Accuracy Methodology

What is measured

The Fan50 accuracy engine measures relative ordering of plausible fantasy starters within a position, not fantasy league wins or overall expertise. Accuracy is format-specific (half-PPR) and tied to the weekly lock policy below.

Guides, podcasts and narrative columns aren't scored as predictions.

Weekly lock

All experts are evaluated using the most recent valid ranking snapshot observed five minutes before the first scheduled NFL game of that scoring week. This prioritizes comparability and auditability over representing final Sunday lineup advice.

Limitation: injuries after the lock remain part of this model's real-world limitation. We don't retroactively remove bad predictions.

Evaluation universe

A predeclared, versioned weekly reference ranking determines the player pool: top 24 QBs, 48 RBs, 60 WRs, 24 TEs, 60 FLEX players (running backs, wide receivers and tight ends ranked together), 16 kickers and 16 team defenses, excluding bye-week players. The universe is stored before outcomes are known.

A ranking is scored when it includes at least 75% of the pool (18 of 24 QBs or TEs, 36 of 48 RBs, 45 of 60 WRs or FLEX players, 12 of 16 kickers or defenses). Players it leaves out share a tied bottom rank, and the omission count is recorded. Shorter previews are captured but not scored, because a tied bottom rank costs too little to compare them fairly with full boards.

Kicker and defense points

Reception scoring doesn't apply to kickers or defenses, so they use one fixed set of rules:

Kickers: field goals of 0–39 yards 3 points, 40–49 yards 4, 50 or more 5; extra points 1; each missed or blocked field goal or extra point −1.
Team defense: sack 1, interception 2, fumble recovery 2, defensive or return touchdown 6, safety 2, blocked kick 2, plus points allowed: 0 scores 10, 1–6 scores 7, 7–13 scores 4, 14–20 scores 1, 21–27 scores 0, 28–34 scores −1, 35 or more scores −4.

Points allowed is the opponent's final score, including points scored against the offense. Because the index compares orderings, a different but consistent rule set changes results only where it changes which player finished ahead.

Fan50 Order Index

For each position universe of N players, predicted and actual ranks are converted to percentile ranks, (rank − 1) / (N − 1). The mean absolute percentile error is computed over all N players. The index is:

Fan50 Order Index = 100 × (1 − mean absolute percentile error)

In words: if an expert ranked every player exactly where they finished, the error is 0 and the index is 100. Larger gaps between ranked and actual positions lower the index. This is a custom ordering index, not the percent of predictions that were correct.

Skill score

Results are shown as a skill score: the Order Index rescaled so a random ordering scores 0 and a perfect one scores 100. A random order of N players has an expected mean absolute percentile error of (N + 1) / 3N, so it scores about 65 to 66 on the raw Order Index rather than 0, and every real ranker lands just above it. Rescaling against that baseline shows the real gaps between rankers at their true size without changing their order.

baseline = 100 × (1 − (N + 1) / 3N)
skill = 100 × (Order Index − baseline) / (100 − baseline)

The baseline uses each position's standard pool size (24 for QB and TE, 48 for RB, 60 for WR and FLEX, 16 for K and DEF). A score below 0 means a ranking did worse than a random order, which can happen when a board leaves out players who then score. Gaps to the consensus are shown on the same scale. The raw Order Index remains in published data and evidence.

Right when they disagreed

Most of any ranking agrees with the consensus, which says little about the ranker's own judgment. This measure looks only at pairs of players the ranker ordered differently from the FantasyPros consensus, and counts how often the player they put higher actually scored more. Pairs whose players tied in points are skipped. 50% means their disagreements were no better than going with the consensus; it's shown once a ranker has at least 30 such pairs at a position over the season.

Baseline comparison

A baseline index is computed from the reference rankings under the same rules, so a high-looking index can be read against what the reference ordering achieved.

Season ranking gates

A season position score requires at least 6 eligible weeks and participation in 75% or more of scored weeks. Below this gate, results are shown as unranked partial samples without a season rank. Minimum sample gates limit small-sample claims; they don't establish statistical confidence. Past results don't establish future superiority.

Published methodology versions

Version labels come from completed, published score runs. A label change marks that results may use different scoring rules.

Version	First published	Example cohort	Code version
v1	2026-10-28	2026 QB	v1

Category leaders

Category leaders compare rankers with the FantasyPros consensus, which serves as the benchmark rather than a contestant. Tiers are drawn from the consensus order captured before kickoff, never from results.

Tier categories (for example deep running backs, ranks 13–48) apply the Order Index to that tier alone, using each ranker's order of those players. A ranker needs 6 scored weeks in the tier.
Bold calls are players a ranker placed 5 or more spots from the consensus. A call is right when the player finished closer to the ranker's spot than to the consensus spot. A ranker needs at least 50 calls.
Predicting usage grades each QB, RB, WR and TE board against expected fantasy points instead of actual points: what each player should have scored from their targets, carries and field position, which removes much of a single week's luck. Expected points come from ffopportunity (ffverse), a model built on nflverse play-by-play, converted to half-PPR. Each week is scored as the gap to the consensus's own usage index at that position, then averaged, so rankers who cover different positions compare fairly. A ranker needs 12 position-weeks.
Most weekly wins counts the position-weeks in which a ranker had the highest Order Index, excluding the benchmark.
When no ranker beats the consensus in a category (or, for bold calls, no ranker is right more than half the time), the top ranker is labeled “closest to the consensus” rather than a leader.

See the standings · Review methodology · Report a correction

How Fan50 measures ranking accuracy