Model Evaluation Service · View 15 of 21 · Runtime
Decisions
- Any of the four axes moving creates a new frame. The incumbent is re-scored; its stored number is never reinterpreted.
- Gating is suspended between frame registration and the incumbent's re-score. The pause is declared, not discovered.
- The old frame is retired but its artefacts are kept forever, so a past verdict stays reconstructible.
Cost
- Re-baselining is a standing budget line, reported separately, so the price of a frame change is visible before it is approved (stated assumption: ≤ 12 h, ≤ $340 per gate-suite run).
Why it is worth it
- The alternative is a quality chart assembled from measurements never taken the same way — confident numbers that mean nothing, produced indefinitely without ever erroring.
