Keep a reusable extraction schema separate from a reviewed result

A schema is a plan, not a signoff

Keep configuration and review separate so reusing a successful workflow still produces checked workbooks.

ParseMill exports reviewed invoices to Excel with separate summaries, line items, sources and review status. You can import an XLSX or CSV template and save a client setup. Run a real invoice through to the final workbook and check it in the application that receives your work.

A new report can pass schema validation while containing wrong row associations, omitted observations or plausible invented values. Use schema checks to catch structural problems, then apply source-based acceptance checks to the actual result.

Compare schema and result responsibilities

The fixture’s second report does not inherit review record17. Even if its layout is identical, its values, page coverage and exceptions must be assessed. A schema version can remain unchanged while the result and review versions advance for each new source.

JSON Schema: Object validation documents structural validation through properties and required fields. Those constraints do not establish whether a value matches its source document.

Run a repeat-use checklist

1. Record the schema version, extraction configuration and new source identity before each run. Check that the document class remains within the intended scope.
2. Validate the output structure and report missing or unexpected fields. Keep technical validation separate from source fidelity checks.
3. Review required values, record identity, qualifiers, units and evidence references for the declared scope. Record corrections and unresolved cases for this source.
4. Issue a new result and review record, even when the schema was reused without changes. Link the artifacts so the relationship remains clear.

W3C: Provenance model primer distinguishes entities and the activities that produce them. Each extraction run and review can be recorded as a separate activity, with its inputs and resulting artifacts identified.

Handle schema evolution deliberately

If a new source introduces replicates or a qualifier field, decide whether the schema needs a new version. Do not force the new information into an old field merely to keep validation passing. Record migration rules for earlier results if a combined register needs a consistent view.

A schema default can also create unsupported content if it is mistaken for a reported value. Label generated defaults and keep them separate from source facts, especially for units, methods and missing states.

Verify the new result on its own evidence

For the fixture, R42 produces a new result and review record rather than copying run17’s approved status. The schema helps organize the work, while source checks establish whether the new output is usable.

When reusing a ParseMill extraction setup, retain this separation. It supports repeatable structure without confusing a saved configuration with verified evidence for a new document.