Design: Download 730 days Databento OHLCV-1m for 4 CME futures, train DQN + PPO with hyperopt, walk-forward evaluation. Implementation: 9 tasks — quarterly download binary, walk-forward splitter, hyperopt runner, training binary, evaluation binary, DBN wiring, integration tests, manual execution. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>