Can a generated environment keep information hidden from the agent?

**Does the agent need access to every generated record?**

No. A simulated environment can separate underlying state from the information exposed to the acting role. The underlying state might contain the cause of an insurance administration error, while the complaint handler initially sees only the customer's account and a processing history.

**How can information become available?**

A valid investigation action can reveal an authorised extract, schedule a response from another role or return a refusal with a reason. Visibility rules can depend on role permissions, the request's scope and whether a prerequisite has been completed.

**What could MirrorFirm generate for this?**

A release could include hidden cause records, role-specific views, disclosure conditions and expected responses to evidence requests. Those components let builders create partial visibility without improvising answers during an episode.

The hidden cause should not become a requirement that the agent somehow guesses. Success should reflect what the agent can reasonably establish through permitted actions. Where available evidence remains inconclusive, a justified escalation or unresolved finding can be the correct outcome. The builder should keep evaluator-only fields outside every agent-visible export and tool response.