R2Orch Uses Recursive Self-Play to Improve Agent Orchestration

R2Orch employs a dual-loop self-play mechanism where an inner loop improves orchestrators on self-chosen tasks and an outer loop optimizes the improvement process. The system uses a Recursive Learnability objective to prevent advantage collapse during training.