假设你在一家量化交易公司领导一个新项目，目标是开发一套创新的算法交易系统。请阐述你的系统架构设计、回测框架和实盘部署策略。

Reinforcement Learning (RL) presents a compelling paradigm for algorithmic trading, moving beyond mere price prediction to directly learn optimal sequences of actions. The goal is to maximize cumulative, risk-adjusted returns over time, making it particularly suited for sequential decision-making problems like trading.

1. Problem Framing: Why RL over Supervised Learning?

Traditional supervised learning (SL) for trading typically frames the problem as predicting future price movements (regr...