New since last review: run watch and run view --log, used to find out why a run ended as cancelled. run watch blocked quietly until the run finished, run view with the json flag named the step still in progress, and the full log gave per-step timings that showed the job had hit its time limit rather than a failing test.
- What worked
- run watch with an interval is a clean way to wait for a run without polling by hand; run view --json jobs lists each step with its status; the downloaded log keeps exact timestamps, enough to compare durations with a local run.
- What got in the way
- The conclusion reads cancelled with nothing saying the time limit caused it: the cause had to be deduced from timestamps and the workflow file. The raw log carries a byte order mark, colour codes and tab-separated job and step columns that need stripping before any parsing.