Per-36 stats is a basketball metric that estimates a player's counting stats over 36 minutes. 36 minutes is a baseline for an average starting role, which makes it useful for evaluating bench players against starters on a level playing field, instead of comparing raw totals from players who play very different amounts of time.
Per-36 assumes production scales linearly with minutes, that a player who scores at a certain rate in 18 minutes would keep scoring at exactly that rate all the way to 36. In reality, more minutes usually means facing better perimeter defenders, more fatigue, and a bigger role with more defensive attention, none of which is guaranteed to scale in a straight line.
Small sample sizes can inflate a per-36 stat a ton. A player who scores well in a handful of garbage-time minutes gets that hot stretch scaled all the way up to a full workload, even though it's a tiny, noisy sample that says very little about what they'd actually do with real minutes. The comparison below shows this happening live, with real players, right now.
There's no hard cutoff, but the more total minutes behind the number, the more it reflects a real, repeatable rate rather than a hot or cold stretch. A per-36 built on a few hundred minutes deserves a lot more skepticism than one built on a couple thousand.
For comparing players who play very different amounts of time, yes, since per-game averages just reward whoever plays the most minutes. But per-36 isn't a full substitute for context: it still needs a real minutes sample behind it, and it says nothing about efficiency on its own.
The math is identical for any counting stat, points, rebounds, assists, steals, whatever you scale. But how trustworthy the result is can vary: a stat that happens in bursts (like blocks or steals) is noisier at low minutes than something that accumulates more steadily (like points for a high-usage scorer).