What “compression” actually changes in a PDF
A PDF is a bag of objects: page trees, font programs, content streams, and embedded images. Text-heavy reports are often already small because Type1/CFF fonts subset to a few dozen kilobytes. The bulk is almost always JPEG or JPEG2000 image XObjects, sometimes at 300–600 ppi from a scanner.
Low and Recommended walk those image streams, decode them, and re-encode as JPEG at a lower quality. Content streams that paint text stay as operators (Tj, TJ), so you can still select a paragraph. High does something different: it rasterizes each page to a bitmap and wraps that in a new single-image page. File size drops hard. Selectable text is gone because the operators are gone.
Worked example: a 28-page lease scan at 42.6 MB is usually 28 full-page JPEGs around 1.5 MB each. Recommended often lands near 6–9 MB. High can land under 3 MB. The same file exported from Word with no photos might already be 180 KB — compressing it can make it larger, because you add a rewrite and JPEG headers around assets that were not the problem.
Why local processing matters for a contract
A “free PDF compressor” that uploads a signed NDA is a second copy of that NDA on a machine you do not control. HR packets, bank statements, and medical forms are the files people compress for email. Those are exactly the files that should not sit in a converter’s temp disk.
This tool uses the File API and pdf-lib in the browser. There is no upload endpoint. Closing the tab drops the bytes. That is the product, not a slogan.