How it works

The Aerie Power Index rates each team by the points it should score in a typical week. In fantasy football your opponent can't stop your players, so the points a team scores are the best evidence of how good it really is. A 3-4 team that keeps scoring 140 is better than a 5-2 team that keeps scoring 120 and getting lucky.

The rating updates automatically within about 15 minutes of ESPN marking a week final, usually early Tuesday morning ET. Every week is kept in the archive.

• Every score this season counts the same. The rating is the average of a team's weekly scores, regular season and playoffs.
• Pulled toward the league average. Weekly scores swing a lot: a typical team's score moves about 18% from week to week (about ±24 points this season), while the real gap between good and bad teams is only about 7% (about 9 points). So each team starts at the league average, worth about 7 weeks of scores, and moves away from it only as its own scores pile up. Early in the season every team is close to average.
• No opponent adjustment. Your opponent doesn't change your score, so there's nothing to adjust. Strength of schedule matters only for wins, and the forecast handles that by playing the real schedule.
• Luck is actual wins minus the wins a team would expect from its all-play record (how it would do against every team, every week).

• We play out the rest of the season 10,000 times. In each run, every team first gets a "true strength" drawn from its rating's margin of error, then every remaining game gets a score for each side drawn around that strength.
• The regular season is ranked by ESPN's settings for this league: record first, then head-to-head among the tied teams, then points for. We checked that rule against ESPN's own seeds for all 15 past seasons: it picks the right playoff field in 15 of 15 and every seed exactly in 10. When the regular season is over, ESPN's real seeds are used.
• The top 6 make a fixed bracket (no reseeding) and the top 2 get byes. Playoff games already played use the real result.
• From the runs we count projected wins (with the 10th–90th percentile range), the odds of making the playoffs, getting a bye, the #1 seed, reaching the final, winning the title and finishing last.

We replayed every past Aerie season week by week, using only games already played to predict that week's matchups. Settings were tuned on 2011–2017 and then tested on 2018–2025 (615 games), which the tuning never saw. Log loss scores the probabilities (lower is better; a coin flip scores 0.693). Picked is the share of winners called correctly; week 1, when every model says 50/50, counts as half.

Season points per game turns the gap in points per game into a probability, and current record does the same with win %; both were fitted on 2011–2017 too. The Aerie Power Index is best on both measures, but the margin over points per game is small (0.681 vs 0.684) and within the noise of 615 games. The honest summary: in this league the best predictor is points scored, and record adds almost nothing.

Log loss with picked % below. Weeks 5–9 were close to a coin flip for every model in the test years; the rating earns its keep late in the season, once it has enough scores.

We also ran the full forecast after weeks 4, 7 and 10 of every season 2018–2025 (24 forecasts), hiding everything after the cutoff, and compared the odds with what happened. Brier score measures probability errors (lower is better).

Playing out the real schedule is what matters: both simulations beat simply assuming today's top teams make it. Adding team ratings gives slightly better win totals and ranges, but playoff, bye and title odds are no better than treating every remaining game as a coin flip. In a league this even, the standings and the schedule tell you most of what can be known.

• Recent weeks weighted more (half-life 4, 8 or 12 weeks). No gain on 2011–2017; every half-life tried did worse than weighting all weeks equally.
• Owner's past seasons as a starting point (last 1 or 3 seasons). No gain: test log loss 0.6814 with none, 0.6815–0.6993 with it. Last year's team strength does not carry over in this league.
• ESPN player projections / roster strength. Not tested. ESPN keeps no week-by-week projection history for past seasons, so there is nothing to backtest against.

• Uses only this league's ESPN data: weekly team scores, the schedule, the playoff settings and ESPN's seeds. Nothing is estimated where data is missing.
• Settings were tuned on the 10-team, standard-scoring years (2011–2017). The league moved to 12 teams in 2021 and full PPR in 2025; the model works in proportions of the league average, so it carries across scoring changes.
• Not modeled: injuries, byes, trades, waiver pickups and lineup mistakes, except as they show up in later scores. Ties are treated as impossible in simulations (there has never been one in this league).