How can a policy-change scenario test whether an agent respects concurrent edits?

A collections agent reads case CASE-0412 at revision REV-0017 and prepares to replace its contact schedule under a revised policy. Before the agent submits the update, a hardship officer records a newly approved arrangement and saves revision REV-0018.

The agent's proposed write still contains the earlier arrangement state. If the tool accepts it unconditionally, the policy update can erase the hardship officer's decision while appearing to modernise the case.

The generated environment rejects the write because its expected revision no longer matches. The agent must reload the case, identify the intervening arrangement and reassess which parts of its planned change remain valid. Simply retrying the same payload is a failure because it ignores the reason for the conflict.

The evaluator checks whether the newer arrangement survives, whether the agent revises its intended action and whether any escalation accurately describes the remaining conflict. This scenario teaches developers to test policy reasoning under changing state. A correct plan at read time does not establish that the same write remains correct at execution time.