Accuracy

How our reference probabilities score

Market pages show a reference probability computed from public data beside the market price. This page scores both against how each contract settled, and publishes the result whichever does better.

Weather: daily high temperature

CheckpointEventsContractsBrier, referenceBrier, marketReference closerMarket closerNo quote, skipped
24 hours before close1937500.1860.12633%66%4,188
1 hour before close2033580.1090.08665%35%4,926

Calibration, 24 hours before close

  • Reference probability
  • Market price
  • Forecast matches what happened

Calibration, 1 hour before close

  • Reference probability
  • Market price
  • Forecast matches what happened

How the reference is calculated: weather method.

Crypto: price at a set time

CheckpointEventsContractsBrier, referenceBrier, marketReference closerMarket closerNo quote, skipped
24 hours before close1772,0840.1000.10965%35%4,449
1 hour before close1852,8480.0290.03288%12%4,425

Calibration, 24 hours before close

  • Reference probability
  • Market price
  • Forecast matches what happened

Calibration, 1 hour before close

  • Reference probability
  • Market price
  • Forecast matches what happened

How the reference is calculated: crypto method.

How this is measured

  • For every settled contract that had a reference probability, we record the reference and the market price 24 hours and 1 hour before the market closed, and whether the contract settled yes or no.
  • The market price is the last trade, else the midpoint of the bid and ask. A contract with no quote at a checkpoint is skipped for both sides and counted in the last column.
  • The Brier score is the average squared difference between the forecast and the result (1 for yes, 0 for no). Lower is better: 0 is perfect, and always saying 50% scores 0.25.
  • "Closer" is the share of contracts where that side's forecast was nearer the result; the rest were ties.
  • Calibration groups forecasts into 10 bands of 10 points and plots the average forecast in each band against how often those contracts settled yes. A well calibrated forecast sits on the diagonal.
  • An event counts once even when it has several contracts (one per temperature bracket or price level), because its contracts settle together. Scores are recomputed daily.

A reference that scores worse than the market is a finding about the reference, and it is shown the same way. These figures describe past data, not future results, and none of this is a recommendation to trade. Not financial advice. See the methodology and the responsible trading page.