Observed rainfall against real HIRDS v4 depths, with the TARP position for the current event and every past storm on the same axes.
MET Norway forecast rainfall for the next 120 hours, screened per gauge and tracked across the TARP plot so you can watch it approach a damaging combination.
The delivered engine. Close any road, route the detour, see what severs and what isolates. Every other dashboard stands on this.
Slope instability separated from surface flooding. Long-memory antecedent drives the ranking; the lag between the two mechanisms is made legible.
Forecast-red links pushed into the routing engine on a button press. User-asserted by design — an automatic display would read as a prediction of closure.
A watch list, not a scour model: bridges on catchments under a severe forecast, cross-referenced with scour history and the weight-restriction schedule.
The week-after artefact: rainfall, peak ARI by gauge and duration, antecedent state and event window, formatted for an NZTA Emergency Works application.
Damage register and the WC141 claim record geocoded to links, so rankings cite the council's own cost history instead of a model's opinion.
Ford crossings as their own instrument — passable, marginal or impassable, rather than a rainfall depth. The question a grader driver actually asks.
The other way this network fails. Shaking intensity against the same links, feeding the same closure engine so a quake scenario routes detours as a storm does.
The join that turns the map from interpolated to real: forecast.json, 29 timesteps, flood and slip scores on 1,200–1,600 selected links.
Every storm adds a row to the calibration set, so the contours on the TARP plots tighten each year. This is the recurring value, and the reason it is a subscription.