Little minds.
Large worlds.

The swarm is allowed to lose.

These six experiments follow a single chain: local coordination → shared prediction → bounded adaptation → robustness → physical constraints → the minimum intelligence needed. Controllers and budgets are declared, failures are retained, and the full source and seeded results are inspectable inside each world.

All environments are synthetic. The three-dimensional sketches illustrate declared models; they do not certify biology, robotics, ocean engineering or general intelligence. Actual model responses, where available, are separated from rule-based simulation.

群体可以输，失败必须留下。六个实验从局部协作、共享预测、有界适应、抗冲击、物理约束，一直追问到最低智能需求。每个世界都有方法、源码与可重播证据。