I must admit I haven’t fully followed all the conditions you outlined, as there seem to be some incorrect assumptions. From the beginning
and end
We don’t know a priori that the file content has not changed, and that’s the core problem: A file which hasn’t changed will look to Syncthing like it has (new mtime), which means the file gets rehashed and metadata about that exchanged. Even if the latter part is somehow worked around, hashing it pointlessly is already a deficiency. Therefore it does not only affect receive-only folders, it’s just that the problem gets very apparent there.
Also you seem to get the mtime 2s window issue totally - what’s still unclear about the proposed solution: Consider two mtimes equal if they are within a 2s window. There’s nothing more to it then conceptually (of course there is stuff to sort out internally, that’s partially what Audrius and me were discussing).