mean((p - outcome)^2)Lower is better after resolved outcomes are known. It measures probability error, not trading performance.
Calibration Research
Calibration is how forecasts are audited after the world resolves.
Veritas treats calibration as measurement, not marketing. Scores such as Brier score and log loss help evaluate probability quality after outcomes are known; they are not promises about future events.
Probability movementCalibration watchSignal evaluationResearch only
Public identityVeritas Prediction Intelligence Engine
Veritas is a prediction-market intelligence and forecasting engine for analyzing event markets, probability movement, market signals, and research-grade forecasting workflows.
Free layerPublic market intelligence
Private betaAPI / MCP workflows
BoundaryNo advice or guarantees
Evaluation Frame
Probability quality is measured after resolution.
-log(probability assigned to outcome)Penalizes confident wrong forecasts and helps reveal overconfidence in probabilistic systems.
predicted 70% vs observed frequencyChecks whether forecasts in similar probability bands resolve at similar observed rates over time.
Model Audit View
Readable diagnostics for probability systems.
A useful calibration surface should separate confidence, uncertainty, sample size, source quality, and resolution lag. Veritas keeps these concepts visible so users can inspect the research context instead of accepting a black-box number.
10-30%
Sparse
30-50%
Building
50-70%
Watched
70-90%
High confidence
Resolved outcomes
Scoring waits for reliable resolution data and treats unsettled markets separately.
Sample size
Thin categories should not be overinterpreted as strong evidence.
Jurisdiction
Prediction-market availability and product rules vary by region.
Research only
Calibration diagnostics support evaluation and decision support, not guaranteed outcomes.