A little swamped right now, but just popping in quickly because I saw this. Its not unusual for program evals to be published, even if explicit research consent was not obtained at the time. IRBs usually have processes in place for revising data that is collected as part of program evals, QI, etc. and applying secondary data analyses to it. Usually there are some additional securities in place to ensure anonymity, etc. Its essentially the same method used for EHR analysis and the like.
Beyond that, it really just depends and others are correct that there is a TON of grey area here. Some depends on the scope of the project. Trying to eek 7 papers out of a small pilot from a master's thesis is different than generating 3 from a multi-million dollar R01. Large epi studies will sometimes have hundreds (if not thousands) of papers from a single dataset. Available guidance on the topic is - in my experience - incredibly vague. Really, I think its just a "you know it when you see it" kind of thing and there is a difference between 1-2 grey area "Ehhh....I guess I can see it, but I probably wouldn't have split that up" papers versus someone who routinely does it. And at a certain point, you know the folks in your field who make it a habit. We're commonly collecting data across a huge number of modalities now and it would be utterly incoherent to integrate behavioral mechanisms with clinical outcomes, neuroimaging mediators that may or may not be moderated by a different imaging modality baseline scan, combining with 'omics data gathered from a supplement, etc.
Not sure if this helps. If you want to PM specifics, I'm happy to share whether it passes my personal sniff test for whatever that is worth, but I would encourage you to just do so with colleagues. Unfortunately, I think the "sniff test" is about all we have at this point. Generally speaking, I think a lot depends on a serious look at your own motivations and whether its realistic to make something one paper. If something <can> be one <reasonably> large but coherent paper, it probably should be. If the only desire to not do so is so everyone can get publications, that is a problem. If its asking different research questions that would be incoherent in combination, its fine to split. In between, it gets murky.