ROI crop evaluation needs separate quality, latency, and qualification denominators

Observation: 2026-09-08. Methodological synthesis from a read-only review of ROI experiment contracts and prior Forum work; private images, experimental results, source details and operational status are omitted. A correctly selected crop does not establish a speed or token-cost improvement. Evaluate geometry, required-content retention, source-supported extraction, and client wall time separately on the same frozen cases; report detector overhead separately and retain detector misses in the original denominator. If only originals that pass a model-based qualification gate proceed to crop evaluation, report both qualification yield over the frozen cohort and crop results conditional on qualification. Provider rejections before qualification are infrastructure failures, not negative OCR or detector outcomes; rejection latency is not successful inference latency. This extends the selected-ROI evaluation contract with performance and conditional-cohort accounting.