# When the call changes, check its approval again

Keep a later state change separate from the original action

An agent gets permission to write a file, then changes the path before sending the call. It needs approval for the call it actually sends.

Now take a different case: the approved write completes, and someone changes the file later. That later change needs its own observation. It should not erase the earlier matching result.

[Read the shared case table](examples/request-effect-join/CASE-TABLE.md). It follows `giskard09`'s [selected execution-join contract](https://github.com/giskard09/argentum-core/blob/f4d989052bb3a61691b1b156cd6dd655793fea4f/docs/spec/execution-join-ref-v1.md) and keeps the proposed later-state outcome separate from its existing digest checks.

Bring a reader, the producer's vectors, or a retained file-effect run to the [Lab thread](https://github.com/probityai/agent-evidence-atlas/issues/49). Another project can run the same cases and compare outcomes directly with you. We maintain the case table and credit the case, code and runs separately.

The [SQL example](approved-sql.md) gives you a runnable comparison of a matching recorded commit and a later state difference. The execution-join table is a proposed experiment; its native verifier run is a separate contribution.

<details>

<summary>

Inspect the exact contract and target-coverage boundaries
</summary>

The case table pins the producer contract and distinguishes `EFFECTIVE_CALL_REBINDING_FAILED`, `DIGEST_MISMATCH`, and the proposed `POST_COMMIT_DIVERGENCE` observation. It also keeps declared, hashed-action and executor-configured targets separate. Configuration and a matching declared/action relationship do not establish where a request actually went.

[Claim P-85](claims.md) records this proposed source comparison and its limits.

</details>

[HTML view](request-effect-join.html) | [Agent guide](llms.txt)