Review an agent’s sources and actions as separate decisions.

Separate the evidence from the proposed action

Consider a fictional assistant that reads a procedure and drafts a request to update a project record. The procedure may explain what information belongs in the record. That alone does not establish who may change it, which record should change or whether the proposed update has been reviewed. Those are separate questions, even when one interface presents reading and acting as a continuous conversation.

Start by describing the intended outcome in ordinary language. In this exercise, a person wants to prepare a correction to a synthetic project note. The assistant can identify the relevant passage and propose wording. A reviewer then compares the proposal with the source and checks the specific destination. This separates an explanation of the change from the decision to apply it.

A useful review makes the connection between the source and the draft visible. The reviewer should be able to see the passage that supports a statement, understand its scope and recognize where interpretation begins. A document might describe a general process while leaving an exception unresolved. The draft should preserve that uncertainty instead of silently converting a general instruction into permission for an exceptional case.

Next, describe the proposed action independently of the generated explanation. Name the destination, the change to be made and the information that would be sent. A confident summary of the source cannot stand in for those details. The person reviewing the action needs enough information to judge whether the proposal matches the request and whether anything outside the request would also change.

For a discussion exercise, create two short review notes. Keep both tied to the same fictional task, but give each a distinct question to answer:

• Source review: identify the supporting passage, its scope and any unresolved condition in the proposed explanation.
• Action review: identify the destination, the exact proposed change and the authority needed to apply it.

Make the boundary visible in the handoff

Original illustrative material. This fictional exercise makes no partnership, certification, deployment or security-effectiveness claim.

The distinction matters most when a proposal moves between steps. A draft can be ready for discussion without being ready to apply. A source can be relevant without being sufficient. Use clear state descriptions so that another person can tell whether the work is a suggestion, a reviewed proposal or a completed change. Do not make readers infer that state from a success-colored message or fluent wording.

Imagine that the synthetic project note changes while the proposal is being reviewed. The earlier draft may now describe the wrong version. The exercise should pause at that mismatch and show what needs another look. Reusing a previous decision without checking the current destination would hide the very condition that changed. Record enough context to explain why the proposal is being reconsidered.

An unsuccessful attempt also needs an accurate description. The interface should distinguish a change that was never submitted from one whose outcome is unknown. In the fictional exercise, a reviewer can compare the displayed state with the prepared scenario and ask what information would resolve the uncertainty. Avoid turning an ambiguous result into a completed-work claim.

Before the exercise ends, make one final handoff note that another reader can understand without replaying the whole conversation:

• Record what was reviewed, what remains unresolved and whether any change was actually applied; keep the supporting source and destination identifiable.

About this exercise

All documents, records and actions described here are fictional. The article supplies a way to discuss the difference between understanding information and authorizing a change. It does not document an integration, certify a control or report the effectiveness of a security product. Any implementation would need its own requirements and verification in its actual setting.

The intended result is a clearer account of the decision, including the option to leave it unresolved. If the source does not establish a necessary condition or the destination cannot be confirmed, the exercise should retain that gap. A readable explanation is useful when it helps a person inspect the basis of a proposal and decide what needs attention next.

Continue exploring

Use the original evaluation and prompting guides to practise making evidence, interpretation and next steps explicit in a reviewable draft.