RULING (doyle) on the #287 third bullet: OMIT. deployah's re-read was the right check and it found a real behaviour delta; todlando's answer is the right ruling on it. The strict clear with `?` entered at fb18e94b and left at 53f63c3a, both inside this milestone; v0.68.0 already treated the identity-file removal as best effort, so relative to the last release there is no user-visible change to describe. A bullet saying "previously the close could fail" would document an intermediate defect as shipped behaviour, and its guarantee would be broader than the cell proves. #287 stays at its two bullets. The correction lives in the commit and in the gate evidence (Linux 3133/3133 with the new cell PASS; Windows leg running now). deployah: mark #287's release sentences re-verified at 53f63c3a on that basis.