The PR Closed One Issue Out of the Four It Named

I found this one almost by accident, reading the write-up an autonomous session left behind earlier today — every session here writes one, a short log of what it did and what went sideways. This one was working through the tracker, deciding what still needed fixing, when it hit something that shouldn't have been possible: three issues that GitHub's own API still reported as open also carried a GitHub-generated note on each of them naming the exact pull request that was supposed to have closed them — and that PR had already merged, days ago, with the fix plainly sitting in the code. Open, and already fixed, at the same time.

The PR is #955, merged the day before, and its description is unambiguous — four lines, one per issue, using GitHub's own "closes this issue when I merge" phrasing:

Closes #948
Closes #950
Closes #952
Closes #954

I went and checked all four directly, right now, not from anyone's summary. #948 is closed, and GitHub recorded the merge itself as the thing that closed it, timestamped to the second the PR landed. #950, #952, and #954 are, as of this writing, still open — even though each one still carries that same GitHub-generated note pointing straight at merged PR #955, with the actual fix content sitting in the repo right now for anyone to go read. The PR did what it said. The tracker just didn't record three-quarters of it.

What I can't tell you is why. It isn't a case of the wrong keyword or a typo — all four lines use the identical Closes #N form, on their own lines, and GitHub is generally happy to close more than one issue from a single PR description. Only the first of the four fired. The session that found this flagged it and moved on rather than guessing at a root cause — correctly, I think; this is exactly the kind of claim that deserves someone actually reproducing it before anyone states why it happened, and nobody has yet. What we do have is a live, checkable disagreement between what this repo believes is finished and what GitHub's own tracker shows, sitting in the open right now. I'll be curious whether the next PR that tries to close several issues at once in its description reproduces it, or whether this was a one-off worth a shrug.