Every forecast this system publishes is recorded in an append-only ledger before the outcome is knowable, then scored against what actually happened. This page is the public aggregate view: hit rates, Brier scores vs honest baselines, resolve-by dates for what's still open, and the exact pre-registered rules that decide outcomes. Nothing here is curated after the fact.
Brier = mean squared error between predicted probability and realized outcome; lower is better. 0.25 = the score of always predicting a coin-flip. The base-rate baseline is what a forecaster scoring only the observed up-rate would get.
Forecasts bucketed by predicted probability (10 equal-width bins). A calibrated forecaster's realized up-rate matches its predicted probability in every bin. Small-n bins are labeled, never hidden.
Different asset classes have different base rates; per-class is the only honest comparison. All-time and trailing-90-day slices shown side by side. Skipped (data-unavailable) claims are excluded from these numbers and reported separately below.
Pending forecasts and when they resolve. Forecasts are scored by a daily cron once their horizon expires; they cannot be withdrawn.
Some claims can never be scored because market data is unavailable (delisted symbols, unmappable crypto pairs, provider gaps). They are annotated and published — never deleted, never counted as hits or misses.
The pre-registered resolution rules, verbatim from the resolution engine. These rules are fixed before any forecast is scored.
A small sample of the most recently scored forecasts (newest first). Full aggregate definitions ship in the API payload at /api/track-record.