I remember reading that in the documentation, and I think it was this part that confused me in the first place, thanks for mentioning.
When I read “each frame” made me think about a small window of samples (with size of the analysis window parameter) and to access the [bufmelbands~] results I would have to read a sample every hopsize, OR, that this frame required some further processing, when says: “Each frame represents a value…”
On top of that, the MEL buffer reported a equivalent duration to the source buffer (I wasn’t aware of the SR change at the time).
I was off the rails…
Perhaps, it’s my english understanding (which happens often), but in my humble opinion I believe it could be written in a more clear way.