{"resourceId":"byu-athletics-video-api-project-update-2026","versions":[{"version":"external-cfa902fb7b96c82d391c0661907ddd340d848077712bf8f2e6b54286b990e17b","resource":{"id":"byu-athletics-video-api-project-update-2026","title":"BYU project deck exposes the implementation work behind sports-video AI","organization":"BYU-hosted Data-Driven Athletics / Sports Research Institute project","sector":"Collegiate athletics","geography":"United States; Utah, including a reported Weber State case","publishedAt":"2026; exact publication date unknown","publicationDate":null,"eventDate":null,"sourceName":"Data-Driven Athletics: AI Athletics & Outreach","sourceLabel":"University-hosted technical project update; operator claims","sourceUrl":"https://apm.byu.edu/prism/uploads/Projects/dde_athletics_Q2Y2.pdf","evidenceClass":"vendor-claim","outcomeClass":"mixed","topics":["knowledge-work","developers-agents","infrastructure","data-security","accessibility-workforce","operating-model"],"finding":"The deck describes an implemented video-analysis pipeline and coaching interface, but does not establish causal athletic improvement.","sledRelevance":"Direct university-linked athletics development; includes outreach beyond college sport, which is not treated as collegiate evaluation.","evidence":"Appendix: 96.9% step-detection accuracy using stratified five-fold cross-validation. Main slides name MLP as best; appendix names HistGradientBoosting. A Weber State slide claims 13% boys' and 22% girls' top-speed increases over October–April without sample size, control or causal baseline.","architectureImplications":"Reported stack includes FastAPI, PostgreSQL, Celery, Docker and video models, with a Claude coaching layer. Interpretation: validate measurement and generated advice separately.","governanceImplications":"Interpretation: freeze the evaluated model and dataset version before approving a coaching use; reconcile conflicting model descriptions.","securityPrivacyImplications":"Interpretation: restrict identifiable footage and prohibit secondary training without an approved purpose; test deletion across derived files.","caveats":"Operator evidence uses vendor-claim as the available category. Athlete-level validation split and independent replication are unreported. PDF screenshots failed; extracted text supported inspection. No injury-reduction conclusion is established.","streamIds":["college-athletics"],"roles":{"sales":"Interpretation: The customer problem is the effort required to turn practice footage into usable feedback. Engage the coaching lead, athletics analyst, campus IT and athlete representatives. Ask which measurements influence decisions, how manual review performs today and whether existing camera workflows already suffice. A bounded engagement could validate one movement metric on authorized footage before adding advice generation. The value hypothesis is reduced preparation effort at acceptable measurement quality. Include annotation, correction and support costs in discovery. The deck does not justify promised speed improvements, injury prevention or a claim that a named institution is seeking a supplier.","engineering":"Interpretation: Treat this as a candidate architecture, not a validated product specification. Map approved video ingestion to a job queue, versioned outputs and a coach review interface. Check camera geometry, calibration, input rights and retention before processing. A proof of value should hold out athletes and sessions, compare against independently annotated video and report measurement error and correction time. Reconcile the conflicting best-model descriptions with an immutable evaluation manifest. Assess the advice layer separately using a coach-approved rubric. Select cloud, local or hybrid placement from institutional requirements; the deck does not establish a compliant deployment boundary.","delivery":"Interpretation: An athletics analytics lead should own the pilot, with coaches adjudicating output quality and IT supporting deployment. Begin with a data inventory, consent and access review, annotation guidelines and a manual baseline. Train users to identify unusable video and record overrides. Proposed acceptance criteria include complete source-clip traceability, held-out errors within coach-approved tolerances, lower total review time and successful deletion tests. Review governance before expanding sports or adding sensitive health data. Document model changes and arrange support beyond student-project turnover. Risks include calibration drift, correlated training examples, unreviewed advice and apparent time savings erased by corrections."},"retrievedAt":"2026-09-12T03:00:49Z","enrichedAt":"2026-09-12T03:01:56Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: provide readable metric descriptions and train analysts to recognize calibration and annotation errors.","procurementImplications":"Interpretation: require data export, model-version documentation, maintenance ownership and full hosting costs before purchase.","operatingModelImplications":"Interpretation: budget annotation, technical support and coach adjudication as ongoing service work.","updateExplanation":"Absent from all 217 canonical archive records scanned at offsets 0, 100 and 200, including URL and related-title checks. Newly archived historical evidence; no new-since-last-run publication or update to an existing resource is claimed.","sourceVerification":{"openedUrl":"https://apm.byu.edu/prism/uploads/Projects/dde_athletics_Q2Y2.pdf","referenceExcerpt":"Compared 7 classifiers with stratified 5-fold cross-validation","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}