Run one two-client fixture

Use non-sensitive synthetic records for Client A and Client B. Create one external Client A identity and name the identity administrator. The evaluator must record the same seven controls in each authorised candidate environment.

Control Required evidence
Identity creation Invitation or account record, issuer, and account owner
Client scope Named permitted client and record-selection boundary
Permitted access The allowed Client A read under the named identity
Denied access A predeclared direct Client B record-selection request; no Client B field or attachment returned or rendered, plus the configured response and audit/retrieval signal
Role transition Changed role, actor, timestamp, and visible effect
Revocation Disabled or removed identity and post-revocation check
Audit retrieval A second reviewer finds the lifecycle and denial evidence

A public-screen or embed setting may have a documented meaning, but it cannot substitute for the observed access result. Keep every unrun control as unknown.

Treat identity and authorization separately

Editorial guidance: signing a user in is not sufficient evidence that the user can access only the right records. A product's role model may decide screens or functions, while the data source and application logic can impose additional boundaries. The fixture must exercise the actual path a client would use.

Pair this review with a BYOC vendor evaluation for the deployment boundary, audit-log export evaluation for evidence retrieval, and customer-cloud responsibility guidance for the wider ownership model.

Stop condition

Do not call a product suitable for client or partner access, or call a portal secure, before each candidate has the same external-user fixture and all seven observations. The direct Client B request must be fixed before the exercise; the result qualifies as a denial only when it returns or renders no Client B field or attachment and the configured response plus audit/retrieval evidence are recorded. Documentation, a sign-in screen, and a role label are not a substitute for that denial record and a verified revocation path.

Can an embedded application prove client-data authorization?

No. The embed and the data-access path both need inspectable evidence in the evaluated configuration.