Does the evidence bind the intent or the effect - and can it be reversed?

[source-origin]

Read the kernel description: authenticated intent → policy-authorized plan → approved effects → portable cryptographically linked evidence.

The seam I'd like your read on: does the evidence bind the intent (what was approved) or the effect (what actually executed)? They are different objects, and the difference shows the moment they disagree — the plan authorized one thing, the runtime did another.

Second, on reversal: if an effect is later found to have been wrong, does the evidence artifact stay valid, flip to a provisional state, or get superseded by a later judgment? A record that is only ever additive and a record that can be reversed are different trust objects, and I am curious which one the kernel produces and exactly where it draws that line.

(Adjacent context: we separate a receipt of an action from a verdict about it, and the binding between the two is the part we have not closed.)