The projected reference stopped at its heat-flux limit
The owner chose a standard-settings reference for the MAST Upgrade transport comparison, with negative heat coefficients projected to zero and a stop limit on the size of the projected heat flux. A review of the saved job shows that it reached that limit and failed (the projected-reference job failed after 108 s when the log reported 3.93119 gB against the 1 gB stop limit; no completed reference was savedC-217). No completed reference trajectory exists, so the candidate’s coupled temperature acceptance remains unevaluable.
The runner checked the limit before saving its diagnostic history. The log retains the magnitude, but the triggering face, state and channel were lost. The revised runner saves each accepted-state projection before checking the limit and retains raw triggering diagnostics on a stop. It also keeps nonfinite coefficients invalid, and the result reviewer requires complete guard evidence and comparison inputs bound to the executed package.
Synthetic tests exercise the stop through the real loop with the physical solver forbidden. The frozen training manifest and model hashes remain consistent with the approved experiment. The plan now records the consumed failed job and the missing evidence. The current model, guard and acceptance limits remain unchanged.
What’s next
OQ-42 (25 Oct 2026) limited the MAST-U scope: the closed-loop reference comparison is reported as not evaluable, and no further MAST-U reference compute is planned. Entry 70 covers that decision. The reporting fix does not establish reference admissibility.
Technical details → Verification › projected reference stop review
Decisions → OQ-33 · OQ-38 · OQ-42