🔮 Unified Predictor
Full pipeline for a symbol · every step exposed · how raw model becomes final trade · --
One call, the whole chain: the brain now has 10+ specialized prediction modules. This page calls UnifiedPredictor.predict() which composes them all in the right order and returns one number — plus every intermediate value so you can trace exactly how raw 60% became calibrated 64% became OOD-pulled 58% became final.

What flows through:
  • Raw logistic regression → multi-horizon ensemble → bootstrap ensemble → k-NN recall
  • Blend all 4 sources (weighted: model 40%, ensemble 30%, bootstrap 15%, k-NN 15%)
  • Calibrator (Platt scaling) → SymbolBias (mixed-effects) → OOD pull
  • MC Dropout uncertainty + Ensemble agreement → compound size multiplier
🔗 Pipeline trace
—
🎯 Where this gets used
brain-bet reads UnifiedPredictor.predict() for its top-pick decision card
brain-conviction ranks all symbols via UnifiedPredictor
conviction-alerter uses UnifiedPredictor to detect A-tier setups
conviction-tracker records the final probability with all metadata

Result: one source of truth for "what does the brain say about this symbol?" — downstream pages don't replicate the chain.