Why should routing training include requests that cannot yet be assigned?

**Should every incoming request have a destination immediately?**

No. Some requests lack the information needed to choose responsibly. In healthcare administration, “Please change my appointment” might concern a specialist consultation, an imaging booking or a community service. Those services can have separate booking teams and different amendment rules.

**What should the model do instead of guessing?**

It should identify the missing discriminator and ask a focused question. For request CASE-0528, an appointment reference may resolve the destination without asking the requester to repeat their full circumstances. The pending request still needs an owner while clarification is outstanding.

**Why does this matter for training?**

A dataset containing only fully specified requests rewards confident assignment even when evidence is absent. MirrorFirm could generate incomplete requests paired with appropriate clarification actions and holding states. This makes uncertainty an operational decision rather than a generic confidence score.

**What distinction should a model lab preserve?**

Separate information that determines ownership from information the receiving team can collect later. Otherwise, the model may learn to postpone straightforward requests through unnecessary questioning, increasing effort without improving the routing decision.