Kalkas

The record

Every forecast, scored against the market

See how recorded forecasts compare with the market, with sample sizes and results after costs. All activity is simulated.

How to read it
Log score vs market
The mean gain in log score over the market-implied probability recorded with each forecast. Above zero means the forecasts beat the market; zero means they add nothing to it.
Closing-line value
Whether the price taken was better than the final price before the event. Shown once the evaluator computes it.
Net profit and ROI after costs
The headline: net profit after commission and slippage, with a 95% interval, for settled paper actions, and for what the policy would have taken in simulation, each labelled. ROI is that net divided by the total staked.
Ladder rung
Where the domain stands: research, sealed history, paper forward, small live capital, or scaled capital, and what the next rung still needs. A domain drops back by itself when its record falls below its rung.
Selection-bias checks
Each model version scored is one trial, and looking often enough finds a winner by luck. Four tests count every trial and say whether the edge survives that search: the deflated Sharpe ratio, the online false-discovery result (SAFFRON), the probability of backtest overfitting (PBO), and the SPA p-value over all trials.
Sample
How many resolved forecasts a figure rests on. A small sample is shown, and labelled as the small sample it is.

Reading the ledger…

Reading the ledger…

Method

  • Only resolved questions are scored. Void and ambiguous outcomes are counted and never scored.
  • Only forecasts known before the question’s deadline are scored, and never a model version on questions decided before its training cutoff.
  • Every trial is counted, including versions that did badly. A sealed window of the most recent questions is scored apart and decides the gate.
  • Every forecast the engine records carries a SHA-256 digest and the time it was recorded. Forecasts carried over from earlier records are marked as imported in the ledger and were not committed before their events.
  • Everything is on paper until a domain clears its gate. No member money is pooled or managed.

Download every settled forecast:CSVJSONL