Maybe this can give you some insight. I downvoted your comment simply because it didn't add anything to the discussion and you made an assumption that has several faults in it.
For one, your "mental model" is off because you assume that the first part of "readable bytes" is accurate. Without actually seeing the PDF in question, you don't know if the "readable bytes" are actually corrupted and Preview is fixing them to make them readable. That would mean that Preview is actually correct in its behavior and the source document is what's flawed.
On the tail end of your mental model, then, is another assumption which is that this results in "unreadable bytes" but that's not accurate either. The PDF that results from a save in Preview may be accurate to the PDF specification and is perfectly readable as a PDF in any PDF application/reader. What's no longer readable is the extra content that was originally in the file that may not have been saved correctly, in-spec, or may have been corrupt to begin with.
A big hint as to what's going on here, now that I've had some time to review this, is that the "corruption" happens consistently - the letter "a" is always replaced by the same "corrupted" character, the letter "b" seems to be consistently replaced with the same character, etc. That points, in my opinion, to a lookup that's no longer correct. What side of that lookup is bad is hard to say without seeing the file in question.