How we build the forecast
A Forecheck forecast is a team's chance to win a KHL game, including overtime and the shootout. We show it as a pair: “62% vs 38% for Barys”. It comes from a model built on KHL data, not from an expert's gut feeling — so every forecast has a “why” breakdown: how many percentage points each factor added.
What goes into it
Chance quality. How many dangerous chances a team creates and allows — by expected goals (xG) at even strength. Expected goals come from our shot model: from the location and type of every shot — how often shots from that spot go in in the KHL. This is half of a team's strength.
Season goal differential. The other half is actual goals, adjusted for special teams: teams spend different amounts of time on the power play and penalty kill, and we even that out.
Early season. While few games have been played, we also lean on last season — it counts as 10 games and gradually gives way to this one.
Goalies. How much a goalie has saved above or below expected over his last 10 games, anchored to last season. A change in net can move the forecast by several points.
Home ice. Home teams win more often in the KHL — this is always included.
Morning and with lineups
The first forecast comes out on the morning of game day, with the expected goalies. About an hour before puck drop the KHL publishes the starting lineups: as soon as they're out, we update the forecast with the starting goalies. Until then the game page shows the expected lineup: who has played most often in recent games.
The forecast never changes after puck drop
Every forecast is stored with a timestamp and never edited. Lineups that arrive after the opening faceoff are ignored. Every forecast — hits and misses — is in the forecast record. The morning forecast is what counts; the lineup update is shown separately so you can see whether lineups help.
How accurate it is
We measure accuracy with the Brier score — the mean squared error over all games: 0 is perfect, 0.25 is a coin flip, lower is better. We don't judge single games: any one game is a lottery.
On past seasons (each tested without the model having seen it) the score is 0.221 in 2024/25 and 0.229 in 2025/26. That's a good level for hockey: there's a lot of randomness, and even the best models are far from zero. Live accuracy since 21 October 2026 is in the record, next to the bookmakers'.
The market
Next to our forecast is the market's view: bookmaker odds converted into a probability. We remove the margin and split the chance of a draw in regulation between the teams in proportion to their chances — giving the same quantity as ours: a win including overtime and the shootout. When we differ from the market by 8 points or more, the game is tagged “Us vs the market”.
What we don't promise
We don't promise to “beat the bookmakers” and we don't give betting advice. A forecast is our estimate of the chances and an explanation of what it rests on. We show the full history of forecasts, not the best hits: a bad first week proves nothing, and neither does a good one.
Other numbers on the site
Openness — how many slot shots to expect from both teams combined: how each team creates and allows in the slot this season. Points vs play — how many points a team has earned above what its chances suggest: from each game's xG we estimate the chances to win, lose or go to overtime. Form in the standings — the last 10 games grouped into home and road runs.
Expected goals on game, player and team pages account for rebounds from the crease. Empty-net goals and shots are left out wherever goals are compared with expected goals (finishing, goals above expected, goalies).