Beta · Probabilistic ForecastsBeta forecasts are probabilistic estimates. Availability depends on current data and model support.
The evidence behind SkyCrate
Model & data
Real coverage, measured performance, and a clear account of what the models can and cannot support.
Historical coverage
Provider universe: — assets · Production-approved models: —
Coverage is being audited. Missing history is never filled with invented prices.
Coverage describes the retained research archive. Public charts use a shorter serving window; missing periods are never synthesized. Raw SkyCofl archives are not downloadable.
Retained model registry
Legacy artifacts remain available for comparison. Live forecasts require the current validation and trading-cost gate; registry status alone does not authorize publication.
| Horizon | Model | Direction accuracy | Brier score ↓ | Calibration error ↓ |
|---|
Accuracy is a historical measurement, not a forecast confidence score. Lower Brier scores indicate better probability forecasts; a fair 50/50 forecast scores 0.25.
Does context improve the forecast?
Controlled experiments compare the same chronological development folds with different feature families. Results below retain their original experiment date and scope; they do not imply that later holdouts remain unopened.
No verified experiment summary is available from this backend.
Where the evidence stops
Historical association, model attribution, and causal evidence are different things. SkyCrate does not treat a price pattern as a guarantee.