{"resourceId":"crang-broadcast-ai-tracking-validity-2026","versions":[{"version":"external-c01a586f3b8742ba177ae92fe407d3e33a6837b2b3f85208acac28c91b7d6856","resource":{"id":"crang-broadcast-ai-tracking-validity-2026","title":"Broadcast tracking accuracy depends on detection coverage and local conditions","organization":"Crang and colleagues","sector":"Collegiate athletics","geography":"International professional football; 2022 Qatar World Cup data","publishedAt":"May 5, 2026","publicationDate":"2026-05-05","eventDate":null,"sourceName":"Journal of Sports Analytics","sourceLabel":"Peer-reviewed concurrent-validity study","sourceUrl":"https://journals.sagepub.com/doi/10.1177/22150218261445834","evidenceClass":"academic-research","outcomeClass":"mixed","topics":["infrastructure","data-security","governance-procurement","operating-model"],"finding":"Tactical feeds achieved 89–96% player detection, versus 36–64% for programme and camera 1 feeds; undetected frames produced poorer accuracy.","sledRelevance":"Transferable validation design for collegiate video procurement, not evidence from NCAA teams.","evidence":"Three providers were compared with TRACAB Gen 5 using 27 outfield players in one match. Methods used RMSE, bias and mixed models across camera feeds and resolutions. Table 1 distinguishes detection from estimated coverage.","architectureImplications":"Interpretation: preserve detected-versus-inferred flags and acquisition metadata through every downstream calculation.","governanceImplications":"Interpretation: approve intended uses by validated metric and condition rather than granting blanket model approval.","securityPrivacyImplications":"Interpretation: license footage and derived data explicitly and control identifiable player exports.","caveats":"Single stadium, match and broadcaster in favorable conditions; goalkeepers excluded. Data are not public because of rights restrictions. No collegiate outcome or prospective effectiveness test.","streamIds":["college-athletics"],"roles":{"sales":"Interpretation: The customer problem is buying apparently complete tracking outputs without knowing which observations are trustworthy. Engage the video coordinator, performance analyst, coaching staff and procurement. Ask what decisions require precise movement measures, which camera feeds are available and whether estimated positions are distinguishable from observations. A bounded vendor-comparison trial could establish fitness for one local use. The value hypothesis is avoiding unsuitable purchases and reducing manual correction, not guaranteed savings. Do not generalize professional-stadium results to every campus or claim any unnamed provider is best. Include rights and reviewer workload in the proposed scope.","engineering":"Interpretation: Fit is an evaluation harness before integration with tactical dashboards. Secure licensed representative footage and a defensible local reference measurement. Preserve timestamps, coordinate transformations, identity corrections and detected-status fields. Test across venue geometry, uniforms, occlusions and camera changes; evaluate aggregate metrics separately from frame-level error. Establish task-specific thresholds with analysts before examining vendor results. Hosting selection must account for video transfer, retention and access controls; the paper does not compare deployment costs. Validate export completeness and version changes. Keep generated narrative explanations outside the proof of value unless their factual grounding is separately assessed.","delivery":"Interpretation: Assign a video-analysis owner with support from data engineering, procurement and coaching. Inventory feeds, obtain rights, annotate test segments and document the reference method before starting a trial. Dependencies include representative conditions and protected analyst time. Train staff to reject unsupported inferences and escalate identity mismatches. Governance checkpoints should review trial coverage and proposed uses before operational approval, then repeat targeted checks after vendor changes. Proposed acceptance criteria include explicit missingness flags in every tested export, documented error by condition and metric, and passage of locally approved thresholds. Risks include unrepresentative test footage, drift and false confidence in aggregate summaries."},"retrievedAt":"2026-09-10T03:01:34Z","enrichedAt":"2026-09-10T03:03:18Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: train analysts to identify missing observations and present uncertainty accessibly.","procurementImplications":"Interpretation: require local trial footage, metric-level error reports and disclosure of imputation before contract acceptance.","operatingModelImplications":"Interpretation: make the video-analysis lead accountable for ongoing quality checks and vendor-change review.","updateExplanation":"URL and related paper absent from the 151-resource full archive scan. Newly adds direct measurement validation to existing exploratory video coverage; publication predates the last run.","sourceVerification":{"openedUrl":"https://journals.sagepub.com/doi/10.1177/22150218261445834","referenceExcerpt":"Another limitation is that goalkeepers were excluded from the analysis","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}