Three ways to split, and which one to use
A separate file for each range is the option for chapters, sections or anything where the document is already divided in your head: type 1-3, 4-6, 7-10 and you get three files, each holding those pages in that order.
One file holding every page you list is the option for gathering: type 2, 5, 9 to build a three-page excerpt, or 1-4, 12 to keep a cover and an appendix and drop everything between them. The pages arrive in the order you type them, which is how you can reorder as you extract.
Splitting every N pages is the option for machinery: a 120-page scan cut every 10 pages gives twelve files of ten pages, numbered part-01 to part-12 in the file names so they sort correctly in a folder or an upload dialog.
How the page list is read
Pages are numbered from 1, and the list accepts ranges, single pages and words: 1-3, 5, 8- means pages 1 to 3, page 5, and page 8 to the end. Odd and even select every other page, which is what you want when a scan put two sides of a sheet on one page and you need to separate them. A reversed range such as 4-2 is read as 2-4 rather than rejected.
Anything past the last page is trimmed instead of failing: a range of 8-20 in a 12-page document becomes 8-12 and the page reports that it did so. Duplicate page numbers are collapsed within one output but allowed across outputs, so the same page can appear in two files on purpose.
Pages are counted from the actual document, not from the number you typed in the preview field. The preview is there so the plan can be drawn before you pick a file, and it is replaced by the real page count the moment a PDF is dropped in.
What the output keeps, and why there is a file cap
Pages are copied as they are, so the extracted file keeps its text, its images at full resolution, its page size and orientation, and its searchability. Nothing is re-rendered and nothing is recompressed.
Document-level features do not follow the pages: a chapter lifted out of a book does not bring the bookmarks with it, and a form field extracted into a new file arrives as page content rather than as a working field. Signature fields behave the same way.
The cap of 20 output files per job exists because every output is held in the tab memory until you download it. Twenty files is generous for chapters and statements; if a scan needs to be cut into more pieces than that, split it in two passes, which also keeps each download small enough to check.