ci(fork): use standard hosted runners - #11
Conversation
૮ >ﻌ< ა ci reviewran on 8b81966 — ci(fork): bound standard-runner test concurrency
|
|
Independent model review is bound to exact head |
|
Independent advisory review PASS is bound to exact head Review boundary: isolated Claude Opus session; no tools, repository credentials, write authority, label authority, merge authority, or deploy authority. The remote head and diff were independently reverified to match before applying Live standard-runner evidence on this exact head:
The remaining failed aggregate is solely downstream of the intentionally absent review label and is eligible for the label-rerun workflow. This label does not authorize merge. |
Summary
Root cause
Exact-head CI run
33740464585created and completed ordinary jobs onubuntu-latest,windows-latest, andmacos-latest, but five required jobs remained queued with an emptyrunner_namewhile requesting:ubuntu-latest-96-coreubuntu-latest-32-corewindows-latest-32-coreThose larger-runner labels are not provisioned for this personal fork. The
Label rerunworkflow then waited for the never-completing CI run, so the independent-review label could not clear the aggregate gate. This is a runner-capability mismatch, not a test failure.Attempt and pivot record
Earlier PR #7 diagnostics tested workflow-identity and transient GitHub orchestration hypotheses. They did not change the unavailable runner capability and are not repeated here. The new decisive evidence is the same-run contrast: standard-label jobs acquired runners and passed; every queued required job had an empty runner name and a larger-runner label. This remediation changes that exact condition.
Prediction: on this branch, each previously queued required lane acquires a standard hosted runner. Kill criterion: if any corrected lane remains queued without a runner, do not broaden permissions or add external collaborators; inspect the exact job labels and provider annotations again.
Exact identity
8178548437d63a2357fce98fa0b76d560d7207139a27a057119c06af942fafc0470ab79284357a6b577e7198332ab1d1fd81868f161dcb7bddc14addaf2de31e5dd600086d4ffec2Verification
uv run --extra dev python -m pytest tests/ci/test_fork_runner_compatibility.py -qfailed becausetests.ymlrequestedubuntu-latest-96-coreuv run --extra dev python -m pytest tests/ci -q— 121 passeduv run --extra dev ruff check tests/ci/test_fork_runner_compatibility.py— passedgit diff --cached --check— passedIndependent review
A fresh, non-persistent, tool-free
claude-opus-5review assessed the complete final staged diff. It had no repository credentials or write, merge, or deployment authority.577e7198332ab1d1fd81868f161dcb7bddc14addaf2de31e5dd600086d4ffec2Security and authority boundary
No trigger, permission, secret, executable step, action pin, branch protection, merge rule, or deployment authority changes. The reviewer produced evidence only; Rook remains the constrained repository broker and Tom retains merge authority.
Documentation impact
FORK_STEWARDSHIP.mdnow records the standard-runner policy, timeout bounds, measured-run follow-up, isolated model-review requirements, mandatory higher-risk escalation, and a current point-in-time divergence inventory.Rollback
Revert exact commit
9a27a057119c06af942fafc0470ab79284357a6b, then verify all five workflow runner labels and the regression test. Do not rewrite shared history.Acceptance gate
Keep draft and unmerged until the exact-head jobs prove that the corrected lanes acquire runners, required checks complete, and a fresh exact-head model review validates any post-review change.