British Transport Police ran live facial recognition cameras across 18 deployments in London's busiest railway stations between February and July this year. Freedom of information data obtained by Liberty Investigates reveals the programme scanned more than 500,000 faces, cost taxpayers £320,786 in equipment hire and officer time, consumed nearly 100 hours of policing, and produced a single watchlist alert — which turned out to be wrong. No arrests resulted from the technology itself. The response to this performance was not a pause or review. BTP extended the trial for four more months and expanded it into London Underground stations. Transport for London endorsed the extension, framing the technology as a tool to "specifically tackle violence against women and girls" — a framing that attaches a sympathetic cause to a system that has not yet demonstrated it can identify anyone at all. BTP acknowledged the gap obliquely, noting that officers made "a number of associated arrests" during deployments — for assault, theft, weapons possession, and breach of court orders — but conceded these "did not result directly from an LFR alert." In other words, uniformed officers standing in stations made arrests the way uniformed officers standing in stations always have. The technology contributed nothing measurable. Former biometrics commissioner Fraser Sampson, now a non-executive director at commercial facial recognition firm Facewatch, drew an instructive distinction: in retail, success means deterrence (people not entering), while for police, success means identification and arrest. By the police's own metric, the trial failed completely. Sampson noted the technology "works" in the abstract but that operational variables — location, timing, watchlist composition, likelihood of targets being present — must all align. The civil liberties dimension is sharpening. Parliament's joint committee on human rights flagged LFR rollout as a "particularly clear example of risk" to human rights from AI. Labour MP Bell Ribeiro-Addy called for a suspension until proper legal frameworks exist. More than half of police forces in England and Wales have now deployed LFR, and the Metropolitan Police plans fixed cameras in London's West End and on the newly pedestrianised Oxford Street. The pattern is familiar in surveillance procurement: a pilot fails by its own performance criteria, the failure is reframed as a learning opportunity, and expansion follows anyway. BTP described its approach as "deliberate, controlled" watchlist growth while "ensuring data quality, governance and safeguarding arrangements were robust" — language that emphasises process compliance over outcome accountability. The system processed special category biometric data from hundreds of thousands of people who were never suspects, for zero operational benefit. What is being built is not a tested crime-fighting tool but institutional infrastructure for mass biometric surveillance, advanced incrementally through trials whose failure conditions are never defined in advance. Each expansion normalises the next. The question is not whether the technology can theoretically work but whether a system that has demonstrably not worked earns the right to scale — and who bears the cost of that scaling.