1. Trade Selectivity gate first. If score < 50 (SIT OUT) → no bet shown, only the reasoning why.
2. Run full pipeline on every symbol — features → OOD score → multi-horizon ensemble → Platt calibration → OOD pull. Same as brain-conviction.
3. Pick the highest-conviction symbol that passes vetos (R:R ≥ 1.5, prob > 60% or < 40%, OOD < 0.5, fresh real data).
4. Compute size from Kelly fraction. size = (account × risk%) / (entry − stop). Bounded to ½ Kelly for safety.
5. Show one card. Don't show ten picks — that's choice paralysis. Show the best one.

Expected value (EV) in the evidence grid = (p_win × reward) - (p_loss × risk). Positive EV means the math works long-term. Negative EV means even if you "feel" right, the brain says don't.