What we did
A designed test planted a known effect and sent that evidence through RefCal's real update path. A separate scheduled production run then checked whether real catalogue monitoring wrote a complete, traceable record for every item it processed.
What we saw
RefCal recovered the planted effect within the tolerance fixed before the test, while unrelated bands and factors stayed stable. The scheduled production run processed 500 items and wrote 500 audit records, with no missing rows and the completeness invariant satisfied.
The production evidence was not strong enough to justify a real update: all 21 evaluations returned insufficient evidence and no verdict was written. Later versions remained sealed rather than active, so no delivered result changed.
What it means
The update and audit mechanism works where the answer is known, and the production path leaves a complete record while refusing to act on insufficient evidence. That supports a mechanism-shown stage. It does not yet demonstrate selective reprocessing, republication or a reproducible before-and-after product change.