We find every table on the page. Tick the ones you want, and each becomes its own sheet. Currency signs, brackets and dashes are cleaned so SUM() actually works.
We converted one source document into every major output format and measured the file that came out. The source was a 37.4 KB DOCX with headings, bold and italic text, a table, numbered and bulleted lists, and one embedded image.
Sizes are from one representative document converted with Pandoc and LibreOffice. Your own files will differ, especially with more images, but the relative ordering holds.
A PDF has no paragraphs. It stores “draw these glyphs at x, y in this font”, often one word or one line at a time, and a scanned page stores only a picture. Drop any PDF to see which kind you have and what its page instructions look like.
English-speaking countries write 1,250.50. Much of Europe and South America writes 1.250,50. A PDF only stores the characters, so the converter has to know which convention the document used: guess wrong and every figure is off by a factor of a thousand.
1,250.50arrow_forward1,250.5Other convention reads 1.25051.250,50arrow_forwardtext: can’t readNot valid in this convention1.250arrow_forward1.25Other convention reads 1,2501,250arrow_forward1,250Other convention reads 1.250,75arrow_forwardtext: can’t readNot valid in this convention12.5%arrow_forward0.125Same either wayA transaction list printed over three pages repeats its header row at the top of each, and page footers sneak in between. Stitching removes the repeats so you get one continuous table you can sort and filter.
Reports often group columns under a shared heading. In Excel, merged cells stop sorting, filtering and pivot tables from working. Flattening gives every column its own complete name instead.
For the prose around them, or the pages you don't need, these do better.