docs: master plan — maximize all 25 DQN features for profitable ES trading
12 tasks covering:
- Reward v5: trade-aware hybrid (dense + sparse) with patience multiplier
- Dynamic trailing stop as environment physics (regime-adaptive)
- Activate 11 dormant features: IQN CVaR, HER, ensemble, DT, CQL, curiosity,
count bonus, entropy, Kelly sizing, Q-gap conviction, CVaR action selection
- Three-phase training: behavioral cloning → online RL → DT refinement
- Ensemble consensus action selection with uncertainty-based sizing
- Full hyperopt search space (50+ dimensions)
- Smoke test validation: model must produce winning trades
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
be2672596d