British Transport Police LFR pilot: location and watchlist determine yield

Original records and their limits: BTP's public deployment register currently lists only eight operations from 11 February–26 March 2026, including one false alert on 24 February at King's Cross (19,313 estimated faces, watchlist 455, threshold 0.64) and no arrests in the listed rows. BTP's pilot page says the trial began 11 February, expanded to Underground stations in August and runs through November 2026; watchlists include wanted people and those with ancillary court conditions, alerts receive officer review, unalerted templates are immediately deleted, and LFR CCTV footage is kept 31 days. A police-authored register of eight rows does not verify an entire six-month trial. Link this to the deployment-value question, not to Met arrest totals.

Original investigative FOI reporting, full FOI response not public here: Liberty Investigates and the Guardian, 29 September 2026 obtained a BTP FOI response and report 18 operations, over half a million faces scanned, £320,786 equipment hire plus staffing and ~100 police hours in February–July 2026, one alert that was false and no arrest directly triggered by LFR. These full figures depend on reporters' account of a response not independently inspected in this note; BTP's published subset matches the early one-false-alert pattern, not all 18 rows or cost. BTP told the reporters officers made associated arrests for unrelated observations during deployments, which must not be recoded as either LFR matches or proof of zero utility of extra officers. This early pilot was subsequently extended, so do not claim the whole 2026 trial ended with no true alert.

Why the disagreement is informative: BTP says watchlist size was initially deliberately limited while data governance and locations were refined; original register's lists have ~381–465 names in February–March. A former biometrics commissioner argues catch yield depends on list, people actually passing, location, and time; a nearly zero direct yield says something about this choice of location/list/window, not that software is incapable of face matching. Comparing BTP's zero direct arrests with the Met's 962 requires police hours, list eligibility, passenger mix, selection and the alternative use of officer time. The BTP 2026 pilot evaluation is promised after completion; inspect that before claiming a cost-effectiveness estimate.