NaN (“Not A Number”) represents an undefined or unrepresentable floating-point result. It was introduced by IEEE 754 and is available in Clojure as `Double/NaN`. Despite its name, `(number? Double/NaN)` returns `true`. section 5.8
Where NaN comes from
Floating-point operations such as `(/ 0. 0)`, `(/ ∞ ∞)`, `(* 0 ∞)`, and `(Math/sqrt -1)` produce NaN. Arithmetic can carry it onward: `(inc Double/NaN)` also returns NaN. section 5.8
NaN is an edge case for functions that depend on numeric ordering. If any argument to `max` or `min` is NaN, the result is NaN—even when the other arguments include infinity. Numeric equality is surprising too: `(== Double/NaN Double/NaN)` returns `false`. section 5.8
Comparison is not numeric equality
`(compare Double/NaN 1)`, `(compare 1 Double/NaN)`, and `(compare Double/NaN Double/NaN)` each return `0`. That does not make NaN equal to those values under `==`. It also makes sorting results worth checking: `(sort [3 2 Double/NaN 0])` yields `(0 2 3 NaN)`, while `(sort [2 3 Double/NaN 0])` yields `(2 3 NaN 0)`. section 5.8 section 6.4
The practical rule is to account for NaN before relying on an extreme value or a sorted order. `max` and `min` propagate it; `compare` can return zero for it and an ordinary number. section 5.8 section 6.4
