A clutch is a round won by the last player alive on their team, against one or more opponents. It is written 1vN — 1v1, 1v2, and so on.
Clutches are the most cited and most misread numbers in Counter-Strike. They are rare, high-variance, and heavily circumstantial: a 1v2 with the bomb down, full armour and an AWP is a different proposition from a 1v2 on a save with a pistol.
That is why raw clutch counts mislead. A player who is often last alive may simply be dying last on a losing team, inheriting clutch situations they never chose. What you would actually want is the conversion rate — clutches won over clutches faced, split by opponent count — but that is not a number the API hands you directly.
How it is measured, honestly: the API exposes a clutch count per player-match, and it is nullable, since not every source populates it. There is no attempts field, no 1v1-versus-1v3 breakdown, and no bomb state on the player row. To build a genuine conversion rate you reconstruct the situations yourself from the round-by-round data at /matches/{id}/rounds, which records who was alive and when.
The impact of a clutch, as opposed to how many a player accumulates, is better captured by round-win-probability metrics than by a tally. Winning a 1v3 to steal a round is one of the largest single swings in the game, which is what HLTV's Round Swing tries to value; a raw count scores that heroic hold and a gift 1v1 identically.
Because clutches are so low-frequency, sample size matters far more here than for ADR or KAST: one tournament's clutch tally is dozens of rounds, not thousands, and reveals far less about a player than a season of either.