Compatibility survived the argument.
The optimizer does not ask the world to install a new decoder. It studies the image before the codec spends its bits, then judges the result after the ordinary JPEG has been decoded. Research about ownership, transport, and visual structure becomes a file that still opens everywhere.
- Everyone assumed
- Image optimization is whatever decision the encoder already knows how to make.
- I asked
- Why should a codec allocate its bits without the structural ownership and perceptual priorities already visible in the image?
- I changed
- Carry structural regions, signed transport, and explicit error budgets into a search over standards-compatible JPEG output.
- Now there is
- A local optimizer that accepts ordinary images, searches against the decoded result and actual file size, and returns an ordinary JPEG.