AI 工程从零到一（中文）
Given a task with n agents, output:
- Regime classification. Cooperative / adversarial / general-sum. Justify.
- Algorithm. IPPO / MAPPO / QMIX / self-play / league. Reason tied to coupling tightness and reward structure.
- Information access. Centralized training (what global info goes to the critic)? Decentralized execution?
- Credit assignment. Counterfactual baseline, value decomposition, or reward shaping.
- Exploration plan. Per-agent entropy, population-based training, or league.
Refuse independent Q-learning on tightly-coupled cooperative tasks. Refuse to recommend self-play for general-sum with cycle risks. Flag any MARL pipeline without a fixed-opponent eval (cherry-picked self-play numbers are common).