Docs
Replay
Replay
Planned deterministic replay and reproducibility for Cascades runs—verification-grade re-execution.
Coming soon — Replay is not shipped yet. The capabilities below describe the target posture Cascades is building toward—they are not guarantees for your deployment today.
Goals
Modern automation needs more than immutable logs—you need confidence that evidence means what operators think:
- Deterministic replay — given the same enqueue-time snapshot, inputs, and adapter versions policy, reproduced runs arrive at reconcilable outcomes.
- Reproducibility — auditors can correlate proof payloads (execution proofs) back to deterministic execution timelines.
- Future verification — pair replay with drift detection so “what reran later” aligns with structural definitions hashed at enqueue.
How this differs from retries
Retries today rerun failed slices under live policy—they are recovery, not reproducibility proofs. Replay will deliberately bind to immutable snapshot material (immutable snapshots) and signed adapter metadata so auditors can differentiate operational jitter vs intent drift.