Skip to content
MirrorFirm

What should we specify when requesting generated sufficiency cases?

A checklist of the choices a buyer should settle before generation, covering case types, labels, defects, multi-turn behaviour and document form.

Before asking MirrorFirm for generated cases, settle these points so the release matches the decisions you want to test.

- Case types and domain, such as mortgage applications, warranty claims or supplier onboarding - The requirement sets, and how they change by amount, region or customer type - The decision labels you need: proceed, request, escalate, or your own vocabulary - Defect axes to include, and the share of clean cases to keep in the mix - Whether substitutes and waivers should appear, and under what conditions - Multi-turn behaviour: full replies, partial replies, late replies and silence - Document form: structured fields only, extracted text or rendered files - Output formats and the field names your harness already expects

If you have internal policy text, say whether it can be supplied as an input or whether generic rules should be used instead. Generic rules are quicker to inspect but may not match the thresholds your reviewers know. Either way, treat the first release as a draft to inspect, and expect to adjust the specification once you see how the cases read.

Deciding these points up front keeps the release matched to the decisions you want to test and avoids rework after the first sample.

Audience:
Evaluation teams
Content type:
Checklist
Buyer question:
What can be generated

Content release: r-a220eb6e099b243b

More on Document sufficiency