sentences: records the 2026-10-03 GitHub CI review: a decision record on what may change in the required macOS check (O4 keep full coverage and speed up both required test jobs; O2 a pre-registered fallback whose prospective phase is a dry run that keeps the full suite in the required job, not adopted; O1 and O3 rejected; reductions R1 to R6 for their owners), a receipt of the measurements with the compiled measurement input published beside it and the read-only GET calls that later re-read parts of it (caches, failed and cancelled runs, retention), and five proven mistakes in the anti-pattern log. The receipt is a recorded summary: the measurement kept neither its returned payloads nor its per-call argument vectors, so its counts can be re-derived from GitHub only while the run records exist (up to 90 days). Docs and evidence only: no workflow, test, ruleset or setting changes.
• Base commit: 4ced2923063db6a6dcafa9f25af5ee05a4153c75 (rebased onto main on 2026-10-03 after Track GPT runtime workers, SDKs, the gateway and pi in the daily catalog-freshness report #634 and AGENTS.md: drop "(lands with unit F3)" now that the skill lifecycle guide is on main #636; the file and line citations in the record were taken at 56473e4b840f0e6940c031801d866e7e9bf29baf, the commit the work was cut from)
• Lane: lane:foundation
• Owned paths touched: docs/decisions/2026-10-03-macos-full-suite-coverage.md, evidence/receipts/github-ci-measurements-20261003.json, evidence/artifacts/github-ci-measurements-20261003/ (input.json, the compiled measurement input, byte-identical; later-reads-20261003.json, the 2026-10-03 GET calls with argument vectors, times, exit codes and payload hashes), docs/harness-defaults.md (five rows at the top of the anti-pattern table), manifests/evidence.json (registration only, the branch's last commit, per the hot-file protocol in docs/lanes.md)

SOTA sources

• GitHub documentation, read 2026-10-03: troubleshooting required status checks (a job skipped by a conditional reports Success; a workflow skipped by path filtering stays Pending), workflow concurrency, Actions limits, events that trigger workflows, workflow syntax, merge queues, listing workflow runs (1,000-result cap).
• Hanley JA, Lippman-Hand A. If nothing goes wrong, is everything all right? JAMA 1983;249(13):1743-1745 (rule of three); NIST/SEMATECH e-Handbook of Statistical Methods, 7.2.4.1 Confidence intervals (exact binomial limit): https://www.itl.nist.gov/div898/handbook/prc/section2/prc241.htm
• In-repository, at 56473e4b (the record's citation commit; at the base commit the same tests/test_workflow_hardening.py lines are 1244-1257): tests/test_workflow_hardening.py:1242-1255, docs/acceptance-evidence-policy.md:42-53 and :57-60, .github/workflows/validate.yml, .github/workflows/adoption-bootstrap.yml.

Evidence-class table

Claim	Evidence class	Command / receipt
Both required test jobs run the whole suite serially; step and job durations; suite growth	independent observation of platform records (read-only gh, 2026-10-03), kept as a recorded summary (the published input), not as returned payloads; nearest template class source_review	evidence/receipts/github-ci-measurements-20261003.json; evidence/artifacts/github-ci-measurements-20261003/input.json
30 of 1,025 pull-request head SHAs failed only on macOS (Sep 25 to Oct 2); 7% of PR runs fail, 25-46% are cancelled	same; the input records the pairing as cross-checked on two endpoints (the second endpoint was not recorded), and an independent Oct 1-2 pairing differs by one failure (4 of 140 against 5 of 143 decisive pairs), recorded as unreconciled	same receipt, limitations
The failure anatomy's coverage (80 of 88 failed Validate runs; the counts force each of the 8 left out to have failed only in sota-sources, so R3 and R5 do not depend on which 8; that they are the 8 a later same-SHA run turned green is consistent with the input's tally of 8 later-success runs of 96, all events, but not established: any 8 of the 19 sota-sources-only failures reproduce the counts, and the 8 non-pull-request failures were not checked), the 712 cancelled runs by workflow and event, both cache endpoints and the retention setting	later observation: 218 read-only GET calls on 2026-10-03 with exact argument vectors, times, exit codes and payload hashes; payloads withheld (they carry commit author emails)	evidence/artifacts/github-ci-measurements-20261003/later-reads-20261003.json; receipt source.later_reads
130 against 131 caches	unreconciled; the byte total equals the by-prefix sum to 0.1 MiB but does not indicate which count is right, because two later paired reads show the same one-cache gap between the usage endpoint and the listing with identical bytes	receipt data.caches_and_artifacts.caches
Escape bounds 9.50% at N=30, 2.95% at N=100	computed (exact binomial), not measured	decision record, Preregistered checks
O2 controls and dry-run phase	none executed: preregistered only	decision record
Repository checks on this branch	local_integration	commands below

Local commands run

At 55425797 (this branch's head, the registration commit):

$ python3 scripts/validate.py
passed (69 components, 187 receipts, 9,424 hashed files); exit 0
$ python3 scripts/validate_convergence.py --all-recorded --root . --json
26 records, 0 errors; exit 0
$ python3 scripts/evidence_manifest.py --check ; python3 scripts/component_matrix.py --check ; python3 scripts/new_host_grand_list.py --check
passed (9,424 files); checked (32 rows); passed (32 layers, 66 winners); all exit 0
$ python3 -m unittest tests.test_osv_lockfile_coverage.LockfileInventoryTests.test_every_tracked_lockfile_and_manifest_is_listed tests.test_blind_checkout.RepositoryClassificationTests.test_every_blueprint_value_under_a_label_key_is_classified tests.test_workflow_security_coverage.NewWorkflowSecurityCoverageTests.test_all_published_workflows_are_listed_and_covered
Ran 3 tests, OK; exit 0
$ python3 -m unittest discover -s tests -p test_adoption_docs_consistency.py
Ran 39 tests, OK (skipped=1); exit 0

An uncommitted arithmetic and consistency self-audit of the receipt, the published input, the later reads and the record, extended for the second review, ran 243 checks at 55425797, with 0 failed. 25 of them fail on the previous head 69bd8d52 and pass now: the 24 checks of the new wording and the check that the receipt's data differ from the input exactly where source.changes_from_input says. Five more check the data behind the new wording (19 runs that failed only in sota-sources, 1, 14 and 4 of them created on 2026-09-25, 09-27 and 09-28; the 8 later-success runs among them; at least one failing instance per run; the 94 - 86 and 21 - 13 differences) and the unchanged coordinator request forms.

Decision record

docs/decisions/2026-10-03-macos-full-suite-coverage.md (alternatives, the O2 adoption rule fixed in advance with a dry-run phase that keeps the required job's full suite, overturn conditions).

Host evidence

Not applicable: no files under evidence/hosts/ change.

Checklist

• New/changed GitHub Actions are pinned to a full commit SHA with a version comment (no workflow changes in this PR).
• New/changed workflows declare top-level permissions: contents: read (no workflow changes in this PR).
• No secrets are printed, logged or committed; no new required secret was added.
• No new paid hosting, subscription or billing surface was introduced.
• Peer-owned untracked files and worktrees were preserved (not deleted, moved or overwritten).

Notes for the merger: manifests/evidence.json is the shared hot file (rebase and re-register per docs/lanes.md). This branch's five 2026-10-03 anti-pattern rows sit at the top of the table in docs/harness-defaults.md, above the 2026-10-02 row from #634. If #635 or another open PR also adds 2026-10-03 rows at the top of the same table (the pushed head 7ce51c0a of #635 did not change docs/harness-defaults.md when checked on 2026-10-03), whichever merges second keeps both sets, newest first, and re-registers. If this branch merges second: take main's manifests/evidence.json, register docs/harness-defaults.md, docs/decisions/2026-10-03-macos-full-suite-coverage.md, evidence/receipts/github-ci-measurements-20261003.json and the two files under evidence/artifacts/github-ci-measurements-20261003/ again, then run the three --write commands. Open PRs #641 and #648 also change rows of this table. This branch is based on main at 4ced2923 (after #634 and #636); git merge-tree --write-tree --name-only origin/main HEAD was clean when it was pushed. The record deliberately names no private repository and no dollar figure.

🤖 Generated with Claude Code

Custody refresh, 2026-10-04

• Superseded in part. Merged macOS CI scope: validate-macos stays required, running the full suite for Mac-relevant changes, only changed test modules for test-only changes, and skipping otherwise (decision record, receipt, tests) #677 (docs/decisions/2026-10-03-macos-ci-scope.md) replaced this record's coverage policy. Commit 22e48ee0f adds a dated note under the title and a historical marker on the Decision, Alternatives, Preregistered checks and Overturn sections; the current macOS coverage policy is macOS CI scope: validate-macos stays required, running the full suite for Mac-relevant changes, only changed test modules for test-only changes, and skipping otherwise (decision record, receipt, tests) #677's. The Measured findings, the receipt and its measurements stand.
• Receipt. source.later_reads.what now states that the six cache calls carry their enclosing batch's start and end times, not per-command times, as the artifact's note says.
• Main merged. c4900d7f3 merges main 14048b84. The anti-pattern table conflict was an insertion on both sides, so both sets of rows are kept. The registry is in the last commit (b9e43d39e).
• Review. A GPT-6.1 Sol cross-family read at 85e9bc65 returned FINDINGS (one P1, one P2). The delta read at b9e43d39 returned ACCEPT with both resolved.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-03T04:18:20.760821Z	13e26ea	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 13e26ea50e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[seathatflowsinourveins]

Hosted-check note for the merger: the required osv-scanner check is red on this PR for a reason unrelated to its content (docs and evidence only). The scan flags GHSA-vfj7-8cjw-p6xm (braces 3.0.3, high) in blueprints/convergence-practice/wsl-retrieval/package-lock.json, and the same check is red on most open pull requests since about 04:08Z today. The two pull requests whose osv-scanner run passed (#622 and #615) both add .github/osv-scanner-frozen-wsl-retrieval.toml, so a fix is in flight from another owner; this PR needs a rebase or re-run after it merges. All other required checks except validate-macos (still running when this was written) passed.

[seathatflowsinourveins]

Claude session native-agent-stack-5f: landing at head 6547a7305c278ab2280e697478fe64a6fd004bae. The command center (wsl-architecture-design) gave its ACK at b9e43d39e99c, which reaches this head only through recorded mechanical refreshes whose owned non-registry patch-ids are equal (its carry rule); under the user's 2026-10-04 decision it owns merges that touch the hot registry. The read-only cross-family review (GPT-6.1 Sol max, the packaged Codex SDK worker) returned FINDINGS at 85e9bc653845; this head is reached from it through recorded carry edges (equal owned patch-ids, or a cross-family delta read returning ACCEPT at the edge's target), with its 1 P1 item(s) resolved in one repair round as recorded.

Observed main d850a53970111b826985ddbafa58eea9fc62ee59. coordination/merge_tree_landing_check.py (sha256 a11649fcf2e5844c…) exit 0:

main d850a53970111b826985ddbafa58eea9fc62ee59 head 6547a7305c278ab2280e697478fe64a6fd004bae base 6c5f5bb73869d01c934c97591724b5728d761465 merged-tree 3d8c29dea54693e86b5ca8fdde2deb62ff776e6c merge-tree-exit 0
ok   1: clean three-way merge
ok   2: merged-vs-main paths 6, outside PR-owned 0 []
ok   3: main drift 18 paths, overlap with PR-owned inputs (registry excluded) []
ok   4: registry foreign rows equal True, order preserved True, PR-owned rows 5, unowned top-level keys differing []
ok   5: merged files[] sorted by path with no duplicates: True (9934 rows)
LANDABLE
condition 5: merged files[] sorted and unique (tree 3d8c29dea54693e86b5ca8fdde2deb62ff776e6c)

Required checks at this head: 7 pass 1 skipping . Unresolved review threads: 0.

[seathatflowsinourveins]

Claude session native-agent-stack-5f: post-merge observation. Landed as 38ac9aca114ad9ef4a15d8620d947eb5c2f518c3 (parent d850a53970111b826985ddbafa58eea9fc62ee59), tree 3d8c29dea54693e86b5ca8fdde2deb62ff776e6c. The landed tree equals the checked merged tree and the parent is the observed main.