{"resourceId":"thames-valley-lfr-deployment-register-202609","versions":[{"version":"external-38f27ee5c2eca806bf672915d3e02c6b9aec2440acb8f6a62ff2b3dc658280a3","resource":{"id":"thames-valley-lfr-deployment-register-202609","title":"Thames Valley register exposes the gap between facial-recognition activity and impact","organization":"Thames Valley Police","sector":"Public safety","geography":"England, United Kingdom; limited U.S. transferability","publishedAt":"Undated live HTML; latest listed deployment September 4, 2026","publicationDate":null,"eventDate":"2026-09-04","sourceName":"Thames Valley Police","sourceLabel":"Government operator deployment monitoring, not causal evaluation","sourceUrl":"https://www.thamesvalley.police.uk/police-forces/thames-valley-police/areas/sd/stats-and-data/live-facial-recognition-deployments-and-results/","evidenceClass":"government-evaluation","outcomeClass":"mixed","topics":["infrastructure","data-security","governance-procurement","operating-model"],"finding":"Operator logs report deployment activity with variable alerts and disposals, without a counterfactual crime-reduction evaluation.","sledRelevance":"New-to-archive recent operational data illustrates what an agency can disclose and what local independent evaluation must still establish.","evidence":"September 4 Wycombe HTML records 33,792 faces seen, eight alerts, zero arrests and six disposals at threshold 0.64. The linked PDF is marked updated September 1 and excludes this deployment.","architectureImplications":"Interpretation: Version watchlists, thresholds and deployment records; reconcile web and downloadable reports from one governed dataset.","governanceImplications":"Interpretation: Treat deployments, alerts, confirmed identities, actions and justice outcomes as separate measures.","securityPrivacyImplications":"Interpretation: Minimize biometric processing, restrict watchlist changes and test retention/deletion independently.","caveats":"Faces seen are not established unique people. No causal baseline, demographic error analysis or September 4 alert adjudication is provided. HTML and PDF disagree on older entries; no totals or disputed figures are used. UK legal authority does not transfer to U.S. agencies.","streamIds":["public-safety"],"roles":{"sales":"Interpretation: Police command, oversight boards, procurement and privacy teams need an evidence-based answer to what deployment achieves. Ask how an alert becomes a verified lead, what alternative deployment would cost and whether data distinguish unique people from repeated observations. A bounded engagement could design an outcome and disclosure scorecard before any local technology decision. The value hypothesis is more defensible evaluation, not a promise of arrests or crime reduction. British operator records cannot establish U.S. authority, demographic fairness or a transferable return on investment. Establish community and legal constraints before discussing product fit.","engineering":"Interpretation: Fit is an auditable monitoring pipeline around an authorized deployment. Link camera sessions, watchlist versions, thresholds, alert reviews and action outcomes without treating matches as independent authority to act. Prerequisites include reliable clocks, data definitions, lawful watchlist inclusion and controlled reviewer access. Test duplicate observations, mistaken identities, stale entries, interrupted connectivity and mismatches between report formats. A proof of value should independently adjudicate sampled alerts, examine missed matches where a valid reference exists and measure operating workload. Hosting choice requires local latency and biometric-data analysis; these logs do not validate any architecture.","delivery":"Interpretation: A deployment owner should work with privacy, community oversight and data-quality staff to establish definitions, training and incident review. Reconcile HTML and downloadable outputs before routine publication. Dependencies include watchlist governance, reviewer capacity, retention decisions and a usable complaints route. Proposed acceptance includes matching figures across formats, traceable disposition of all pilot alerts and successfully tested outage and deletion procedures. Set local error and workload thresholds before deployment, then stop or revise use if they are exceeded. Risks include counting activity as impact, overlooking people repeatedly observed and importing foreign operating practices without legal review."},"retrievedAt":"2026-09-08T03:01:11Z","enrichedAt":"2026-09-08T03:02:12Z","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Interpretation: Offer accessible HTML and consistent downloadable data; train operators on uncertain matches and public questions.","procurementImplications":"Interpretation: Require auditable error adjudication, version exports and reconciliation checks, rather than relying on aggregate arrest marketing.","operatingModelImplications":"Interpretation: Deployment command owns lawful action; independent oversight reviews errors and data quality.","sourceVerification":{"openedUrl":"https://www.thamesvalley.police.uk/police-forces/thames-valley-police/areas/sd/stats-and-data/live-facial-recognition-deployments-and-results/","referenceExcerpt":"The threshold settings for all deployments shown in the table is 0.64.","promptVersion":"sled-research-v3.1","model":null,"basis":"agent-reported inspection"}}}]}