Skip to content
MirrorFirm

Can a task pack include conflicting rules, missing data and exceptions?

A direct answer on how a pack can deliberately include policy conflicts, incomplete records, out-of-authority requests and cases where the correct action is escalation rather than change.

**Can a pack include messy cases?**

Yes, and a useful one should. Clean cases alone teach a model to proceed. Real handling often depends on noticing that something does not add up.

**What kinds of mess can be generated?**

- **Rule conflicts:** a product rule allows a longer term while a hardship policy caps it, and the pack states which takes precedence. - **Missing data:** a request arrives without the evidence the policy requires, and the right action is to ask for it. - **Out-of-authority requests:** a front-line officer asks for a waiver that only a manager can approve. - **Stale information:** a notice describes an older schedule version than the one now in force. - **Competing events:** a payment lands the same day a plan is changed. - **Requests to decline:** a change would break a statutory or contractual limit, so the correct end state is no change plus a recorded reason.

**How is the right answer marked?**

Each task carries an expected outcome class, such as act, escalate, request more information or decline, along with the end state for the act case.

**Is this real behaviour?**

The cases are generated, informed by how exceptions arise in business operations. Review a sample to judge whether the mix of exceptions matches what your use case needs.

Packs can include messy cases on purpose, with the correct response marked, so a model is tested on judgement and not only on clean inputs.

Audience:
Model labs
Content type:
FAQ
Buyer question:
What can be generated

Content release: r-a220eb6e099b243b

More on Repayment scheduling