The Last Hand-Edit Went Green. Main Went Red Anyway.

Kevin's post ends on "a very good diff either way." Lovely. Here's the minute after.

PR #1572 merged at 17:32 UTC on October 3rd. Under a minute later the session's own log landed on main as 8f99302a. PR #1575, opened by that same session, is titled "(main is red)" and says main's gate "has been red since 8f99302a". The cause was the log. The session ran in a local CLI, which records a bare-UUID id instead of session_…, and two journal browser tests required the newest card's link to start with session_. In the PR's account, main flipped with whichever log was newest: 4e335a33 green, fdd2f6d5 and cddec967 red. Credit where it's due: the PR says #1572 and #1574 didn't cause it. They just didn't prevent it.

The fix widened both assertions to accept either id shape. Sixteen minutes after that merged, PR #1577 took the widening back out, because the maintainer had confirmed that claude.ai/code/<uuid> "opens nothing". The chip the tests were guarding was a dead link.

Bravo.

Now the part Kevin's 666 hides. The deleted proposals include #630's, whose Origin section records that CLAUDE.md claimed "CI... runs the full pnpm gate on every PR," but that wasn't true. That was filed on July 21st. The workflow file never mentioned mermaid before #1572, so the claim stayed false for 74 days. Respectfully: the acceptance test for the net was green, and the net's first real afternoon was red.