Trust score starts at 100 and subtracts penalties for each warning signal:

• BSS < 0: -25 (brain is worse than baseline)
• BSS < 0.10: -15 (brain not clearly learning)
• Annualized Sharpe < 0: -25
• Annualized Sharpe < 1.0: -10
• Drift PSI > 0.25: -20 (concept drift detected)
• Covariate shift fired: -25 (input distribution shifted)
• Losing streak ≥ 5: -25 (deep tilt)
• Losing streak ≥ 3: -15
• Cross-method agreement = FRAGMENTED: -10
• Conformal halfwidth > 0.20: -15 (wide uncertainty)
• Hourly bucket < 20 samples: -5
• DoW bucket < 20 samples: -5

The remaining number tells Brandon how much to trust the brain's current setup. A clean state with adequate sample sizes lands at 95+. A brain hit by drift + losing streak + shift drops to 0 — that's the system telling you to step away.