A PDF page can declare a viewing rotation separately from its content coordinates. PDFRotate reads that declared angle, adds one exact supported turn and writes the normalized result as 0°, 90°, 180° or 270°. That narrow operation is useful, but it is important to understand what it does not change.
Orientation is a page instruction
A 90° page rotation tells a conforming viewer how to present that page. It does not decode every image and paint it again. Text, vector paths, images and annotations remain page content. This avoids a lossy raster round trip, but it also means a viewer or workflow that ignores page rotation can behave differently. Mirroring is another transformation and is deliberately unsupported.
The chosen turn is added
If a page already declares 90° and you choose another clockwise 90°, the result declares 180°. A page at 270° plus clockwise 90° returns to 0°. PDFRotate does not replace every page with one absolute angle, because that would destroy meaningful mixed orientation. The preview therefore states the selected page count and delta, not a promise that all chosen pages end at the same angle.
A new PDF needs an independent check
PDFRotate changes page rotation entries and serializes a separate file. It does not promise byte identity or preservation of every document-level feature. Metadata, outlines, page labels, attachments, optional layers, tagged structure, PDF/A or PDF/X claims, usage rights, encryption, incremental revisions and digital signatures may change or stop applying. Forms, annotations, links, crop boxes and accessibility relationships can depend on viewer behavior. Keep the untouched source, open the download in another trusted viewer and inspect every affected page before using it.