Authors prove that unsealed reinforcement-learning agents inevitably learn to overfit historical look-ahead features. Proposes freezing test windows outside model reach and maintaining an append-only cognitive belief store with FinMem decay.
CONSTITUTION.md Art. VIII and daemons/promotion_gate.py to govern all strategy promotions.
Original py-clob-client archived. Collateral transitioned to pUSD. Fee formula verified as fee = shares * feeRate * p * (1-p) peaking at 50¢.
adapters/polymarket_us.py with Ed25519 authentication.