[Castor-o7]

RAL-182, from the watchman: a_redraw_moves_the_state_bar_with_the_counts failed once for jaunty-harrier under full-suite load.

What the failure was: it failed at in_browser's assert!(ok) (tests/web.rs:1102 at 90de3d3), so the script ended badly, not one of the probe's assertions. Its output wasn't kept, so the exact error can't be named. No Chrome core was dumped between 05:30 and 05:50, and ci-local's logs have never shown this test failing alone.

The test's own timing assumptions, fixed (the timeouts are unchanged):

• Waiting for the board: it went on after 10 s whether or not the board had drawn. Now it waits up to 20 s for the first draw, and if there's none it returns an error that the test asserts on.
• Reading the widths: it read them after a fixed 1 s, which could catch a width mid-transition (0.6 s) when frames are starved. Now it awaits the bar's getAnimations(), so it reads only once the transitions have finished.
• The page's own poll: a 30 s poll could redraw the bar with the service's status in between. If one did, the probe draws its own status again, at most three times, and says how many times as redrawnByPoll.

Checked:

• Passes 3/3 alone, in under 3 s each.
• Passes in the full suite with a second web suite running alongside (load 22).
• With the viewer's style update removed, it still fails: "the bar kept its first draw".

🤖 Generated with Claude Code