Find a row split across a PDF page break

Establish whether two fragments are really one record

Check the boundary before using the export so the recipient does not have to discover and rebuild the missing relationship later.

For multipage invoices, use ParseMill to review extracted invoice data beside the source and export separate summary and line-item sheets. Include a page break in your trial and check that every required line made it through. Other report types need their own fit assessment.

Do not fill a missing amount from the next page simply because it is the next number in reading order. It may belong to a new item or a page total. Keep each fragment separately until the record boundary is supported. A flat table extractor can return technically valid rows while losing this visual relationship.

Keep reconstruction outside the untouched extraction

Microsoft: PDF connector provides a route for importing detected PDF tables. Microsoft: Table.Combine appends tables; appending by itself does not establish whether two fragments describe one record.

1. Save the original output and mark the last two records on the first page and first two on the next. Include physical page and row references for each fragment.
2. Transcribe the visible fragment strings into a reconstruction sheet. Leave absent fields empty and identify which page supplied each field.
3. State the evidence linking the fragments, such as a repeated item identifier or an unbroken description with a matching column alignment. Have another reader check an ambiguous join.
4. Create a new reconstructed row only after deciding the link. Retain both fragment references and the reviewer decision instead of overwriting one fragment with the other.

Work through a split description

Synthetic example: item A17 starts at the bottom of page 3. Its description continues at the top of page 4 before item A18 begins.

After inspecting the actual layout, the reconstructed description may be “Steel bracket, galvanized for exterior use,” with quantity 4 and price 12.00. The reconstruction record must retain both source locations. If page 4 lacks a convincing link to A17, do not publish that joined row as certain. Mark the amount and quantity association unresolved and request a better source or manual review.

Watch for joins that look plausible but are wrong

A repeated heading can sit between the fragments. Remove it from the proposed description only when its role is confirmed; otherwise a header filter may disguise the missing connection. A footnote can also look like continuation text but qualify several records instead of extending one description.

Different columns may wrap independently. One cell continuing onto the next page does not mean every top-of-page value belongs to the same previous row. Review each field assignment and preserve the precise punctuation and units. If the layout requires guessing, document it as unsupported for the automated workflow.

Check the reconstructed result before handing it off

Count logical records in the source region and compare them with the reconstructed output. The fixture should contain A17 and A18, not three unrelated items or a combined A17/A18 row. Confirm that no fragment was both retained separately and included in the reconstruction. Ask a reviewer to follow both source links and explain every populated field.

Use the extraction validation guide for the final checks. In an invoice trial, include a line that crosses a page break and check each resulting field beside the source. ParseMill supports native invoice PDFs and clear printed English PDF scans, but a split record can still require manual correction. Count that work when deciding whether the workbook saves time.