Twinned from https://github.com/cfsh/ninefold/issues/843 by tools/beads/import_github.py's reconcile pass.
Filed unclaimed from nf-dev-delete-board. Not claimed, not labelled — leaving the record so it is not lost, no more than that.
The post-merge UI gate for #838 recorded a new failing probe on main:
``` gate: FAILING — main @ 8602a71 has new failing probes (1 failed of 34) ```
./tools/pr/probe_gate.py --show on nf-dev-sg:
```json "failing": { "split-notes-editor": { "since": "8602a71...", "sinceDate": "2026-08-11T05:43:38Z" } } ```
It is attributed to the wrong commit, and probably to no commit at all. #838 deletes the session board — it touches no .cs, no .tscn, no project.godot and no probe JSON, so there is nothing in it a probe can observe. Checked rather than assumed:
2b10361 (the merge before): probe passes. Fresh detached worktree, dotnet build, then ./probe.py tools/probes/split-notes-editor.json — every expect_focus / expect_prop OK, [probe] done.8602a71 (the merge itself): probe passes. Same command, three consecutive runs, exit 0 each time.So the sweep and an on-demand run disagree at the same commit. The interesting question is which conditions differ — the gate sweeps ~34 probes back to back in a dedicated detached worktree, and this box logs All audio drivers failed, falling back to the dummy driver on every launch. Contention or ordering is the obvious suspect; I did not chase it.
⚠️ The reason this is worth a record: the entry is self-suppressing. A probe in failing is a known failure, so it cannot fail the gate again. notes/probes.md and runbooks/provision-a-box.md both name that shape — *"absorbs it into standing where it can never fire again — silent, permanent coverage loss"*. Whoever picks this up decides between re-sweeping to clear the entry and hardening the probe; I was told to leave it, and did.
Baseline lives on the orphan screenshots branch at probe-baseline/main-linux-nf-dev-sg.json.
None.
No comments.