Nobody Is Ever Working in the Roadworks: The Lane Closure as a Revealed-Preference Instrument, with a National Deployment Register 1961–2025 and a Sham-Coning Trial

There is never anyone working in the roadworks. Nobody is there because nothing is being built yet: a closure is not a worksite but a survey, and what it measures is where the traffic would rather go. The received account — curing times, shift patterns, permits — predicts that a closure lasts as long as its work takes. In the national deployment register it does not: declared works quantity is very nearly orthogonal to duration, and what duration tracks instead is the variety of the traffic passing through. That is the sample-size requirement of a survey, not of a job. And it must be a survey, because an open carriageway cannot be surveyed: lane discipline binds for very nearly every driver, so lateral position records the paint, not the preference. Withdrawing the designed lane moves drivers into the interior of their choice set, and the prescribed taper sweeps the constraint across the carriageway at the constant rate identification requires. We recover the desire field from telemetry the cars already keep. The next scheme on the link paves it at a gain of $0.83$ (95% CI $0.79$–$0.87$), a median of $8.2$ years later, which is why nobody has connected the two. A gain below one leaves a residual, and the residual is the next closure. Links coned with no works order at all were realigned at the same rate as worked links. The road is not being repaired. It is being read.