Metric glossary
Every proprietary metric ships with public documentation: what it measures in plain language, the formal definition, required inputs, minimum samples, the model behind it, and its known limitations. A metric without documentation doesn't ship.
Counter-Strike 2
8 metricsHow likely a player was to win each fight before it happened, based on weapons, health, distance, flashes, smoke, and nearby support. A kill in a 20% situation is worth more than one in a 90% situation.
Duels won above what the situations predicted, per 100 duels. Positive means winning fights they 'should' lose; negative means dropping favorable fights.
How much each player's actions moved their team's chance of winning rounds. A 1v3 clutch kill swings far more probability than a cleanup frag in a 5v2.
Value created in the first duel of each round, adjusted for how difficult those opening situations were and how often they convert to round wins.
Share of a player's deaths answered by a teammate kill within five seconds. High values mean deaths that keep round equity; low values mean deaths that hand over map control for free.
Share of deaths that were classified as isolated: no teammate close enough to trade, not on an objective, not a last-alive or sacrificial situation.
Round-rate value of grenades: enemy blind time that leads to team action, utility damage, and flash assists. Measures utility that does something, not utility thrown.
Average distance to the nearest living teammate at the moment of each death. A positioning fingerprint: low = plays inside team structure, high = plays independently.
VALORANT
5 metricsDuels won above what each situation predicted (weapons, health, flashes, smoke, support), per 100 duels.
First-contact value adjusted for difficulty: winning hard opening duels on site hits scores highest.
Round-rate value of utility that affects enemies: blind seconds, enemies revealed or displaced, and damage from abilities.
Share of deaths answered by a teammate kill within five seconds.
Average impact generated per ultimate used: enemies affected, kills enabled within its window, and damage.
Fortnite
5 metricsHow much better a player placed than the survival baseline expected, phase by phase. Rewards outliving the lobby's hazard curve, not just one lucky endgame.
How early and safely a player reaches the next zone: time inside the safe zone before it closes, storm damage taken, and distance traveled under pressure.
Whether fights are taken from advantage: health/shield edge, elevation, and third-party risk at engagement start.
Damage dealt per 100 building materials spent — output per unit of the scarcest endgame resource.
How often knocks are converted into eliminations before a teammate revives or resets.
Rainbow Six Siege
5 metricsOpening-engagement value adjusted for difficulty, using the R6 coefficient set of the duel model.
Combines how often a player's deaths get traded and how often they trade teammates — presence in the kill network.
Round-win probability created through defuser plants and plant denials, using round-state swings.
Team kills that follow this player's drone spots within a short window — turning intel into eliminations.
Defender gadgets, cameras, and drones removed per round, weighted by importance.
League of Legends
5 metricsGold lead at 14 minutes over the opposing player in the same role — early-game pressure measured against the direct matchup.
Win-probability swing from objectives this player's team secured while they were alive and nearby — dragons, heralds, barons, towers.
Ward output weighted by risk and denial: control wards, sweeps, and wards placed deep in enemy territory.
Team damage share minus team gold share — output relative to resources consumed.
Win probability surrendered through deaths, weighted by game phase and shutdown gold handed over.