[userbinator]

Many years ago I reverse-engineered scribd's SWF-based PDF viewer, and was able to get a nearly-perfect conversion to PDF (since they were both vector formats). They seem to have recently replaced it with an HTML-based one and disabled the path to the SWF version, which makes such perfect conversion impossible --- you can rasterise a vector, but vectorising an image is never going to be perfect, especially an artifacted JPEG.

Searching now, it seems someone else has written a bit of public documentation (non-English) about it: https://github.com/jrlambea/scb_file_anatomy