The Fire Extinguisher Is an Inspection Instrument: Charge-Blinded Randomisation in 41,208 Rooms, a 1.24-Million-Room-Year Instrumented Cohort, and a Standing Directive on Attendance

A fire extinguisher hangs within thirty feet of most people who will read this. Almost nobody has ever seen one used, and the reason is not that fires are rare: the agent inside the cylinder makes no contribution to whether a room catches fire. We linked service records to ignition records for 61,412 rooms across 1,243,900 room-years and modelled room-level ignition against two exposures carried together — whether the unit’s service tag was current, and how much agent the unit held. Attendance predicted ignition strongly (adjusted hazard ratio $0.19$, 95% CI $0.16$–$0.23$). Charge mass did not predict it, and the dose–response curve is flat from an empty cylinder to a full one. The association survives instrumentation on a 2014 redraw of contractor depot catchments, which moved rooms between inspection circuits for reasons of road geometry alone. We then randomised 41,208 rooms to charged units, to mass-matched inert units that cannot be told apart from them by lift, gauge, seal, or paperwork, or to a bracket and a service tag with no cylinder at all. Over 181,315 room-years the arms were indistinguishable (hazard ratio $1.02$ for inert versus charged; pre-specified non-inferiority met). The cylinder is a scheduling device. What protects the room is not the cylinder but the fact that somebody came.