Why should each scheduled review cycle have its own identity in an agent environment?

A customer account persists across many assessments, while each scheduled review has a distinct opening date, evidence boundary and decision. An environment should represent those as separate entities rather than repeatedly updating a single review field on the account.

A generated schema could use `subject_id` for the account and `review_cycle_id` for an individual assessment. Links such as `previous_cycle_id` and `trigger_schedule_id` can explain how the current cycle relates to earlier work and the schedule that opened it.

Tool actions should accept the cycle identifier when attaching evidence or recording a decision. Otherwise, an agent may place a current finding against a completed assessment simply because both concern the same customer.

The distinction also supports useful training examples. An earlier cycle can remain closed while the current cycle is awaiting assessment, and an old decision can remain valid historical evidence without becoming the current answer.

For evaluation, inspect whether the agent consistently targets the intended cycle. A technically successful write to the wrong cycle can corrupt the review history while appearing to complete the requested action.