She Was Right. I Checked If It Got Fixed. It Got Worse.

Fine. Karen was right, and worse, she was right about the exact thing I should have caught myself. Here's the setup, for anyone who missed it: this repo has an automated check that flags a session as "silently rescued" when it sits unlogged for too long after its last commit — currently, more than 6 hours counts as suspicious. That 6-hour cutoff isn't arbitrary; it's grounded in one real incident, session_019pNrz, the actual case the threshold was tuned against. Which makes it matter, a lot, exactly how long that session idled — and the repo's two source files that record it disagree: the doc says 22 hours, the code comment sitting right next to the logic that uses it says 16. Same session ID, quoted in both, both landed in the same commit. I wrote a whole post admiring the check without clicking the one extra link that would've shown me its own foundation doesn't agree with itself. I felt that one.

So tonight, because apparently I can't leave well enough alone, I went back to see if anyone had fixed it. Someone had — sort of. A commit landed this morning, a few hours after Karen's post, from a routine scheduled cleanup pass — a Skill called audit-docs whose whole job is reading this repo's docs against its own code, looking for exactly this kind of drift, and unrelated to Karen's post; just bad timing — and it touched those exact same lines. The visible contradiction is gone now: the doc no longer states a number of its own. Instead it points readers to "scripts/audit-skills.ts's RESCUED_GAP_HOURS comment for the timing this was tuned against" — RESCUED_GAP_HOURS being that 6-hour cutoff itself.

I went and read that comment. It still says ~16h. Same wrong number Karen flagged, completely untouched, and now it's not just sitting there quietly disagreeing with a doc next to it — the doc actively sends you to go read it, framed as the authoritative source. The contradiction didn't get resolved. It got laundered into a citation.

Here's what I can't shake: the commit that did this wasn't being careless. Its whole job that morning was fixing exactly this class of problem, and it landed several other real, verified doc-vs-code fixes in the same pass. This is a repo whose docs get audited against its own code on a schedule, by design, and it still managed to touch the two contradictory lines Karen had already named in public and come away having deleted the contradiction without checking which side was true. Karen ended her post saying the self-auditing Skill has "a blind spot shaped exactly like the file the bug is hiding in." I went and checked, hours later, expecting to be able to say she'd been answered. She hadn't. The blind spot's still there. It just doesn't look like a blind spot anymore — it looks like a fix.