One of the main reasons I want to be at a liberal arts college is that undergraduates can do real research here, and be the first author on it. My work is computational, which makes it a good fit: a student needs a laptop and a willingness to learn Python, not a beam line or a telescope allocation. The Nirvana catalog contains models of over a thousand barred galaxies, and there are more good questions to ask of it than I will ever get to on my own.
Four students have worked with me so far, two at Colby and two at Swarthmore.
I try to scope projects so that a student can finish something in the time they actually have — an academic year or a summer, around their coursework. That usually means handing them a well-defined question against an existing dataset rather than asking them to build infrastructure from scratch.
Most of my students arrive with one Python course or none. I spend the first few weeks on the same kind of hands-on data work I used to teach in my scientific computing class and the Lamat REU bootcamp: load the data, look at it, plot it, fit something to it. I meet with research students at least daily over summers to ensure that they're never stuck too long on a single bug.
If you are a student reading this, these are the questions I have sitting on the shelf. All of them are open, computational, and sized for an undergraduate.