Replay Engine Model

Replay re-evaluates recorded execution decisions against policy. It is useful for incident review, policy simulation, and before/after comparisons.

What replay does

Replay loads recorded action context and runs policy evaluation again:

recorded ActionEvent + selected policy -> new verdict

This helps answer operational questions:

• Would a proposed policy have blocked this action?
• Which historical requests change from allow to deny?
• Why did a request require approval?
• Did a policy update broaden access unexpectedly?

Before and after example

Recorded request:
  capability: payments.transfer.production
  payload: {"amount": 5000}

Policy v1:
  amount_gt: 1000 -> require_approval

Policy v2:
  amount_gt: 2500 -> deny

Replay result:
  decision changed from require_approval to deny

Determinism

Given the same recorded event and same policy, evaluation should produce the same decision. Replay does not re-run the model or reconstruct hidden chain-of-thought. It re-runs policy against recorded execution context.

Limits

Replay quality depends on what was recorded. If raw payload storage is disabled or an adapter omits important fields, replay cannot recover that missing context.