Which features make this workflow worth including in a model training corpus?

Use this checklist to decide whether a candidate case contains enough operational substance for model training.

- A rule changes a permitted action or required condition. - Work already exists in more than one processing state. - The affected population requires a reasoned selection. - At least one dependency sits outside the policy document. - An incorrect choice produces a visible business consequence. - The correct response includes an actionable next step.

These features distinguish a propagation task from a document comparison exercise. A retail returns amendment that merely replaces a contact address may offer little reasoning depth. A rule that changes whether an opened item needs inspection can alter queue routing, stock availability and the customer response. The model has to connect the rule to those consequences.

Not every case needs maximum complexity. A narrow task can isolate one reasoning step, such as identifying which processing states still permit an inspection request. More demanding cases can combine several dependencies once the simpler behaviour is understood.

MirrorFirm could generate a progression of cases using these selection criteria. A buyer can then choose training material according to the business reasoning it exercises, rather than the length of the policy text.