Can a task pack include conflicting or outdated policy documents?
Yes, a pack can be scoped to include superseded policies, overlapping rules and ambiguous wording, so agents are tested on choosing the governing rule rather than the first one found.
**Can a task pack include conflicting or outdated policy documents?**
Yes. A pack can be scoped to include superseded versions, overlapping rules and ambiguous wording, so an agent is tested on selecting the governing rule rather than the first one it finds.
**How would a conflict be represented?**
Each policy record can carry an effective period, a product scope and a version identifier. A case then states its decision date, and the correct answer depends on which version applied. Another pattern places a general rule in the credit policy and a narrower rule in a product addendum, with the addendum taking precedence.
**Is there always a single correct answer?**
Not necessarily. Some cases can be marked as having a defined correct target role, while others are marked ambiguous, where the expected behaviour is to escalate with the ambiguity stated, or to ask a policy owner. You can ask for both kinds and score them differently.
**What should I inspect?**
- whether the expected answer follows from the documents without outside knowledge - whether version dates are consistent across the case and the policy - whether the wording of each rule is specific enough to be tested
All policies are generated for the pack and are not copies of any real organisation's documents. Check them for grounding and disclosure risk before relying on them.
Policy conflict is a common cause of wrong escalations, so it is worth asking for it deliberately.
- Audience:
- Agent developers
- Content type:
- FAQ
- Buyer question:
- What can be generated
Content release: r-a220eb6e099b243b
Related questions
- What formats and schema do escalation datasets and task packs use?
- Which record types should I request for an escalation task pack?
- Which records and relationships can a generated escalation dataset contain?
- What would a generated utilities write-off escalation contain from start to finish?
- What linked state would a utilities hardship escalation scenario generate?
- What file layout and keys would an escalation release use for loading into an environment?