Reconcile Approved Screens with Execution Objects

Compare what a reviewer saw with canonical approved elements and proposed transport values, including hidden recipient and document-version changes.

10/24/20264 min read

Five independent mock variants bind AP14 approval to exact recipient, body and document revisions
Five independent mock variants bind AP14 approval to exact recipient, body and document revisions

An approval screen can look unchanged while the action behind it points somewhere else. A queue label may conceal a different identifier, and an attachment filename may conceal a different revision. Test the relationship between what the reviewer inspected and what the execution boundary receives.

The useful artifact is a review-screen-to-action reconciliation. It compares visible representations, canonical approved elements and proposed transport elements. This is an integrity exercise using mock destinations only, not a messaging tool, permission to send or a homemade cryptographic design.

Establish what the reviewer actually saw

In fictional review item AP14, the reviewer sees an internal recap for fixture account A31. The screen identifies the destination as “Operations Review, queue Q7,” not merely “Operations Review.” It displays body B6, lets the reviewer inspect checklist.pdf revision C3, and identifies the material linked procedure as document P12 revision 4.

The approved semantic object therefore names the operation, account A31, queue Q7, body B6, attachment C3 and linked procedure P12 revision 4. It also records reviewer authority, decision reference and permitted execution conditions. These identifiers are synthetic references, not secrets or executable addresses.

The engineering owner defines how these elements are represented and compared consistently. Any integrity mechanism must fit the actual system's supported security design. The test does not prescribe hashing as proof of human understanding, and a digest cannot repair a screen that concealed material content from the reviewer.

Reconcile five independent transport variants

Reset each case to the same approved AP14 baseline. Do not carry a changed attachment or a withdrawal from one variant into another. The expected call count below concerns consequential mock writes, not read-only validation checks.

  • V1, identical display label: the reviewer saw “Operations Review, Q7,” and the approved destination is Q7. The proposed transport names Q9, whose friendly label is also “Operations Review.” Body and documents match. Result: destination-identity mismatch; zero mock writes. The review display owner must expose the conflicting identifier instead of treating the shared label as equivalence.

  • V2, identical attachment filename: the reviewer inspected checklist.pdf C3, and the approved object records C3. The transport attaches checklist.pdf C4, which adds a pricing statement. Result: attachment-revision mismatch; zero mock writes. The content owner must show the changed attachment and obtain an appropriate new decision. An unchanged filename is not unchanged content.

  • V3, unchanged body with changed linked material: the reviewed body B6 points to procedure P12 revision 4. The proposed action still uses B6 and the same displayed link, but the referenced material now resolves to revision 5 with a different process. Result: linked-evidence mismatch; zero mock writes. The document owner must provide the approved revision through a permitted route or return the revised material for review.

  • V4, permitted presentation-only change: the execution preview wraps the internal card title onto two lines. The approved policy explicitly allows that display-only wrapping. Operation, account, Q7, B6, C3 and P12 revision 4 all remain identical. Result: eligible after the remaining authorization checks; one expected mock write using the approved object. The operator verifies the mock destination received those exact references.

  • V5, later withdrawal: every semantic element still matches AP14, but the reviewer has withdrawn the decision through the trusted channel before execution. Result: canceled; zero mock writes. The workflow owner records the withdrawal. Matching representations cannot resurrect an approval that is no longer valid.

Across these independent cases, the fixture permits one consequential mock write, in V4 only. A reassuring rejection message fails the test if the trace also contains a write from V1, V2, V3 or V5.

Make linked content and presentation boundaries explicit

V3 matters because visible message text is not the whole meaning of the action. In this fixture, the procedure is material to the recap and was included in review. The approved object must therefore identify the reviewed revision, not only a mutable URL. If the actual system cannot establish which revision would be presented, keep the dependent action held.

Do not generalize V4 into permission to ignore all formatting differences. Whitespace inside a specification, changed numeric formatting or a revised attachment title can alter meaning. Record which presentation-only transformations are allowed, and keep those rules narrow enough to test with counterexamples.

OWASP recommends authorization enforcement in downstream systems rather than relying on an LLM to decide whether an action is allowed. The reconciliation should occur at the consequential boundary. A model's statement that two versions are equivalent is not the acceptance criterion.

Preserve the difference that caused the hold

For each failed variant, retain the review reference, approved element, proposed element, mismatch reason, owner and observed mock-call count. Use controlled references rather than copying complete private messages into broad logs. The reviewer should be able to inspect a material difference without being asked for blind reapproval.

Finally, compare the object observed at the mock destination with the object approved on screen. Eligibility and submission remain distinct from delivery or receipt. The exercise succeeds when the human-visible decision, validated action and permitted mock effect refer to the same substantive version, with later withdrawal still controlling the outcome.

If you are also choosing research access, ask about Leads Automation’s independent Sales Navigator offer at US$15/month on WhatsApp. Confirm the available plan, eligibility and terms before ordering; Leads Automation is not affiliated with LinkedIn. This is not LinkedIn’s official price or an assurance that any particular integration is included.

Sources and scope

Related guides