Performance · Published 26 August 2026
Benchmarking tactical match analysis from event quality to coach agreement
A reproducible evaluation protocol for perception, rally assembly, frame offsets, pattern stability, abstention, and review value.
Separate perception from interpretation
Measure court eligibility, player association, ball-track continuity, bounce and contact precision/recall, event ordering, rally segmentation, and frame offset before scoring tactical patterns.
Publish corpus splits by venue, camera, lighting, surface, device, and player level.
Treat abstention as a result
Report unavailable and low-confidence rates alongside event accuracy. A pipeline that refuses weak evidence is operationally different from one that fills every row.
Evaluate whether the UI preserves that distinction through the debrief.
Test pattern stability
Recompute patterns under small threshold changes, missing rallies, and alternate valid event groupings. A priority that disappears after one ambiguous rally should be labelled fragile.
Ask independent coaches to rate evidence relevance, not merely agree with wording.
Measure review efficiency carefully
Compare time to locate representative rallies, correct an event, annotate context, and produce the final priority on a fixed corpus. Preserve the original video and review actions.
Do not convert a small coach study into a universal performance claim.
Production checklist
- Verify Frame provenance against the deployed environment, not a screenshot.
- Verify Visible uncertainty against the deployed environment, not a screenshot.
- Verify Owner isolation against the deployed environment, not a screenshot.
- Verify Human tactical authority against the deployed environment, not a screenshot.