- The author bought an ebook from Amazon but found the Kindle Android app buggy and crash-prone.
- They discovered Amazon no longer allows downloading purchased ebooks for use with other readers.
- Motivated by frustration, they reverse-engineered Amazon's web obfuscation to extract the book.
- Amazon's obfuscation includes randomized glyph IDs every five pages, making mapping inconsistent.
- Multiple font variants (bold, italic, bolditalic) and ligatures add complexity to the obfuscation.
- OCR failed to recognize half the glyphs due to poor single-character accuracy.
- The successful solution used rendering SVGs as images, generating perceptual hashes, and matching characters via SSIM with Bookerly fonts.
- The final decoded output perfectly reconstructed the 920-page book into a lossless EPUB with formatting.
- The author emphasizes this is for backing up legally purchased books and should not be used for piracy.