RunSpec 2.0 off-policy runtime Accepted rollouts are durable before acknowledgement; only a checkpointed contiguous WAL frontier may bepruned. Resolve once Actor · local process orremote game host Authenticateddurable ingest Learner process parse record action rollout submit persist, then ACK apply in order capture; pruneafter save activate nextepisode RunSpec 2.0 strict YAML · explicitcomponents Resolve contracts seed · imports ·fingerprint Immutable manifest map · geometry ·feature/action IDs Frozen actor policy one version perepisode Game adapter telemetry → features→ transition Durable actor spool retry until accepted Loopback gRPC token · size ·fingerprint SQLite WAL accepted rowsordered by ID Applied frontier contiguous replaycommit Replay + sampler uniform · PER ·sequence Typed batch n-step · masks ·IDs Learner update validate →backward → step Checkpoint state · RNG · WALfrontier Policy snapshot portable tensorstate Run evidence JSONL · W&B ·artifacts Invariant: acknowledged chunks remain recoverable until the exact applied frontier is inside adurable checkpoint. PPO is a separate local on-policy lifecycle; it never enters this actor/WAL/replay path.