Replay

Replay reconstructs run behavior from trace and state evidence. It is inspect-only by default.

What replay can show

• Percepts received during a run.
• Policy outputs and proposed actions.
• Constraint and verifier decisions.
• Action outcomes and denial reasons.
• State commits and snapshots.
• Message causality and parent/child run links.
• Work-order, approval, circuit-breaker, policy, and safety facts where present.

What replay does not do by default

Replay does not:

• call filesystem, HTTP, database, shell, webhook, artifact, or device adapters;
• re-run policies or verifiers as fresh authority;
• re-send messages;
• contact approval or revocation services;
• install or clear governance state;
• mutate live state heads;
• control physical simulators or devices.

CLI shape

splendorctl replay \
  --db <trace-path> \
  --state-db <state-path> \
  --run <run-id>

Replay validates trace ordering, run scope, trace event identity, integrity metadata where available, and referenced state snapshots before reconstructing a run.

Failure behavior

Replay fails clearly when trace or state data is missing, corrupt, out of order, wrongly scoped, or incompatible with the requested run.