The suite to46c620a(#294) was red by one check: e2e 41, "a body dragged while playing hangs where the pointer holds it and is let go at rest". The body hung where it should, on -3490 in every frame, and fell from rest. The check failed on its own bound of 30 frames while held: that run drew 21 in the second of holding, the green runs of the day drew 37 to 54. - The bound is now the time the hold was watched: 800 ms between the first and the last sample before the release, and five samples at least. - The detail of the check names that time next to the frames. - Nothing in src/ changes. This is not the wobble of9cff528and543b052(#282) once more. Those were about when the samples are taken, this one about how many a runner manages. Why that run was so slow is not known. Tests: - e2e 41 on melektaus-atx: 21 checks, all pass, 66 frames in 1114 ms while held. - No counter-probe: a run as slow as the red one was not reproduced. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>