Fantasy football rankings you can check.

Every ranking here comes from a model tested on eight seasons of history before we trusted it. The code, the data and the methods are public — including the ideas that didn't work.

This week's boards Read the findings

Draft 2026 draft board

Season projections built from eight years of stats — usage, touchdown luck, age, draft capital, contracts. Tested against the draft market: it out-ranks ADP at QB, WR and TE.

beats ADP at 3 of 4 positions

Weekly Weekly boards

Who to stream at defense and kicker, ranked by the Vegas lines — the one signal that actually predicts those positions. Skill-position boards join in September.

D/ST & K live now

Research The findings

28 short write-ups on what wins fantasy leagues: what works, what's a myth, and which popular stats are noise. Each one shows its data and says how confident to be.

28 findings

Live Draft assistant

The draft-day tool: a live board that syncs picks straight from your ESPN or Sleeper draft room, flags value, and suggests who to take.

launches full screen →

We publish our misses. When an idea fails the test, the write-up still goes up — about a third of the findings are "this doesn't work" results (WR-vs-cornerback matchups, "injury-prone" labels, team ratings). That's how you know the wins are real.

Weekly boards

Ranked by opponent's implied total — the score Vegas expects the other team to put up. Lower is better. Why this works →

Same list for ESPN and Yahoo — here's why

ESPN and Yahoo score defenses differently. Yahoo pays more at the extremes: a shutout is worth +10 there vs +5 on ESPN, and getting blown out costs as much as −10 vs −6. But both systems reward the same thing — facing a bad offense — so when we re-ran our model under each system, the order barely moved. Your point totals will differ by platform; the ranking won't.

2026 season board

veterans only — for rookies, follow ADP until our rookie model ships

The 2026 proj PPG column is what this model expects each player to average per game over the 2026 regular season. It reads last season's usage, age, contract and team, and it was graded on seasons it was never trained on. How it works, and how often it beats the draft market →

Scoring

Kickers and defenses, on draft day

Take them last, and don't think hard about it. Ranking kickers by how well they kicked last year predicts almost nothing — a rank correlation of .08, where 0 is a coin flip — and telling our model who the kicker is actually makes it slightly worse. Kicker points belong to the offense, not the guy. Defenses are a little more predictable, but the only thing that matters there is who they play. So use these two lists to break a tie at the end of your draft, then move on. Kicker model → D/ST model →

Kickers — best offenses to own

Defenses — easiest schedules

What Vegas expects from every 2026 offense

Add up the betting lines for all 272 games on the 2026 schedule and you get an expected points-per-game for every offense (snapshot: Aug 2026). More team points = more fantasy points to go around. Use it as a tiebreak — when two players are close, take the one in the better offense.

Research

newest first · every finding shows its work

The stats we use, in plain English

Spearman / rank correlation: How well two rankings agree, from −1 to 1. 1 = identical order, 0 = no relationship. Our usual yardstick, since fantasy is about who finishes ahead of whom.
Pearson correlation: How closely two sets of numbers move together, −1 to 1. Like Spearman, but it also cares how big the gaps are, not just the order.
Implied total: The score Vegas expects a team to put up, worked out from that game's point spread and over/under.
ADP: Average draft position — where real drafters are taking a player. The market's opinion of him.
All-play: Your record if you played every team in the league every week. Measures team strength with schedule luck stripped out.
Backtest / LOYO: Testing a model on past seasons it never saw while being built ("leave one year out"), so it can't grade its own homework.
MAE: Mean absolute error — how many points a prediction missed by, on average.
95% CI: The margin of error: the range the true number sits in, 19 times out of 20.

← all research