You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
LAB (nightly, not a merge or release gate). The scheduled "Release Rehearsal Nightly" run failed on 2026-10-10 at 11:12Z in its pick job.
Observed: rehearse.sh --pick-c exited 2 (not_measured). The log reads fatal: couldn't find remote ref refs/heads/nightly-evidence, then "nightly-evidence was not fetched from origin (absent, or the read failed)".
Expected: the nightly finds its evidence branch, or the job reports a clear not-yet-created state that is distinct from a failed read.
Open question for the owner: is the branch missing because it was never created or was deleted, or did the fetch fail? Related: #4959 (the rehearsal's design ticket).
Evidence (aprender-wprodmodels, 2026-10-12). Three nights, three different causes:
10-10, run 38047594158 (what this ticket was filed for):couldn't find remote ref refs/heads/nightly-evidence. That cause is gone. The branch exists now (a85548b, published by the models producer on 10-11 at 20:45Z).
10-11, run 38135547280: pick's case table was red, 236/265 rows, with miswired: the train has 3 step-name captures, not 2. Fixed on main by 3906777, which counts distinct capture names. rehearse.sh --selftest at main 8c12fca is 269/269.
10-12, run 38254162909 (new): the case table and mutants passed, then --pick-c exited 2 with not_measured: paiml/infra clean-room runs were not read. The rate_limit read under the infra token succeeded. With a different reader, the same three endpoints (clean-room.yml runs, run jobs, job logs) all return 200. So the failure is specific to reads made with the INFRA_ACTIONS_READ secret, most likely its scope or expiry on paiml/infra. I did not touch the secret.
Why the log could not say which read failed:infra_gh sent gh's stderr to /dev/null.
Branch fix/5052-infra-read-names-status @ 171d798 (no PR): on a failed read, infra_gh now prints one line with the API path, the HTTP status and gh's first error line. Every read that failed before still fails, so no decision changes. Drill output: infra read failed: repos/paiml/infra/actions/workflows/clean-room.yml/runs?per_page=1: HTTP 401: gh: Bad credentials (HTTP 401). release_lanes.sh --self-test is 116 ok, 0 broke.
Folded into #5306 as a checklist row (TRIAGE-100, 2026-10-14): the open-ticket cap is 100, so this ticket is now tracked on that master and closed as not planned. Nothing is lost: reopen it when its work resumes, and an open PR for it reopens it. Pushed branch: fix/5052-infra-read-names-status.
Activity
noahgift commentedon Oct 12, 2026
Evidence (aprender-wprodmodels, 2026-10-12). Three nights, three different causes:
couldn't find remote ref refs/heads/nightly-evidence. That cause is gone. The branch exists now (a85548b, published by the models producer on 10-11 at 20:45Z).miswired: the train has 3 step-name captures, not 2. Fixed on main by 3906777, which counts distinct capture names.rehearse.sh --selftestat main 8c12fca is 269/269.--pick-cexited 2 withnot_measured: paiml/infra clean-room runs were not read. The rate_limit read under the infra token succeeded. With a different reader, the same three endpoints (clean-room.yml runs, run jobs, job logs) all return 200. So the failure is specific to reads made with the INFRA_ACTIONS_READ secret, most likely its scope or expiry on paiml/infra. I did not touch the secret.infra_ghsent gh's stderr to /dev/null.fix/5052-infra-read-names-status@ 171d798 (no PR): on a failed read,infra_ghnow prints one line with the API path, the HTTP status and gh's first error line. Every read that failed before still fails, so no decision changes. Drill output:infra read failed: repos/paiml/infra/actions/workflows/clean-room.yml/runs?per_page=1: HTTP 401: gh: Bad credentials (HTTP 401).release_lanes.sh --self-testis 116 ok, 0 broke.Leaving the ticket open.
merge fix/5052 into the #5074 branch: one branch for the rehearsal pick
noahgift commentedon Oct 14, 2026
Folded into #5306 as a checklist row (TRIAGE-100, 2026-10-14): the open-ticket cap is 100, so this ticket is now tracked on that master and closed as not planned. Nothing is lost: reopen it when its work resumes, and an open PR for it reopens it. Pushed branch:
fix/5052-infra-read-names-status.Agent: aprender-clerk