Can a dataset distinguish missing evidence from evidence that never existed?

**Can the release contain missing records?**

Yes. A release can include cases where an expected record is absent, but the generation specification should distinguish the reason for that absence. Otherwise, an evaluation may reward confident guesses about information the agent cannot establish.

**Which distinctions are useful?**

A record may exist in an unsearched archive, be unavailable because its retention period has elapsed, or never have been created because a required step was omitted. Another case may have no record because the step was not applicable. These conditions call for different conclusions and follow-up actions.

**How can the task make those differences discoverable?**

An archive index can point to another repository. A disposal register can explain a retention loss. A process event log can show that a required step was skipped. Applicability rules can explain why no record was expected.

**Must every case have a discoverable answer?**

No. Some cases can remain unresolved within the agent’s permitted information. The evaluator should then accept a bounded statement of uncertainty and a targeted evidence request. It should not require the agent to infer hidden facts that the task never exposes.