A public, automatically updated record of what an automated forecasting system predicted and how those
predictions scored once the questions resolved. Every score is printed beside the simplest possible baseline -
always answering 50% - so a reader can see at once whether the system adds anything. As of 2026-11-04.
Graded forecasts on public markets
Worse than always answering 50% (Brier 0.2823 against 0.2500).
Rows scored
67
Brier score (lower is better)
0.2823
Baseline: always 50%
0.2500
Baseline: the market
not available - no signed record of the market's price at the moment each forecast was made exists yet, so the market cannot be scored on these rows.
Graded forecasts, independently graded only
Worse than always answering 50% (Brier 0.2823 against 0.2500).
Rows scored
67
Brier score (lower is better)
0.2823
Baseline: always 50%
0.2500
Baseline: the market
not available - no signed record of the market's price at the moment each forecast was made exists yet, so the market cannot be scored on these rows.
Daily market panel: the market's own price at capture
No scored rows yet.
Rows scored
0
Brier score (lower is better)
-
Baseline: always 50%
-
Baseline: the market
not available
How to read this
The Brier score is the mean squared distance between the forecast probability and what happened (1 or 0).
Always answering 50% scores exactly 0.25. Lower is better.
Only questions on public prediction markets are shown; 6 internal planning forecasts are
scored privately and not listed. Markets that resolved to a partial probability instead of yes or no cannot be
scored that way: 1. Graded rows refused because their records did not check out:
19.
Daily panel questions appear only after their capture day has closed, their test round has finished and the
market has settled.
Each row carries the content hash of the record it was built from, so any number here can be traced back to
its source.