🚦 Trade Selectivity
When to trade · when to sit out · six independent meta-signals · refreshes every 10s · --
⭐ Conviction Picks
The principle: Even a great prediction model loses money if it trades every day. Some conditions destroy ALL strategies — extreme VIX, opening 15 minutes, post-news whipsaws, when the model has been cold lately. The brain needs to know when to not trade.

The math: Six independent meta-signals each contribute a 0-100 score. They're weighted (VIX and rolling accuracy dominate). The composite score drives the tier:
  • 80-100 ACTIVE — normal trading, all picks shown
  • 50-79 SELECTIVE — only A-tier picks, half size
  • 0-49 SIT OUT — brain refuses to publish picks at all
Brain Conviction reads this and behaves accordingly. So does any other downstream page that imports TradeSelectivity.
📊 Six meta-signals
🔬 Why these six
1. VIX (weight 25%): the dominant volatility regime indicator. Optimal trading range is 12-22. Above 25, prediction accuracy collapses for all strategies. Below 11, complacency creates fake setups.
2. Session timing (15%): the first and last 15 minutes of regular US hours are dominated by liquidity-seeking flow, not signal. Pre/post-market = different microstructure entirely.
3. Day of week (10%): Tue/Wed/Thu have the cleanest signal. Friday afternoon = weekend risk premium distorts everything.
4. Rolling accuracy (25%): the most important internal signal. If the brain has been 30% accurate over the last 20 calls, something has changed. Sit out, retrain, come back.
5. OOD ratio (15%): when many recent inputs are statistically far from training distribution, the model is in unknown territory. Confidence is fake.
6. Concept drift (10%): when the rolling-accuracy drift detector has flipped to "adapting" mode, the model is currently relearning. Don't trade through retraining.

Weights sum to 100%. They reflect that internal accuracy + market vol are the most critical proxies for "can the model trust itself right now."