Recognize two unrelated tables on the same PDF page

Where the extra work starts

Identify each table's purpose so the extraction can be reviewed against the right set of columns.

For supplier reports and technical tables, start with ParseMill’s synthetic material-register example. Compare the source references and output with the register you need, then discuss a document pilot with an agreed scope and fee. The self-service trial covers invoice PDFs.

The boundary should reflect row meaning, heading scope and surrounding context. Proximity on the page is not evidence that two regions form a continuous table.

A practical way to get it right

Identify each region by page and a stable local reference such as top table or lower-left table. Record its title, row definition, headings, units and expected row count. Inspect notes between regions that may qualify only one of them.

Review candidate extraction separately for each region where the chosen tool supports that selection. Do not assume ParseMill offers arbitrary region selection or semantic separation of unrelated tables. If the file does not fit its bounded workflow, choose a reviewed manual or alternative method.

Before appending results, compare the row definitions and field meanings, not just column names. If the tables serve different roles, keep separate datasets or add an explicitly reviewed type field without implying the values are interchangeable. Preserve the region reference on every output row.

Microsoft: Power Query PDF connector exposes detected PDF tables in Navigator for selection or transformation. Compare those detected regions with your inventory; detection does not establish that their row meanings match.

What this looks like on a real task

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

The same word Result could appear in T1 and T2 without establishing the same meaning. Copy the region, row-meaning and decision columns and add expected counts. The combined output should not turn a permitted limit into an observed measurement simply because both cells contain numeric text.

Catch the cases that create rework

A parser may return several detected tables or one merged region. Inspect the candidate structure against the inventory rather than accepting either grouping automatically. If a heading appears between datasets, ensure it has not become a data row or been discarded along with a necessary region label.

Side-by-side regions can also produce interleaved reading order. Review both the field values and their region association.

W3C: PDF table markup describes distinct table, row and cell structure. Preserve separate region relationships rather than flattening every similarly labeled block into one table.

Decide whether the change is worth keeping

Every accepted row should identify its source region and retain the intended meaning. Reconcile expected counts within each region before considering any combined total. A useful inventory makes excluded content and separate datasets explicit, allowing later users to understand what the extraction covers without reinterpreting the entire page.