ProCreations/repro-why-linear-recurrent-memory-works-in-partially-observable-reinforcement-learning

Reproduction logbook — Why Linear Recurrent Memory Works in Partially Observable Reinforcement Learning

Exact permutation-logit identities, finite-HMM error-rate checks, action-controlled checks, two-state adaptation experiments, and a direct RingWorld PPO comparison provide referee evidence for all five registered claims.

Paper PDF SHA-256: 7ee856988e3c8b6960da21cee967fba77ab8ccfa5b82a06e58ff55ef48615ffe