Public questions awaiting their deadlines.
We score the misses, too.
WorldPredicta launched its first public forecast cohort on July 30, 2026. No forecast has resolved yet, so any claimed accuracy number would be fake.
Not enough outcomes to judge performance.
Published after the first forecast resolves.
Requires a meaningful set of resolved probabilities.
CALCULATION INTEGRITY6/6 probabilities reproduced
EXACT INPUT COVERAGE83% across the first cohort
VALIDATION BOUNDARY2 component diagnostics · 0 full-model backtests
Audit every input and weight →AUTONOMOUS CALIBRATION TRACK
The system now creates evidence while it sits.
Separate one-day weather trials exercise the complete unattended loop: issue before the event, lock the probability, wait for delayed observations, score the result and learn a bounded location-specific correction.
- Pending trials
- 72
- Resolved trials
- 228
- Trial Brier score
- 0.0632
- Learning state
- assessable
HOW SCORING WORKS
A probability should be judged as a probability.
Brier score measures the squared distance between a forecast probability and the eventual yes-or-no result. Lower is better. Calibration asks whether events forecast at roughly 70% actually occur about 70% of the time.
Never delete a miss
Resolved questions remain in the archive with their original probability and evidence.
Use the first published result
Each forecast states whether later data revisions can change its resolution.
Separate domains and horizons
Performance will be reported by topic and time horizon, not hidden inside one flattering average.
Record model changes
When a miss changes the model, the version and reason will be published.
No resolved forecasts yet.
The six strategic forecasts have not reached their deadlines. The separate adaptive-trial ledger can accumulate short-horizon outcomes without pretending those results validate geopolitical or market forecasts.