Your theory is largely consistent with Yvain's theory.

Yvain says the brain eliminates dreams because of a noticed property of dreams as such. My theory (not really mine, just endorsing without remembering where I first read it, but I'll keep the terminology) says that the brain is just applying normal hypothesis update procedures, with no need to identify the category "dreams".

Maybe it's competing for the short-term part of his theory, but your theory simply doesn't address why we dreams fade after waking up.

I think it does. To wake up is to be bombarded with overwhelming evidence that one's most recent inferences ("dreams") are false. So, whatever neural mechanism (synaptic strengths + firing patterns) represented these inferences is crowded out, if not replaced outright, by a radically different one.

You might say, "But when I change my mind after believing something stupid, I remember that I used to believe it." Sure, because that belief was there much longer and developed more inertia compared to a dream, and you "self-stimulated" that belief, which, lo, helps you remember dreams too.

"But when I briefly believe something stupid and then correct it, I remember it." Compare the set of all beliefs you've held for under twenty minutes, to the set of all your dreams. Do you think you remember a higher fraction of one than the other?

But I see his theory rather differently than you do: it is evolution, not the brain, that has made the observation.

I wasn't claiming Yvain left out the possibility of evolution doing the learning -- that's what I meant by "over period _?" Was this entanglement noticed over the person's life, evolutionary history (the Baldwin effect), or what? But I didn't know how to concisely say that any more clearly.

Of course, if dreaming is about rewriting long-term memory, then other effects on memory could be side-effects.

True, and that would be a parsimonious way to handle the phenomenon of dreaming, but that wasn't Yvain's theory.