Distinguish a source unit from a unit inferred by the reviewer

Make unit provenance explicit

Keep them distinct so the next person can tell what was reported and what still needs confirmation.

Before paying for technical-document extraction, inspect how the source meaning survives in the output. ParseMill’s public material-register example gives you a synthetic PDF and downloadable evidence to compare. Bring your own document type and acceptance checks to a scope discussion; the self-service invoice plan is a separate offer.

Use a small vocabulary such as explicit in cell, confirmed heading, confirmed note and unresolved. Reserve inferred for a separately labeled hypothesis that has not yet become accepted source data.

Use a unit-provenance worksheet

These invented rows show why a single fill-down is unsafe. A3 has a different source-supported unit from A2, while A6 has no confirmed unit. Preserve the reviewer’s proposed unit in an exception note if useful, but do not present it as reported.

NIST: SI notation checklist stresses clear value-unit association. Your register should make that association visible even when the source places the unit outside the individual cell.

Confirm the applicable source range

1. Inspect the result cell and its heading, then locate any unit note. Record each candidate source location before assigning a unit.
2. Determine the range governed by the heading or note, including section changes and page continuations. Mark unclear boundaries for review.
3. Assign the unit with an evidence-basis label and source reference. Keep the original unit spelling or symbol in a raw field.
4. Review rows immediately before and after every unit change. Check that a copied or filled value did not cross the verified scope.

W3C: PROV Data Model distinguishes provenance entities and activities. Record the source evidence separately from the reviewer activity that assigned context to a row.

Do not confuse plausibility with evidence

A magnitude that looks reasonable in millimeters does not establish that the source used millimeters. The document could use centimeters, a scaled value or a different quantity. Treat plausibility checks as flags for investigation rather than conversion authority.

If two source locations conflict, preserve both and request clarification. Do not prefer the cell or heading automatically without understanding whether one is a general label and the other an explicit exception.

Check the unit assignment can be reproduced

A second reviewer should be able to follow each confirmed assignment back to the cell, heading or note and agree on its scope. For the fixture, A6 remains unresolved and A3 retains its distinct unit. Record any later clarification as a revision rather than an invisible overwrite.

Use the unit-provenance worksheet to define requirements for a scoped supplier-report workflow. Inspect the worked example first, then identify which headings and notes your real documents require. Automatic interpretation of those relationships is not part of the invoice offer.