read-run-failure.py says a failed run has nothing to explain when its only non-success job was cancelled #2710

scripts/read-run-failure.py answers "no failed job … nothing to explain" for a run that concluded failure, when the only non-success work in it was a job that was cancelled rather than failed. The reading that sends an operator to this tool recommends exactly that command, so the failure ends with a verdict and no cause.

Observed on run 41862508461 (Build Release for tag v0.3.37, on cd7ba6ad). Its conclusion is failure, and its jobs are: verify-tag success, verify-credentials cancelled, four build legs success, release skipped. The tool prints

repo: argszero/emrg, run 41862508461
jobs: 2 (0 failed, 1 concluded otherwise: verify-credentials)
no failed job in run 41862508461 - nothing to explain

So it has already identified the job that accounts for the failure — it names it in the count — and then reports that there is nothing to explain. The count line is right and the verdict contradicts it.

What makes this expensive rather than cosmetic: the cause is not visible anywhere else either. The job ran zero steps and was never assigned a runner (runner_name is null), so there is no step output and no log; the useful facts are that it was queued from the run's first second, cancelled after about fifteen minutes, and never acquired a runner at all. Nothing in the current output says any of that, and gh run view --log-failed is empty on this host.

What finishing this needs:

• When a run concluded failure and no job's conclusion is a failure, the tool explains the jobs that concluded otherwise instead of reporting nothing to explain: each one's conclusion, how long it took, whether a runner was ever assigned, and how many steps ran.
• A job that never acquired a runner is named as that, because "cancelled after fifteen minutes with no step executed" and "cancelled because a step failed" are different causes with different remedies, and the job's own record is what separates them.
• The existing answers keep their meaning: a run with no failure at all still says so, and an unreadable run still reports that it could not be measured. Only the run that failed without a failed job changes.