In MWI there is no collapse event, because there are no observers in the traditional sense that may collapse the system. In MWI we resign and just say that observers are subject to QM, and when a "measurement" is made, what happens is that observers and the observed system get entangled so that certain combinations of properties of the observer-observed system are not possible. This entanglement of two projections of a composite system can be described in the language of QM.

> what is predicted by standard QM, because the MWI is supposed to recover the same predictions.

MWI recovers the same predictions. Where in standard QM if one says an observer observes that a system may collapse 40% in state a and 60% in state b, MWI says that in systems (of the observer and the observed) where the observer makes the observation in the same exact manner, the observer will record state a with the system in state a 40% and the observer will record state b with the system in state b 60%, and there are no states where the observer observes something and the subsystem is in the other state.

Of course, you can simply model the entire system of observer and observed in standard QM and recover the same relative predictions without invoking the absolute observation of an observer. In MWI we do away with the conceit that there is an absolute observation of an absolute, but ill-defined observer.

If standard QM has no problem to make predictions then I agree than in principle the MWI could make the same predictions (but it seems to me that in practice it can only make predictions by using standard QM and saying that those are the same predictions that could in principle be derived).

Edit: it's too late to alter my original ancestor comment, but I think the way I wrote it may have yielded a misunderstanding. I didn't mean to suggest that MWI has value because it can predict phenomena. I meant to say that it has value because it is a metaphysically cleaner model of reality than the collapse interpretations.