The multivariate test is trying to increase ad clicks, and the result is a poor UX. That's not exactly an alternate theory.

Is it a dark pattern? That's not the first term I'd use to describe it, but perhaps it falls under "Misdirection" mentioned at https://arxiv.org/pdf/1907.07032.pdf (page 12).

I don't think it is a 'perfect balance' so much as trying to maximize the number and length of advertisements that people will tolerate at this point in time. The nature of the business model is that some product manager will be tasked with getting the advertisement revenue higher at some point in the future and they will no doubt try to push the envelope. I also feel like 'total view time' is a metric that is in service to advertisements and not the primary goal itself.

>Alternate theory: A large-scale, ongoing A/B test

Except "ongoing A/B test" isn't really a thing. The whole point of A/B is to identify and discard the worse of the two. Especially on google scale and this being applied to everyone you'd expect the data to point one way or the other really fast.

This is just straight up dark pattern.

This seems so much more likely

Agreed. Unfortunately when it comes to ads (which I also hate! I have adblockers installed, subscribe to premium services like Youtube premium when it makes sense...) HN commenters take leave of their wits and just start posting conspiracy theories, hysterical criticism, or if you're lucky balanced sensible criticism that nevertheless doesn't make sense in the context of the article and is just an instance of preaching to the choir.