A matching highlight can still be wrong

Inspect the value and its label together so source links reduce review work instead of creating false reassurance.

In ParseMill, select an extracted invoice field to inspect its source, correct the value and export the reviewed workbook. Try your own invoices as well as the interactive sample. Compare upload, checking and correction time with your current way of building the same spreadsheet.

Keep the source document version and physical page visible during review. The same coordinates in a revised PDF can point to different content, and an apparently precise box can be misplaced if the page was rotated or cropped.

Use a three-part highlight check

CheckSynthetic sourceExpected evidence
TextA7 result 12.0Highlight contains 12.0
Row associationA8 also reports 12.0Highlight remains on A7
Column meaningA7 limit is also 12.0Highlight is under Result, not Limit
VersionRevision 2 shifts tableReference resolves revision 2

The repeated values are intentional. They prevent a reviewer from treating text equality as proof of correct provenance. If a highlight covers both Result and Limit, record that its precision is insufficient for the intended cell-level check even if the number is visible.

Microsoft: Document layout analysis documents table cell indices and source regions in layout output. These are useful evidence coordinates, but their association still needs validation against the source.

Inspect the supporting context

  1. Open the exact source version referenced by the output. Confirm the physical page before examining the highlight.
  2. Read the record identifier, field heading and neighboring cells. Compare them with the exported row and field.
  3. Classify failures as wrong text, wrong row, wrong column, wrong page, wrong version or overly broad region. Keep these categories separate.
  4. After correction, repeat the check on the affected field and neighboring records. Save the reviewed output revision and source reference.

W3C: Provenance model primer explains provenance relationships. A source reference should support the particular derived field, not just demonstrate that a similar string appears somewhere in the document.

Handle merged and multiline cells

A merged heading may span several columns, while a wrapped result may occupy several lines. Inspect the complete cell and the applicable heading hierarchy. A box around only one line may omit a qualifier or unit printed beneath it.

If the tool exposes only a page reference, perform the row and column check manually and record that limitation. Do not describe the evidence as cell-level verification unless you actually resolved the cell and confirmed its association with the exported record.

Record what the highlight established

For the fixture, acceptance requires the A7 Result cell in revision 2, not either of the other 12.0 values. The review record should state the identifier and heading checked, allowing another person to reproduce the decision.

In the invoice trial, choose repeated amounts in adjacent invoice fields, such as unit price and line total. Select each extracted field and check its label and source row. Judge the source navigation by whether it helps you complete those checks with less work.

Put it to work with ParseMill

Make one real invoice your buying test

Choose an invoice you would otherwise enter by hand. Upload it, inspect the extracted fields, correct what needs attention and download Excel. Include every checking minute in the comparison. Subscribe if the result and the time saved make sense for your recurring work.

Eligible new workspaces: 10 pages for 7 days after email verification. No card or automatic subscription.

Invoice processing: $49 USD/month · 500 PDF pages · one user. Buy only after you have checked the fit on your own invoices.

Sources and editorial notes

Written by ParseMill. Worked examples are illustrative process examples, not customer results. Software availability can vary by subscription and region; check the linked documentation for your account. Review decisions and approvals remain with your team.

Explore the wider workflow