DEVIATEThe score is only part of the situation
A lead means something different with a quarter left than it does with a minute left. Good comparisons need context.
Two games can show the same score difference and still describe very different situations. A team with most of the game left has more time to lose a lead than a team near the finish. In some sports, who has the ball or who is batting changes the situation too.
That is why Deviate’s comparison starts with game state. The aim is to ask how comparable situations resolved, rather than treating the scoreboard as the whole story.
Time changes the comparison
For NBA and WNBA basketball, the historical lookup uses the quarter, time remaining, and score difference. The same margin at different points in the game belongs to different states. Those tables average over possession; they do not use a live basketball possession flag as an input.
Keeping that distinction clear matters. A model should be described by the inputs it actually uses, not by every detail that might be visible on a broadcast.
The useful inputs depend on the sport
For MLB baseball, the state includes the half-inning, outs, base runners, and score difference. The same score with empty bases and with runners on base is not the same comparison. The model uses decades of play-by-play outcomes to find comparable situations.
For NFL football, the quarter, clock, score difference, and possession are part of the historical lookup. Possession must be known for an NFL alert. Other visible details, such as down and distance, are not separate inputs to that lookup.
For tennis, sets already won and the game score in the current set matter, along with tour and tournament context. For soccer, the model reads the minute, both sides’ goals, and red cards. These are different ways of describing what has already happened and how much opportunity remains.
History needs enough comparable examples
Making a state more specific can also leave fewer historical examples. For the measured models, a state without enough supporting history produces no alert. The absence of a number is sometimes the appropriate result of a comparison.
Coverage also varies. Hockey and some basketball and football leagues use estimate formulas. Golf, motorsports, and cricket currently appear in Live without live model alerts. Being able to browse a market does not mean a model is evaluating that sport.
Context still has limits
A game state does not describe every fact about the players or the event. A historical comparison is useful because it puts the situation in context; it does not make the next outcome certain or explain every market movement.
When you open an alert, read the situation and the explanation together with the two probabilities. The Sports page lists the current inputs by sport, and How it works explains how the comparison becomes a notification.