What a PDF records about itself
A PDF keeps facts about itself in more than one place. The oldest is the Document Information dictionary, called Info. It holds a title, an author, a subject, keywords, the program that wrote the document (Creator), the program that made the PDF (Producer), a creation date, a modification date and a Trapped flag from the print trade. Programs may add entries of their own, such as a company name, and those are listed too.
Newer files also carry an XMP packet: a block of XML that says the same things again, with a property for each (dc:title, dc:creator, xmp:CreatorTool, pdf:Producer, xmp:CreateDate) and often more, like document identifiers and an edit history. A viewer may read either place, so they can disagree when a program updates one and forgets the other. The report marks each entry where they differ.
The report also lists what else is inside: bookmarks, attached files, JavaScript, form fields, the names in comments, and private data. Private data is the notes a design program keeps for itself (PieceInfo), metadata attached to single pages and pictures, small page thumbnails, and objects left over from saving on top of an older file. Dates are shown as stored, with their time zone, and an invalid date is marked.
Editing fields while keeping Info and XMP in step
Only the entries you change are written. A change goes to the Info dictionary and, when the file has an XMP packet, to the matching property: title to dc:title, author to dc:creator, subject to dc:description, keywords to pdf:Keywords (and the keyword list, if the packet has one), creator to xmp:CreatorTool, producer to pdf:Producer, and the dates to xmp:CreateDate and xmp:ModifyDate. The rest of the packet stays as it is. If the packet cannot be changed safely, because it is not well formed XML, it is replaced by a small correct one made from the entries, and the result says so. A file with no XMP packet does not get one.
Dates are entered in your own time zone and written with that offset, like D:20240131120000+01'00' in Info and 2024-01-31T12:00:00+01:00 in the packet, which is the same moment. A date you do not touch keeps its stored text. ClipLime does not write its own name into the file: after an edit, Producer is what it was or what you typed, and nothing is stamped with the current time.
If a viewer's title bar shows only the file name, switch on Show the title in the viewer's title bar, which sets a flag asking viewers to show the title. If the file claims to follow PDF/A, the page says so: editing keeps Info and XMP in step but cannot check that the file still meets that standard.
Removing metadata, and what "removed" means here
Removing takes the objects out of the file instead of hiding them. The Info dictionary is dropped and deleted, the catalog's XMP stream is deleted, metadata streams on pages, pictures and forms and all PieceInfo entries are deleted, page thumbnails go, and so does every object that nothing points to, which is where an older copy of the title or author hides when a file was saved by adding to the end of the previous version.
The switch Also remove attachments and JavaScript is separate because it can change how a document behaves. Attachments are other files carried inside the PDF. JavaScript is code a viewer may run when the file opens, when a field changes or when a link is clicked, and forms often use it to add up or check fields, so a form can stop calculating without it. Scripts inside an XFA form definition are not removed, and the page tells you if the file has one. Comments keep their authors' names, and a signature names its signer, so the page lists the names in comments before you remove anything.
Pictures are not rewritten. A JPEG photo is stored in a PDF as the original photo file, and that file can carry its own EXIF block with the camera model and the place and time the photo was taken. That block is not PDF metadata, so it stays. The report counts the JPEG pictures that carry such data, and the result repeats the count, so you know before you send the file.
What changes in the file and what is checked
pdf-lib, the library used here, writes the whole file again. The new file is not byte for byte the old one: it is saved as PDF 1.7, the "fast web view" arrangement is lost (the file still opens normally), objects are packed together, and a digital signature stops being valid. Objects that nothing points to, such as old copies left by earlier saves, are dropped, which is also why a file can come out smaller.
The pages are not touched. To show it, the tool computes a fingerprint of every page's content and size before and after, and the checks say how many pages match. It also compares the numbers of bookmarks, form fields, comments, attachments and scripts. If a check fails, the file is not offered for download.