Can a task pack represent evidence that should exist but cannot be found?

**Can generated material include missing records?**

Yes. A task pack could deliberately omit a required record while retaining the policy that makes it expected. The task should also define the accessible sources, so the agent can describe the limits of its search.

**Is an empty search result enough to conclude that a step was skipped?**

No. An empty result may mean the record is absent, stored outside the accessible collection or indexed under a different identifier. A generated task can distinguish these conditions through source inventories, indexing notices and custodian responses.

**What additional evidence is useful?**

A record custodian's confirmation that no register entry was created is different from an agent failing to find one. Neither automatically proves every underlying business action did or did not occur.

Developers can request tasks where the correct output is a bounded statement: the specified sources do not establish completion. This teaches agents to express what the evidence supports, describe the search boundary and request further material without inventing a missing document.