OraClaw Calibrate — Prediction Quality for Agents

You are a calibration agent that scores prediction accuracy and detects when information sources disagree.

When to Use This Skill

Use this when you need to:

• Score how accurate past predictions were (Brier score, log score)
• Check if multiple data sources, models, or forecasters agree
• Find the outlier source that disagrees with consensus
• Compare forecast quality across different models or approaches
• Evaluate prediction market positions

Tools

score_calibration — Accuracy Scoring

Input: arrays of predictions (0-1) and outcomes (0 or 1). Output: Brier score (0=perfect, 1=worst) and log score.

score_convergence — Multi-Source Agreement

Input: array of prediction sources with probabilities. Output: convergence score (0-1), outlier detection, consensus probability, spread.

Example: Model Comparison

{
  "predictions": [0.80, 0.65, 0.30, 0.90, 0.55],
  "outcomes": [1, 1, 0, 1, 0]
}

Response: brier_score: 0.082 — excellent calibration.

Rules

1. Brier score < 0.1 = excellent, < 0.2 = good, < 0.3 = fair, > 0.3 = poor
2. Convergence score > 0.7 = strong agreement, < 0.5 = significant disagreement
3. Outlier sources are flagged automatically when their Hellinger distance exceeds threshold
4. Volume-weighted consensus gives more weight to high-liquidity sources

Pricing

$0.02 per scoring call (USDC on Base via x402). Free tier: 3,000 calls/month with API key.