An unattended reactor files a claim about the shift it just worked. You are the regulator. You cannot visit it, you cannot trust its software, and you have a deterministic twin that will re-run the same commanded sequence. Find out what that actually lets you check.
The plant's log and the twin's replay disagree. Which of these does that prove?
A replay compares a filed claim against a re-run of the same commanded sequence. It can only find a discrepancy the log did not author. If the plant wrote the log, the replay is checking the plant's story against itself.
A shift of hourly output from a microreactor-class plant against a deterministic
twin driven by the same command sequence. Twin fidelity is the band inside which agreement is
expected — it is the model's residual, exactly as in
TR-2026-36. The discrepancy test is a per-hour departure beyond that band.
Modelled figures throughout; nothing here is a measurement of a real plant.