I read the abstract and skim the intro before committing to a read. The authors have to convince me that they know the field, they think they’ve done something interesting, and I think what they’ve done is plausible. If it passes that bar, I assume the most adversarial possible mindset and look for holes in their methods. If their methods are junk, I may skim the conclusion just to see what kind of unfounded nonsense to watch out for in the future, but otherwise I’m done, and really most papers are done at this point. Papers in my field are mostly bogus, unfortunately. Every now and then, somebody uses plausible methods, and only then do I really bother to sit down and read the whole thing.

I tend to work in more theory-heavy CS (at least when I'm reading papers), so sometimes even the abstract is obscured by lots of scary terminology, so if I want to understand even the basics, I need to do it recursively.