A lot of basketball analytics has been a chase for a holy grail: one number that tells you how good a player actually is. Box score stats can't do it — they only capture what's easy to count. Metrics built on top of box scores inherit the same blind spots. The breakthrough was Regularized Adjusted Plus-Minus, and the idea behind it is almost unreasonably simple: measure how the score changes when a player is on the court versus off it, adjust for who else is out there, and regularize so the noise doesn't eat you alive. What you get back is a single number that captures something the box score never could — a player's actual impact, stripped of context.

RAPM shaped how I think about a lot of things, not just basketball. The lesson I take from it is that you don't need to be able to measure everything to learn something real. The box score misses most of what matters on a basketball court — screens, rotations, gravity, decision-making — and for a long time, people assumed that meant you couldn't quantify those things. RAPM showed that you could, indirectly, just by being clever about what you do observe. The signal was always there. It just needed a better question. I think that's true far more often than people realize, in a lot of domains — there's more signal sitting in existing data than we've figured out how to extract yet.

That's the kind of problem-solving I care about. A lot of applied statistics operates in a world where the answer is "get more data," and that's often right. But the problems I find most interesting are the ones where more data alone doesn't get you very far — where samples are small, context shifts between settings, and the raw numbers can mislead you if you don't think carefully about what's generating them. The real work is structural: what should you actually measure, what's a fair comparison, how do you borrow strength where you don't have enough signal on your own. Basketball is full of problems like that, and it's a domain I know well enough to have good instincts about when something is off.

Outside of basketball, I do research on AI and labor markets at the Burning Glass Institute. The overlap is less weird than it sounds — both are about understanding how systems evaluate people, and where those evaluations break down. I grew up reading about political forecasting and going on long Wikipedia spirals about basketball stats, and somehow that turned into actual work. This site is where I put it.