Foundation release
Verified Forecast Registry
A public evidence layer for comparing what was forecast, when it was published, whether its price was executable, and how it performed against reality and the same market baseline.
Registry
Provider-supplied claims remain distinct from VS Predicts calculations.
| Entry | Evidence | Decided | Sample | Ranking state |
|---|---|---|---|---|
| VS Predicts VS Predicts is the operator of this registry and is subject to the same published evidence and sample rules. | certified feed | 6 | insufficient sample | Not ranked |
| No-vig market consensus This is a planned comparison baseline, not a commercial forecasting provider or a current ranking. | verified integration | Not yet evaluated | Not yet evaluated | Not ranked |
Three scoreboards, never one opaque score
Evidence completeness, probabilistic forecast quality, and market/performance diagnostics answer different questions and remain separate.
evidence quality
- pregame timestamp coverage
- price or line coverage
- settlement coverage
- complete history coverage
- comparable closing line coverage
- source provenance coverage
- correction count
probabilistic forecast quality
- brier score
- log loss
- expected calibration error
- reliability by probability bucket
- incremental skill vs same market baseline
market and performance diagnostics
- mean and median clv
- positive clv rate
- standardized flat stake roi
- maximum drawdown
- average price and break even probability
- price availability after publication
- abstention selectivity
Sample maturity rules
| State | Minimum | Maximum | Ranked |
|---|---|---|---|
| insufficient sample | 0 | 24 | No |
| provisional | 25 | 99 | No |
| ranked developing | 100 | 299 | Eligible if evidence is complete |
| validation sample | 300 | 499 | Eligible if evidence is complete |
| mature sample | 500 | No maximum | Eligible if evidence is complete |
Entry and correction rules
- Use identical event, market, and cutoff cohorts for direct comparisons.
- Keep live published records separate from backtests.
- Do not rank selectively observable or incomplete cohorts.
- Do not recreate prices after an event.
- Do not rank by win rate alone.
- Apply the same methodology to VS Predicts.
- Preserve immutable, reproducible scoreboard snapshots.
Initial third-party entry requires a complete certified feed or an authorized verified integration. A dispute or correction creates a visible new state and snapshot; prior public history is not silently rewritten.