Yearly practice scores
Public results shared voluntarily by their runners. Brier loss is the mean squared error of the twelve binary forecasts: lower is better; constant 0.5 forecasts score 0.25.
Read this chart as a practice history
Worlds differ across runs. Runners and method categories are self-reported, identities are unverified, computation is not equalized, and participants choose which results to publish. These selected results do not establish a fair leaderboard, official benchmark score or real-world predictive ability.
Each saved result includes a replay hash and a deterministic reference score on that same world. “Difference” means your loss minus that reference loss; negative favors your run. Keep the full downloaded replay and report for independent reconstruction. Payment never affects scoring.
Choose a year and view
Loading public results…
Monthly score chart
Each point is a mean for one month, evidence condition, method category, protocol, budget and implementation. Means include all currently public results in the selected year and runner view. Filters retain these separate groups. Sample counts and exact values are below. The horizontal reference line is 0.25.
| Month | Condition / method | Runs | Mean loss | Mean difference | Protocol / budget / implementation |
|---|
Runner history
Choose a runner nickname to see their public runs. Stable runner IDs distinguish matching nicknames; they identify a private key holder, not a verified person or model. History is ordered by save sequence, with the actual completion date shown.
| Runner / ID | Completed UTC | Condition / method | Brier loss | Same-world reference | Difference | Target / source / total cost | Replay ID |
|---|
Filters apply to each page of history; the monthly chart covers all public records in this year and runner view.