The first CI run after 9cff528 failed e2e 41 again, at the release this
time (task 405): the hold was fine for 54 frames, but the first sample
after the release was 6.9 units below the hold, over the bound of 3.
The test took its release time after the round trip of the mouse event
and one more evaluate. The frames in between were missing, so the
first counted sample was already about 100 ms into the fall.
- The page stamps the release itself with a capturing `pointerup`
listener. Held and released samples are split at that stamp.
- The fixed bound on the first sample after the release is gone. How
far that frame has fallen follows the frame time. Instead every
sample in the first 200 ms must stay below the hold and keep
falling, which is what "let go at rest" means: no jump, no kick up.
- The 200 ms fall still has to be more than 1 and less than 40 units.
The comment now says about 25, as measured, instead of under 20.
Tests:
- e2e 41 on melektaus-atx: 8 of 8 runs ALL PASSED, 21 checks. The
first sample came 6 to 42 ms after the stamp.
- Counter-check: a release velocity of y -2 instead of 0 in
PhysicsSystem makes the body rise to -3510 and the check fails.
- Gate on melektaus-atx: 195 files, 2829 tests, CHECK EXIT 0.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
adminandClaude Opus 5.5
543b052136