NHL analytics glossary: expected goals, GSAx, xGF% and more

Plain-language definitions of 55 hockey stats and model terms, each as this site calculates it: Calibration, Expected goals (xG), Goals saved above expected (GSAx), xGF%.

5-on-5 (5v5): Both teams have five skaters and a goalie on the ice. It is most of every game, so it is the fairest way to compare teams and players; stats labelled 5-on-5 or 5v5 use only this state.
5v5 net/60: A skater's 5-on-5 rating per 60 minutes in a neutral role, in goals: his offence plus his defence. Because it is per minute, it does not depend on how much he plays.
5v5 offence and 5v5 defence: The two halves of a skater's 5-on-5 rating, in goals over the rest of the season at his expected minutes: goals his team adds while he is on the ice, and goals it prevents. Positive defence means fewer goals against.
All situations: Every game state together: even strength, power plays, penalty kills and empty-net play.
Calibration: Whether the percentages mean what they say: when the model gives the home team about 60%, the home team should win about 60% of the time. The track record page plots this.
Confidence tier: A label on every game forecast for how strong the favourite is: Toss-up, Slight edge, Lean or Strong. Hover or tap a tier to see how often favourites like that won in past seasons.
Confirmed lineup: The dressed skaters and the starting goalie from the NHL's official Game Roster report, read shortly before puck drop. Once it is posted, a lineup-confirmed forecast replaces the morning one on the game card.
Cup odds: The share of simulated seasons in which a team wins the Stanley Cup, after a bracket of best-of-seven series with home ice by seed. Series are hard to forecast, so treat these as rough.
Decision: The official W (win), L (loss) or O (overtime or shootout loss) credited to one goalie per team for each game.
Derived lines: The NHL does not publish line combinations, so lineups are arranged into four forward lines and three defence pairs by who played the most 5-on-5 time together in the team's last game, from the official shift charts. Each line's centre is the forward who takes the most faceoffs this season, since listed positions are often out of date. “New” marks players who did not play together then.
Expected goals (xG): The chance an unblocked shot becomes a goal, from its location, type and the play before it; added up, it says how many goals a set of chances was worth. 2026–27 values come from this project's model on official NHL play-by-play (empty-net attempts get none); earlier seasons use MoneyPuck's.
Expected league rank: A team's average finishing position across the whole league in the simulated seasons, where 1 is the most points.
Expected lineup: The best guess at a lineup before the team announces it: the skaters who dressed for the team's last game and are still on its official roster, plus each goalie's chance to start. Players ESPN lists as on injured reserve or out are left out, and regulars who sat out that game are listed as scratches. The confirmed lineup, read from the NHL before puck drop, replaces it. A guide, not a team announcement.
Finishing: The part of a skater's rating for scoring more or fewer goals than his own shots' expected goals, in goals over the rest of the season.
GA/60: Goals a goalie allows per 60 minutes in net.
GF%: GF ÷ (GF + GA): a team's share of the actual goals in its games (for a player, while he is on the ice). 50% is an even split.
Goals above average: A skater's projected contribution over the rest of the season at his expected minutes, in goals compared with a league-average skater: 5v5 offence, 5v5 defence, finishing, power play, penalty kill and penalties added together. It is not WAR; the season-by-season chart uses the minutes actually played.
Goals saved above expected (GSAx): Expected goals on the unblocked shots a goalie faced, minus the goals scored on those shots. Shots at an empty net, or by an extra attacker while he was in net, carry no xG and are left out of both. Positive means he allowed fewer goals than the shots were worth.
Goes to overtime: The chance a game is tied after 60 minutes and needs overtime, and possibly a shootout.
High-danger chance: An unblocked shot attempt worth at least 0.15 expected goals.
Injury status: Whether a player is on injured reserve (IR), out, or day-to-day (DTD), with the injury and an expected return when known. It comes from ESPN's injury list, checked each morning and before each game, because the NHL publishes no injury report. It shapes the expected lineups only; the forecasts never use it.
ixG (individual expected goals): Expected goals from a player's own unblocked shots.
Lean: The third confidence tier. Favourite's chance 60–65%; such favourites won 62.6% of 2,655 games from 2015–16 to 2025–26.
Likely range: The span of final point totals covering the middle 90% of simulated seasons (5th to 95th percentile).
Lineup-confirmed forecast: A second game forecast issued once the NHL posts dressed rosters before puck drop, using the players who are actually playing (and the starting goalie when it is marked).
Live win chance (experimental): During a game, an estimate of each team's chance to win from the score, the clock, the players on the ice and the forecast made before puck drop, updated with the live feed. Tested on 2022–23 to 2025–26 it was usually within 1 to 3 points of how often teams actually won, a little short of the accuracy bar set for it, so it is labelled experimental; its record this season is on the Track record page.
Log loss: A score for probability forecasts: minus the logarithm of the chance given to what actually happened, averaged over games. Lower is better; calling every game 50–50 scores .693.
Market (no-vig): Bookmakers' odds turned into win chances with their built-in margin (the vig) removed, then the median across bookmakers. Market close is the last check before puck drop; it is shown for comparison only.
Model closer / Market closer: On a result, whichever of the model and the market gave the team that won the higher chance.
Morning forecast: The game forecast issued each morning from team results so far. It does not know the starting goalies or the final lineups.
Penalties: The part of a skater's rating for penalties he draws minus penalties he takes, in goals.
Penalty kill (PK): Playing a skater short because of a penalty. Team and player stats on this site use exactly 4-on-5; the PK rating is a skater's value in those minutes, in goals.
Playoff odds: The share of simulated seasons in which a team makes the playoffs: the top three in its division or one of the two wild cards in its conference.
Position percentile: Where a skater ranks among forwards or among defencemen (99 is best): on the player page by projected goals above average, for skaters with a roster listing or a 2026–27 game, 200 projected minutes and 200 5-on-5 minutes in 2023–26; on the season chart among skaters with at least 20 games that season.
Power play (PP): Playing with more skaters than the opponent because of a penalty. Team and player stats use exactly 5-on-4; power-play expected goals on game pages count any time a team has more skaters.
Presidents' Trophy: Awarded to the team with the most regular-season points; the standings show each team's chance of winning it in the simulations.
Projected games and Min/game: A skater's expected games over the rest of the season and expected minutes per game, from his recent usage.
Projected points: A team's average final points total over the simulated seasons: the points it already has plus simulated results for every remaining game.
Regulation win: The chance a team wins within 60 minutes, without overtime or a shootout.
Roster listing: The NHL's current roster for each team, checked daily. It has no injury details, and a player missing from it is not necessarily unavailable.
Save % (SV%): Saves ÷ shots on goal against. Missed and blocked shots do not count.
Saves above expected per 100 shots: A goalie rating from each goalie's history in the model: saves made beyond what expected goals predict, per 100 shots. The forecast updated with confirmed lineups uses the starting goalie; the morning forecast does not.
Scratches: Players on the official Game Roster report who are not dressed for the game.
Simulated seasons: The rest of the schedule played out 50,000 times from each team's current strength, with official tiebreaks, then the playoff bracket. Standings and playoff odds are shares of these simulations.
Slight edge: The second confidence tier. Favourite's chance 55–60%; such favourites won 56.9% of 3,847 games from 2015–16 to 2025–26.
Strong: The top confidence tier. Even these favourites lose about three games in ten. Favourite's chance 65% or more; such favourites won 68.8% of 2,671 games from 2015–16 to 2025–26.
Toss-up: The lowest confidence tier: close to an even game. Favourite's chance 50–55%; such favourites won 53.3% of 4,339 games from 2015–16 to 2025–26.
Track record: How past forecasts did, from walk-forward tests: each past season was forecast by a model fitted on earlier seasons only, so the results are a fair guide.
Unblocked shot attempts: Shots on goal plus missed shots; blocked shots are left out. Expected goals are counted on these.
Unrated, Limited and Full NHL history: How much NHL data is behind a skater's rating. Full history: past 5-on-5 seasons inform it. Limited: rated, but the 5-on-5 part starts from a typical player. Unrated: no appearance history to rate.
Wild card: The two playoff spots in each conference for the best teams outside the top three of each division.
Win chance: The chance a team wins the game, including overtime and the shootout.
xG differential: Expected goals for minus expected goals against, per 60 minutes.
xGF and xGA: Expected goals for and against: the expected goals of the shots a team takes and allows (for a player, while he is on the ice).
xGF%: xGF ÷ (xGF + xGA): a team's share of the expected goals in its games (for a player, while he is on the ice). 50% is an even split, not a league percentile.