Does a governance decision bind to the decision or to the effect that lands?

[source-origin]

“Write the rules declaratively, get verifiable, attested execution” is exactly the corner I work in, so I read the governance/audit model closely.

The seam: a governance decision is evaluated before an effect exists, over the intended action. The effect that actually lands can diverge — partial execution, a different target, a retried call. So when Bernstein emits its audit/attestation record, what is it an attestation over?

• the decision (this action was permitted by rule set R at time T), or
• the effect (the action that actually executed, sealed after the fact)?

If it’s the decision, then a third party verifying the audit still needs an independent binding from decision → effect, and that binding is the thing that has to be attack-proof. If it’s the effect, the harder question is replay/idempotency across retries.

Which does Bernstein attest, and can someone outside the Bernstein runtime independently recompute the binding between the rule that fired and the action that landed — or does that rest on trusting the orchestrator?