Skip to main content
Record schemas define the structured data model. Candidates are proposed records or field changes generated from Vault evidence, Workbench activity, workflows, or other configured sources.

Schemas

A record schema defines:
  • Record type name.
  • Fields and field types.
  • Required and optional fields.
  • Relationships to other record types.
  • Evidence expectations.
  • Review rules and ownership where configured.
Examples:

Candidates

TextMine can suggest candidates from source evidence. A candidate may propose:
  • A new record.
  • A field value for an existing record.
  • A relationship between records.
  • A changed value.
  • A missing-field recommendation.
  • A duplicate or merge candidate.
Candidates should be reviewed before acceptance so records remain trustworthy.

Candidate review flow

  1. Generate or discover candidates from documents, tags, entities, tables, or workflows.
  2. Group candidates by schema, record, field, or source document.
  3. Inspect the proposed value and confidence where available.
  4. Review the source evidence.
  5. Accept, reject, edit, or defer the candidate.
  6. Record the review outcome and keep the evidence link.

Evidence

Each record field should remain connected to the documents and extracted evidence that support it. Evidence can include:
  • Vault document IDs and names.
  • Extracted default or custom tags.
  • Entity extraction outputs.
  • Table rows or line items.
  • Source passages and page numbers.
  • Image tags or embedded image regions.
  • Workflow run outputs.
  • Reviewer decisions.
Use evidence links and audit events to understand where a value came from and when it changed.

Schema design guidance

Good schemas are specific enough to be useful but not so rigid that every document creates exceptions.
  • Start with the business object users already understand.
  • Keep fields stable and reviewable.
  • Use relationships for connected objects instead of packing everything into one record.
  • Prefer controlled fields where reporting depends on consistent values.
  • Keep source evidence mandatory for high-risk or regulated fields.

Agent guidance

Agents should read schemas before creating candidates. They should avoid inventing fields when an existing schema can be used, and should ask before accepting, deleting, or bulk-updating records.
Last modified on June 6, 2026