Measure read start and finish, display update, and log-write start and finish separately on the same replay. Compare with a controlled slow writer. Keep observation time attached to the sample so a delayed save does not look like a later controller reading.
Move persistence off the display path with a bounded queue and a visible logging-failure policy. Test a full queue and unavailable destination too. The diagnostic screen should report stale readings honestly if acquisition stops, even after the writing delay is fixed.