Skip to content
MirrorFirm

What records and relationships can a generated sufficiency dataset contain?

A dataset can contain policies, requirement lists, case files, documents, decisions, requests and replies, linked so each judgement can be traced to its evidence.

A generated release for this workflow is built from linked records rather than isolated documents. A buyer can ask for the following.

- **Policies and requirements.** Versioned rules stating which items a case type needs, with conditions such as applicant type, amount threshold or region. - **Case files.** A case with its parties, product, status and timeline. - **Documents.** Each with type, issue date, signatory, entity, condition (clear, partial, expired, wrong entity) and content in text or structured fields. - **Decisions.** The gold outcome of proceed, request or escalate, with the gap list and the clause behind each gap. - **Requests and replies.** Messages to the customer, broker or internal team, and the replies that resolve, partly resolve or ignore the request. - **Exceptions.** Waivers, supervisor overrides and conflicting evidence.

The relationships matter most. A gap links to a requirement, a requirement to a policy version, a document to a case, and a request to the gaps it covers. That chain lets a grader ask not only whether the decision was right but whether the cited reason was.

The documents and cases are generated. They are informed by direct observation of how real businesses handle these decisions, but they are not real records. Inspect a sample for grounding and disclosure risk before relying on any release.

Linked records let a lab train and grade the reasoning behind each decision, not only the final label.

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

Content release: r-a220eb6e099b243b

More on Document sufficiency