For what it’s worth, I prefer to have the edits saved in the metadata of the jpeg/tiff/png that the edits produced. Putting them in the raw means to me that only one edit chain is valid or current for that image, and in my workflow that’s not the case - small jpeg for proof, full-sized tiff for further editing, etc.
It’s what I do in my hack software now. My file menu has an “Open Source…” selection, where such an encoded image can be specified; what the software does is to read the edit chain from the output image, identify the starting image, open it instead of the and re-apply the edits. My workflow is fundamentally dependent on this capability.
Right now, I save the edit chain in the EXIF ImageDescription. Now that I know of the Adobe tags, I may consider modifying my software to use them instead, if they’ll support my editing semantics (that is, the chain of operations).