- Shipped
- September 7, 2026 at 7:31 AM UTC
- Author
- Kamo
- Commit
- efea203
A source-level guard rather than a rendered assertion, because none of what went dark is visible to a rendered one. Three of the four faults were dark for only part of a keyframe cycle or only over one band's fill — jsdom sees identical markup either way, and a screenshot of a paused frame is as likely to catch the readable half of a pulse as the unreadable one. Checked against the pre-fix tree before being kept: it fails 7 of its 9 cases there, so it bites. It also carries the previous commit into a build that runs. That commit was verified sound — both image guards pass and tsc is clean against a pristine worktree of it — but its only CI run died in four seconds, which is a runner giving up rather than anything the tree did; a real guard failure takes ~2m24s and next build takes ten minutes. Re-running the same sha rebuilds without redeploying, so a fresh commit is the only thing that actually ships it, and an empty one would have been a worse use of the push than this.