the refusal to explain a run that has not concluded is measured by no test #2712

scripts/read-run-failure.py reads a run's own conclusion to decide whether a failure exists at all. A run that has not finished carries conclusion: null, and the tool raises rather than accepting that as a verdict, so a still-running run is answered as unmeasured — exit 2 — which is the right answer.

Nothing measures that branch. Deleting the null check leaves every one of the file's 31 tests green: the single test that reaches this path hands the reader a transport error (run_rc=1, an HTTP 502) and never a payload whose conclusion is null.

What the deletion would cost. str(None) is the string "None", which is not "success", so the run would fall into the branch that explains a failed run. A run that is still queued or in progress would then be reported as "the run itself concluded None, and the job(s) that account for it are: …" — a determination about a run that has reached none. That is the defect this area just had removed, one step further along: an unmeasured question answered as a verdict.

Observed live on run 41938836857 while it was still running: the tool exits 2 and says "whether this run failed is exactly the question, so it is unmeasured rather than 'nothing to explain'". The behaviour is correct; the measurement of it is what is missing.

What finishing this needs:

• A test whose run payload carries conclusion: null, asserting exit 2, the word unmeasured and the "no conclusion yet" reason on stderr, and that no explanation of a failure is printed on stdout.
• The fixture must be able to model a run that has not finished, so the reason the tool prints names the state such a run actually has (in_progress) rather than completed.
• Deleting the null check must fail that test — that mutation is the acceptance criterion, not a courtesy.