A perfectly calibrated brain is on the dotted diagonal: predicted = actual. Above the line = under-confident (could size up). Below = over-confident (gates should tighten). ECE (expected calibration error) summarizes overall miss.
📊 Reliability diagram (50%-100% confidence)
📋 Per-bucket detail
BucketPredictedActualΔ (Error)NWins
🔗 Related
📈 Original reliability diagram ⏪ Backtester 🎯 Brier Skill