Methodology
How 94×50 turns box scores and schedules into ratings, predictions, and schedule strength — in everyday language.
What is Elo?
Elo is a rating that moves after every game. Beat a strong opponent and you gain more; lose to a weak one and you lose more. It’s built from wins and losses only — not from how many points you scored.
After each result, both teams’ ratings update. The math is tuned so that over time, the gap between two teams’ numbers matches how often we’d expect each side to win.
On 94×50, Elo numbers sit in a band around 1300–1700. Higher means stronger. The league average drifts toward the middle of that range over a season.
Think of Elo as “who’s winning and who they beat.” It doesn’t care how you won — only that you did.
What is SRS?
SRS starts with point differential: how much you outscore opponents by per game. Then it adjusts for how hard your schedule was. Beating good teams by a little can mean more than blowing out bad teams by a lot.
On the site, SRS is shown in points per game (roughly). Positive means you’d expect that team to outscore an average opponent; negative means the opposite.
Where Elo is “who won,” SRS is a bit closer to “how the games looked.”
Recent form
“Form” is a read on how a team has played lately, using margins in recent games. We use a 15-game window and weight recent games more heavily than older ones (exponential weighting), so a hot streak matters more than a game from six weeks ago.
Form is a tie-breaker of sorts: it captures momentum and fatigue that the raw record might miss.
Combined rating
The power ranking number is a blend of three pieces:
- 40% Elo
- 40% SRS
- 20% Form
League average is set to 0. Positive means above average; negative means below. The blend keeps one signal from drowning out the others.
Home court
Playing at home is worth a few points in the model — travel, crowd, and sleep in your own bed add up. We apply a single league-wide bump for the home team in predictions, tuned on historical data.
In a future version, we may split that by team or venue when we have enough signal to do it fairly.
Strength in predictions
Each game prediction compares the two teams’ combined ratings. The wider the gap, the more the model leans toward the team with the higher number. That “strength” line on game cards is the piece of the spread that comes from talent alone, before rest and travel.
Rest and back-to-backs
Days off matter. The model gives a small edge to the team that’s rested more when the gap is meaningful. If one side played yesterday and the other didn’t, that shows up in the rest factor.
Back-to-backs are the clearest case: less recovery time, sometimes travel, and a higher chance of fatigue.
Player availability
Injuries and rest days for stars can swing a game. That layer is planned for a future release. The model will fold in who’s expected to play once that data is reliable enough to ship.
How SOS+ works
SOS+ is a schedule strength view for what’s left on the calendar. It’s not just opponent win percentage — it’s built from how strong our model thinks each team is, including home and away context.
The “weighted SOS” number answers: how tough is the path ahead, given who you play and where? Expected wins over the remaining slate uses the same ratings — so it’s consistent with the rest of the site.
What makes it different from a simple strength-of-schedule table is that it ties directly to the team ratings you see on rankings and game cards, not a separate stat.
Calibration and accuracy
We check predictions against real results: when the model says a team has a 70% chance to win, how often do they actually win? If those line up over time, the model is well calibrated.
On a full-season holdout (2023-24 validation), the model’s win–loss accuracy was about 67.5%, with average error on the margin of about 10.7 points. Live numbers for the current season update on the accuracy page.