A PDF contains no table — it contains glyphs at coordinates, and a table is something a reader infers from the gaps between them. So does this. You see every row before you download anything, and where the document prints a running balance, the arithmetic is checked against it.
Result
tables
Often, and not always, and you are shown it before anything is downloaded. There is no table in a PDF to read — only glyphs at coordinates — so every column here is inferred from the gaps the author left. Ordinary statements, invoices and reports come out clean. Merged cells, tables ruled with lines and no gaps, and cells wrapping onto several lines are the three that go wrong, and the preview is there so you find out here rather than in Excel.
Because the column could not be read the same way twice. A column becomes numbers only when four fifths of its cells parse and the whole column agrees on one decimal convention; a column mixing 1,204.55 and 1.204,55 has evidence for both and is left as text. That is deliberate — guessing there would make some cells a thousand times too large, and nothing on screen would look wrong.
If your document prints a running balance, each row is checked against the one above it plus that row's movement, to the cent. It is the only check available that does not rely on the same reading that might be wrong: the balance was computed from the real rows, so if a row went missing between two pages the arithmetic stops adding up even though every remaining row looks perfect. You still get the file — a spreadsheet with a doubtful row is worth having, unlike a redaction that cannot be proved — but the row is named so you can go and look at it.
No, it means there was nothing to check against. Most tables have no running balance, and where amounts are printed as magnitudes with a separate debit or credit marker, a sign error cannot be detected at all — so the tool says the check was unsigned or unavailable rather than reporting a pass it did not earn.
Not from this tool, and not in one step. A scan is a picture of text: there are no characters in the file, so there is nothing to place in a cell. OCR a PDF on this site reads the letters out of the image and gives you the text, which you can bring back here — but recognised text is a guess about shapes, and a wrong digit in an amount reads perfectly and costs the most.
Take .xlsx when there is more than one table, because each one becomes its own sheet in a single file. Take .csv when the data is going into another program: it is plain text every tool can read, but it holds one table per file and carries no types at all, so several tables arrive as several files inside one archive. Both formats carry the value rather than the printed characters — a statement's “(1,150.00)” becomes -1150 in either, which is what makes it add up on the other side. Switch off “read amounts as numbers” to get the characters exactly as printed instead. One thing only the CSV needs: a cell beginning with an equals sign is given a leading apostrophe, because a spreadsheet opening a CSV runs it as a formula, and these cells came out of a stranger's document.
No. This site is static files and there is no server to send it to: the reading, the column inference and the writing of the workbook all happen in this browser tab. Open your network panel, or disconnect from the internet after the page has loaded, and the tool keeps working.