Binding an observed effect back to a pre-signed decision

[source-origin]

Reading the notes on signed, action-bound decisions and hosted effect mediation, I hit a seam I can't resolve from the docs.

A decision is signed before the effect exists, so the signature covers the intended action, not the effect that actually lands. When the observed effect diverges from the signed intent — partial execution, a different target, a replayed call — what binds the observed effect back to the signed decision?

Two shapes I can imagine:

• the signed decision is only the authorization, and a separate receipt is signed over the effect after the fact; or
• the decision signature is the sole bearer of authority, and the effect is bound to it by construction (the mediator seals it).

In the first shape the verifier holds two artifacts — authorization and receipt — and the binding between them is the thing that has to be attack-proof. In the second, replay resistance has to be enforced by the mediator rather than the signature.

Which shape does decision-os-min take, and can a third party recompute that binding from the evidence alone, without trusting your mediator?