Check whether a table continues without repeating its headers

Where the extra work starts

Check where the table carries on and compare its first row with the prior page before accepting the combined workbook.

ParseMill processes native invoice PDFs and clear printed English PDF scans. Try a typical supplier invoice and a difficult one. Check the extracted fields, correct exceptions and open the workbook before committing to a monthly plan.

ParseMill can carry recognized invoice headings onto following pages of the same invoice, but each continuation still needs a coverage check. Compare every expected row with the result, especially when columns shift, headings disappear or notes interrupt the table. Report tables outside the invoice workflow need their own fit assessment.

A practical way to get it right

Build a page inventory with the table's title, header presence, first and last row references and any continuation note. Compare the column positions and units across pages. A similar appearance is a clue to a relationship, but the reviewer should confirm that the later rows belong to the same table.

Record expected rows by page before inspecting output. If a later invoice page has no headings, verify whether the preceding column mapping was carried forward correctly. Where rows are omitted or misassociated, use a source-backed correction or reviewed manual recovery and retain the original page reference.

When combining reviewed outputs, preserve page and row identity and check for overlapping boundary rows. A carry-forward line or repeated last row can create a duplicate. Keep the recovery method explicit so the combined register does not appear to have been produced entirely by one successful automatic run.

Microsoft: Power Query PDF connector documents the MultiPageTables option for similar tables on consecutive pages. That is product-specific behavior and must not be assumed for ParseMill or accepted without review.

What this looks like on a real task

Synthetic example. The people, references and figures below are invented to demonstrate the method.

The source contains ten expected rows, but page one's four rows are not a complete result. Copy the page inventory and add the observed output count and recovery reference. If page two is transcribed manually, retain that provenance and review it against the same raw-field standards as the extracted pages.

Catch the cases that create rework

Do not paste headings onto the source and then describe the result as unchanged extraction. Any prepared derivative needs a recorded transformation and verification. A footer containing a familiar heading word is not necessarily the actual table header.

Other tools may combine similar tables across pages, but that behavior still needs source checks. Automatic combination can join the wrong region or hide a layout change.

W3C: PDF table markup distinguishes header cells from data cells. Preserve the relationship that the missing continuation header would otherwise leave implicit.

Decide whether the change is worth keeping

Reconcile every expected page and row to an accepted output or explicit unresolved record. Check boundaries for missing and duplicate rows, then verify units and field assignments on the continuation. Report the scope honestly: a mixed automatic and manual register can be useful, provided its evidence and remaining limitations are clear.