Saved-response replay stalls too, with prompt reads; the long interval is inside the synchronous file write.

Then move persistence off that display path. Bounded queue, visible failure handling.

Decide what the operator sees when the queue fills. Silent loss of diagnostic records would make the next fault harder to explain, even if the screen looks beautifully smooth.

Writer now runs separately, with a queue limit and logging warning; delayed storage, full queue and unavailable folder tests leave the display responsive and expose the logging problem.

Delayed-read replay still produces a stale-reading indication, and normal replies restore it; the write stall is fixed without pretending every communication delay has disappeared.

Put the logging warning in the handover with the location of retained records and the named maintainer. Someone on the covering shift will otherwise dismiss it as a nuisance message.