The e2e 68 check "a clip that starts later" was red in 3 of the last
8 suites (CI tasks 697, 703, 712). Two causes sat on top of each other.
- The check measured too late. It started sampling after play(), and
under CI load the clock was already at 0.05 to 0.22 s. It now starts
before the click, like 68d does.
- Measured early, a real bug showed: the video held for one frame at
its first picture. The first picture triggers a redraw, and the game
takes its first step in the same frame. readSpeed got both calls with
the same frame time and divided by 1 ms. The speed came out at 16.7x,
which counted as a jump, so the player paused.
- readSpeed now measures a step without elapsed wall time against one
frame at 60 fps. A real jump still counts as a jump.
- e2e 34 "the pause holds the stream" also failed in task 712. Its
cause is still open. The failure message now names the state before
the pause, so the next red run shows where the samples got lost.
Tests:
- video-follow.test.ts 18/18, the new test is red without the fix
- e2e 68 on melektaus-atx: ALL PASSED, 21 checks
- Gate on melektaus-atx: CHECK EXIT 0, 314 files, 4853 tests
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
adminandClaude Opus 5.5
5b9d084dfe