From ee853972675d93ca78ff26d9b61e6a264b4b1147 Mon Sep 17 00:00:00 2001 From: ci-check Date: Fri, 6 Nov 2026 22:42:00 +0800 Subject: [PATCH] docs: correct "~30 min window" claims in ci.yml comments to measured schedule delivery --- .github/workflows/ci.yml | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-) diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index 78cf7280f7..bbe5edc165 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -18,10 +18,13 @@ on: # and release lanes fire on their own master-push triggers), so every # merge paid full-shard minutes for a signal nobody saw. Master coverage # does not lapse: the :03/:33 schedule cron below re-runs the FULL - # matrix whenever master moved since the last green scheduled run - # (bounded ~30 min window), and PR-side blocking still gates everything - # pre-merge. Release/bugfix push arms stay — they exist because - # bot-authored PR checks sit in action_required (#772). + # matrix whenever master moved since the last green scheduled run. + # Schedule delivery is throttled under load: target 30 min, but measured + # 2026-11-06 ≈5 runs/day against the 48-tick spec (median gap ≈4.8h, + # worst ≈10.6h over 107 runs) — treat post-merge coverage as a + # multi-hour backstop, not a 30-minute one. PR-side blocking still + # gates everything pre-merge. Release/bugfix push arms stay — they + # exist because bot-authored PR checks sit in action_required (#772). branches: ["*_release-v*", "bugfix-release/**", "bugfix-v*", bugfix] pull_request: # agent-relay: #3468 agent PRs target the relay branch. @@ -41,8 +44,11 @@ on: # Three-layer CI (#3295): PRs pay for the CORE matrix only (master pushes # stopped paying in 2026-11-01 — see the push arm above; # this cron is now master's standing verifier). - # Every 30 minutes this cron asks the plan job whether master moved + # Every 30 minutes (spec) this cron ASKS the plan job whether master moved # since the last green scheduled run; only then does the FULL matrix run. + # Observed delivery is far sparser than the spec (measured 2026-11-06: + # ≈5 runs/day vs 48 ticks, median gap ≈4.8h, worst ≈10.6h) — see the + # push-arm note above. # Lane workflows (e2e/registry/image/theorem/zcode) follow the same gated # cron pattern on their own offsets. schedule: