What is real and what is simulated
The table or standings you start from are the real ones, and every game already played keeps its real score and cannot be edited. You can type a score into any game still to be played. The Simulate buttons fill in the season, or just the current round, with the model described below. Clear takes everything back to the real standings.
A simulation is one possible season, not a forecast. Run it twice and you get two different seasons, because the model draws each result at random from the range of results two teams like these produce.
Soccer
Each team has a rating centred on zero. The difference between two ratings sets how many goals each side is expected to score, and the actual scoreline is drawn from a Poisson distribution, the shape goals follow in real matches. Two equal teams average 1.50 goals for the home side and 1.25 for the away side, the long-run Premier League figures, so home advantage comes from the data.
- In the top leagues, a team's rating comes from its finishing position last season and the market value of its squad.
- In the lower divisions it comes from last season's tables across the whole pyramid, with a step between one division and the next.
- Tables are sorted by points and then by each league's own tiebreak: head-to-head first in La Liga, LaLiga 2, Serie A and Serie B once the level clubs have met, goal difference first everywhere else.
In the Carabao Cup there are no scores: you pick the team that goes through in each tie. Rounds that have not been drawn yet are drawn on the page at random, as there are no seeds after the early rounds, and a button draws them again.
NFL, NBA, WNBA, MLB and NHL
These leagues are decided by records, not goals, so the model draws a margin of victory and a total score instead of goals. Each team is rated from its point differential per game (goal differential in the NHL, run differential in MLB), and the difference between two ratings sets the expected margin. What the ratings cannot explain becomes randomness, measured from real games.
- NFL: home teams win 53.7 percent of games, by 2.07 points on average, measured over all 272 games of the 2025 season. Ties are possible, and scores are rounded to ones football can produce.
- NBA: home teams win 55.4 percent, by 1.72 points, over 1,239 games of the 2025-26 season.
- WNBA: home teams win 53.5 percent, by 1.94 points, over the first 269 games of the 2026 season.
- MLB: home teams win 51.7 percent, by 0.05 runs, over 1,921 games of the 2026 season.
- NHL: home teams win 52.2 percent, by 0.13 goals, over 1,313 games of 2025-26. The model sends 24.8 percent of games to overtime, where the loser still gets a point.
Standings are sorted the way each league reads them: by points in the NHL, by winning percentage elsewhere, then by wins and point differential. The simulators mark the playoff places, and the play-in places in the NBA; the full official tiebreaker procedures, such as strength of schedule, are not modeled.
College Football Playoff predictor
College football has no table to rebuild, because a 13-member committee ranks the teams, so this tool works differently and its ranking is an estimate. It starts from the committee's latest ranking, or the AP Top 25 before the first one, and turns each place into points: 30 for No. 1, down to 6 for No. 25. A win adds 0.8 points plus 0.12 times the opponent's points. A loss subtracts 3 points plus a quarter of the gap between the opponent's points and 30, or 10.5 against a team outside the list. A conference champion gets 3 more points, and losing a title game costs 1.5. To set the odds, it plays the remaining games 2,000 times, giving each home team a 2-point edge and using a logistic curve with a scale of 24. These weights are our judgment and have not been fitted to or tested against past seasons. The rules of the 12-team field come from the College Football Playoff itself.
AFL, NRL, NBL and rugby union
The ladder predictors of the AFL, the NRL and the NBL, and the table predictors of the Premiership, Super Rugby Pacific and the Six Nations, draw a margin and a total score at random around what the ratings expect. In the AFL, the NRL and rugby union each club is rated from its points difference per game: last season's at first, and the current season's once there are about two games per club, and replaying a season from a round never uses a result from after that round. The NBL, like the NBA, rates each club on its points difference per game over the whole of last season.
- AFL: home teams won 58.5 percent of games, by 7.1 points on average, over the 207 home-and-away games of 2026. The average game had 178 points, and three of those games ended level.
- NRL: home teams won 54.9 percent of games but by only 0.05 points on average, over the 204 regular-season games of 2026, so the model gives the home side almost no edge. The average game had 48 points.
- NBL: home teams won 50.3 percent, by 0.65 points, over the 165 games of the 2025-26 season. The average game had 182 points.
- Premiership Rugby: home teams won 64.2 percent, by 6.4 points, over the 53 games played from January to June 2026. Super Rugby Pacific: 63.9 percent, by 5.8 points, over the 83 games of 2026.
- Six Nations: home teams won 80 percent of the 15 games of 2026, too small a sample to trust, so the model gives the home side a four-point edge instead.
Draws are possible in the AFL, the NRL and rugby union, where the model keeps a small share of level scores; the NBL never ends level. Rugby tables need tries as well, because the bonus point depends on them: when you pick a winner, the model writes a score and a try count that fit it, at about one try per 6.8 points, and you can change either. The bonus rules are each competition's own: four tries in the Premiership and the Six Nations, three more tries than the opponent in Super Rugby Pacific, and a point for losing by seven or fewer in all three.
What the models leave out
Injuries, trades, lineups, rest days and weather are not in any model. A rating is a summary of how good a team was, not a scouting report, and that is the main reason a simulated season should be read as one plausible outcome rather than a prediction.
The blog's data pieces use the same models and say in a closing note how their numbers were built. Questions or corrections: futsimulator@proton.me.