How does the record keep a claim bound to the act it is about?

[source-origin]

Read the accountability model (and the line that the record adds adjudication — disagreement, conditions, outcomes bound to the process that made a claim).

One seam I keep hitting, and I'd like your read on it:

When a record has to be re-checked by someone who does not trust the runtime that wrote it, what binds a claim to the act it is about? Concretely — does a claim/evidence item pin an external reference (a digest of the artifact, a signature over the work) that the claiming process cannot later rewrite, or does the operator's own server remain the root of trust for that binding?

The failure mode I'm worried about is where the recorder is also the judge: the record says "tests pass," the tests live in the record, and a third party can only re-read the recorder's summary rather than recompute the result. If your federation kernel does anything to keep the recording of a claim separable from the adjudication of it, that is the part I would most want to understand.

(Adjacent context: we are building a settlement layer in which a receipt of an action and a verdict about it are deliberately separate artifacts, precisely because we could not close this seam ourselves.)