Methodology

Fantasy projections, and what we will and will not claim

We anchor every projection to the preseason expert consensus, wrap it in a range we actually test, and re-rank the whole board for your exact league. We do not claim to beat the consensus or the draft market, and the section below is the receipt for why, including a claim we had to take back.

Correction · June 2026

An earlier version of this page claimed our board beat average draft position "by a wide margin" and beat the expert consensus on accuracy out of sample. Both were wrong, and we are leaving this note instead of quietly deleting them. The historical projection feed we graded against turned out to be contaminated: the vendor's "preseason" numbers were silently updated during the season, so they correlated about 0.80 with final results when a real preseason number correlates about 0.65. They had effectively seen the future. When we re-ran the draft test against an independent, genuinely pre-draft source, drafting by consensus lost to a real-ADP field rather than beating it. So we retracted both accuracy claims, removed the bias-correction layer that was tuned on that contaminated feed, and now ship a straight composite of the major preseason sources. What survives is everything that never touched that feed: the tested ranges, the league-aware ranking, and the custom scoring.

The answer

The number you see is the preseason expert consensus for that player, in component stats so your league's scoring applies to a real stat line. We do not add a secret accuracy edge on top, because we tested for one across six clean seasons and could not honestly find it: the expert consensus ranks players better than our own model (within-position rank correlation 0.70 vs 0.60, in every season 2019 through 2024), and blending our model in makes it slightly worse, not better. What we add is the part the consensus leaves out. Every player ships a floor-to-ceiling range from 200 correlated season simulations, and that range method is calibrated: on held-out 2019-2024 seasons the 80% range contained the real outcome about 80% of the time. Rankings are value-over-replacement for your league size and roster, so an elite tight end can outrank a higher-scoring quarterback in a one-quarterback league. And we price your per-game bonuses, which a single-number projection cannot. We are honest about the rest: the draft market (ADP) is efficient and we do not beat it.

What we tried, and why we stopped trying to beat the crowd

We started where everyone starts: build a player model from opportunity and efficiency, project the box score, score it. It beats a naive "repeat last year" baseline (within-position rank correlation 0.60 vs 0.58). The honest question is whether it beats the consensus, and we tested that the clean way: against a genuinely pre-draft expert consensus (FantasyPros August rankings, which we contamination-checked), across six held-out seasons. The answer is a clear no. The consensus ranks players better than our model in every season 2019 through 2024 and at every position (0.70 vs 0.60 overall), and a 50/50 blend of the two is slightly worse than the consensus alone, so our model does not even add signal at the margin. A hundred analysts reading the same depth charts and injury news price a player better than any single model, and averaging them is hard to beat.

So we stopped selling an accuracy edge we do not have and pointed the work at the three things the consensus does not give you: a tested range, a ranking built for your league, and correct treatment of bonus scoring. Those do not require beating the crowd. They require being honest about uncertainty and doing the bookkeeping the crowd skips.

What we actually ship

The number: a multi-source consensus

The base projection is our own composite of the major preseason projection sources (Sleeper, ESPN, and FantasyPros), averaged at the component-stat level (pass yards, receptions, rushing touchdowns, and so on). Two honest reasons we blend rather than copy one source, and one thing we will not claim. We blend for robustness: averaging several sources keeps any one source's odd take on a player from dominating. And we blend at the component level so your league's scoring applies to a real stat line and the published number is our derived composite rather than any provider's figure. What we will not claim is that the blend is more accurate than the best single source. We tried to verify that and could not: the only clean, genuinely pre-draft historical data is a single source, and the live sources agree with each other about 95% of the time, so the averaging is a de-noising step, not a proven edge. Whether the blend actually ranks better than its best component is something we can only learn from this season's real results, which is why we froze the 2026 board to grade in January. We also do not add a bias correction or a model tilt: we tested both the clean way and neither helped.

The range: 200 correlated simulations

For every player we run 200 correlated season simulations. Each one resamples that player's real game-to-game lines (so receptions, yards, and touchdowns move together the way they do in a real big game), draws a games-played count from his own availability history (the injury left tail), and applies a season-level shock from the historical spread of how far players land from their baseline. Scoring all 200 under your settings gives the floor, median, and ceiling on the board.

This is the one part of the system with a clean predictive test, because it never touches the vendor projection feed. On held-out 2019 through 2024 seasons, the stated 80% range (10th to 90th percentile) contained the actual outcome 79.8% of the time, and the 50% range 49.0% of the time, both close to ideal — and re-centering the bands on an external clean consensus held the 80% coverage at 81.7%. Two honest caveats: that external check is on a thin, quarterback- and veteran-heavy sample, and the coverage is an aggregate — per-position the bands run slightly wide at quarterback and slightly narrow at receiver, and they are least tested on the hardest tail (rookies and players who changed teams). We re-verify and re-grade this calibration each release rather than treat it as settled.

How predictable is each position

Not every position's ranking is equally trustworthy, and we measure exactly how much. On the same leakage-clean 2019 through 2024 splits, we scored how well a preseason rank predicts the actual finish, within each position. Quarterbacks come out the most reliable (rank correlation 0.74), then wide receivers (0.71), then running backs (0.69), with tight ends the least reliable (0.66). Tight end is the noisiest position to rank league-wide, which matches what every drafter already feels.

We surface those numbers on the board, plus a per-player range tag, so you can see how much to trust a given ranking and how volatile a specific player is. But this is reliability information, not a draft-order lever. We tested whether tilting the board toward high-confidence players improved outcomes, and it did not: high-confidence players were not more likely to beat their projection, and a confidence-weighted draft order did not beat the straight value-over-replacement order. So we show the confidence and we do not reorder by it. A wide range or a low-reliability position is a prompt to make your own call, not a verdict that the player is worse.

The ranking: value over replacement, for your league

The board does not rank by raw points. It ranks by value over replacement: a player's projection minus the last startable player at his position, given your league size and starting lineup. That is why the order shifts when you change settings, and why position scarcity, not raw point totals, drives the early rounds. The replacement line we used for each position is shown on the page.

The bonuses: priced from the simulations

If your league pays a bonus for a 100-yard game, the average stat line almost never crosses that line, so a points projection quietly assigns roughly zero bonus. We instead estimate how many bonus games to expect from the simulations, so a boom-prone player gets the credit a mean projection erases. Turn the bonus dials on and the board re-ranks.

What we do not claim

We do not beat the consensus on accuracy. We do not beat average draft position; ADP is an efficient pre-draft market, and when we tested drafting by a clean preseason signal against a real-ADP field across the seasons we could test cleanly (2021-2024, one ADP source, best-ball scoring), the signal lost. We removed the bias-correction layer rather than ship a number we could not validate. Rookies and players with no recent game history carry the consensus number with a position-typical range and are flagged on the board, because we have no signal of our own on them yet.

The only honest path to ever claiming an accuracy edge is forward, not backward. We are freezing this 2026 board now and will grade it against real 2026 results in January, with a projection feed we capture and timestamp ourselves so the same vendor leak cannot recur. If that clean forward test shows an edge, we will publish it. Until then, the claim is exactly what is above and no more.

Sources

Player projections and average draft position come from public fantasy-industry feeds. Historical game logs, rosters, and availability come from the nflverse public data set. Player identities are crosswalked across feeds through the open DynastyProcess id map. We ingest and transform these; we do not republish any provider's raw figures. The range calibration above is reproducible from held-out season splits.