A self-scoring public forecast · Dublin
Dublin Bikes Forecast
Twice a day, a frozen model publishes the probability that each Dublin Bikes station will have a bike, and a free dock, at the next commute window. Every forecast is committed to a public git ledger before its target time exists, then scored against what actually happened. The track record cannot be edited, selected, or quietly restarted; the git history would show it.
Pre-registered gate: after 28 scored days, the model must beat both baselines — station climatology and persistence — at the overnight horizon, with day-clustered bootstrap 95% confidence intervals excluding zero, for bikes and for docks. Anything less is published here as NOT PROVEN, in this same box. The gate was committed before any data was collected.
§1 Will there be a bike?
Loading the live forecast feed…
§2 How it works
A cron job records every station's state each 10 minutes from the public GBFS feed.
At 07:00 and 16:00 Dublin time, the frozen model turns current state, trend, calendar and weather into probabilities.
Forecasts are appended to the ledger and pushed. The git timestamp is the before-the-fact proof.
Nightly, matured forecasts meet reality. Gaps and excluded stations are counted, never hidden.
§3 Evidence
Loading…
§4 Calibration
Each dot is a bundle of forecasts at one confidence level, plotted against how often the event actually happened. An honest forecaster's dots sit on the diagonal.