While I love 2x2 matrices, there are two incredibly important things missing. The first is the career of the PhD students you (as the faculty) are mentoring and supervising. Sure, you might do fun things, but if it ends up not helping their career, then you as the faculty have really failed them.
This issue of students is something often missing in discussions of the modern research enterprise. Speaking as a computer science faculty, if a PhD student wants to compete for an academic job, they need about 1 publication per year. Sure, back in the 1970s and 1980s you could get a CS faculty job with just 1 good publication, but that's not the case anymore. Of course, everyone is doing this, which then leads to more publications overall.
The second is the overall funding of science. There's an unstated compact that the public is funding science to help solve important problems and lead to breakthroughs that can advance the economy. This is the framework that has been in place in the USA since WWII thanks to Vannevar Bush, and is why NSF, NIH, and other funding agencies exist. Sure, as researcher, I strongly agree that we should be doing things we view as fun. But those fun things should have some angle as to why others should care. Another way of framing this is Pasteur's Quadrant, which is good science and good applications.
The more general advice I give to all students (undergrad, MS, and PhD) when looking for jobs: do something that is (1) fun (because there are many job opportunities in our field and life is short), (2) something that will keep pushing your boundaries (because you're smart and will be bored otherwise), and (3) good for humanity.