Sentinel: shared:01m23vygyffx5hg2f5vm8r74sr:r24r5jtj4e is past its budget: trevato/forge: "the Files page of a 40-file diff: cold median of 3 and repeat median of 5 stay under their ceilings" fa… #5492

The Files-page timing test now judges its fastest view instead of the median. That should stop it failing when the check machine is busy. bun test test/pull-timing.test.ts passes locally (4 tests). I haven't seen it fail under load, so whether the flake is gone depends on the forge's next checks.

Cause: the old test took the median of 3 cold and 5 repeat views against wall-clock ceilings. Load on the machine could push a majority of those views over the ceiling even though the code was fine. The existing load scaling (#5000) only partly covered that.

Change: the fastest cold view (of 3) must be under 250 ms and the fastest repeat view (of 5) under 100 ms. Load only adds time, so the fastest view is the one least affected by it. The test still fails if the page itself gets slower, because then every view is slow.

I renamed the test to "fastest cold view of 3 and fastest repeat view of 5 stay under their ceilings", so its name matches what it checks. The sentinel's cause is keyed on the old name, so it should clear once the old name stops failing.

Not done: I left the ceilings and the load-scaling code unchanged, and I didn't touch any code under src/.

To try it: bun test test/pull-timing.test.ts.

🤖 Generated with Claude Code

Closes #5459

The Files-page timing test now judges its fastest view instead of the median. That should stop it failing when the check machine is busy. bun test test/pull-timing.test.ts passes locally (4 tests). I haven't seen it fail under load, so whether the flake is gone depends on the forge's next checks.

Cause: the old test took the median of 3 cold and 5 repeat views against wall-clock ceilings. Load on the machine could push a majority of those views over the ceiling even though the code was fine. The existing load scaling (#5000) only partly covered that.

Change: the fastest cold view (of 3) must be under 250 ms and the fastest repeat view (of 5) under 100 ms. Load only adds time, so the fastest view is the one least affected by it. The test still fails if the page itself gets slower, because then every view is slow.

I renamed the test to "fastest cold view of 3 and fastest repeat view of 5 stay under their ceilings", so its name matches what it checks. The sentinel's cause is keyed on the old name, so it should clear once the old name stops failing.

Not done: I left the ceilings and the load-scaling code unchanged, and I didn't touch any code under src/.

To try it: bun test test/pull-timing.test.ts.

🤖 Generated with Claude Code

I haven't changed anything since my last push. The sentinel's note is the same failure, and it is still reported because the fix isn't on main yet.

The sentinel's failure sentence names the old test, "cold median of 3 and repeat median of 5". That test only goes away when the pull merges, because the fix is on this branch (commit 0ac8e5e0, already pushed).

What the pull changes: test/pull-timing.test.ts no longer judges the median of 3 cold and 5 repeat Files-page views against wall-clock ceilings. A busy check machine can push most of those views over a ceiling even when the code is fine. It now judges the fastest view of each set: under 250 ms for the fastest cold view of 3, and under 100 ms for the fastest repeat view of 5. Load only adds time, so the fastest view is the one least affected by it. The test still fails if the page itself gets slower, because then every view is slow. The existing load scaling stays as it was.

Renamed test: it is now "fastest cold view of 3 and fastest repeat view of 5 stay under their ceilings". The sentinel's cause is keyed to the old name, so it should clear after the merge.

Not done: I left the ceilings alone and made no changes under src/.

To try it: bun test test/pull-timing.test.ts. It passes locally (4 tests).

🤖 Generated with Claude Code