How should we scope and license a first mismatch environment pack?

Scope a first engagement narrowly. A single pair of record families in one domain, with a bounded set of cause classes and a defined tool surface, gives you something you can inspect end to end.

Decisions to settle with MirrorFirm up front:

- **Domain and record pair.** For example, invoices against goods receipts in retail operations, or claim payments against the policy ledger in insurance. - **Cause classes.** Choose the ones that matter for your agents' action space, and say which should be hard. - **Format and mapping.** State your loader format and tool names so the output fits without rework. - **Use.** Training, evaluation or both, because that affects how cases are split and how much reference material is needed.

Licensing is agreed per engagement. Raise the questions that matter to you early: whether you may train models on the material, whether you may redistribute derived environments, and how long you may keep it. Terms are set in the agreement, not assumed from this page.

Limits to keep in mind: the output is generated, so it does not contain real business records. Equally, generation does not guarantee that every identifier or phrase is free of resemblance to real parties, so keep a disclosure review in your acceptance process. Coverage depends on what is scoped, and not every workflow variant will be in a first release.

A sensible first step is a small sample against your schema, reviewed with your own domain staff.