How Ratings Work

The Rating Scale

50 is league average for that position in that season. A rating of 75 means "significantly above average" — roughly top 15% of starters. A rating of 90+ is elite, reserved for the best handful of players at their position that year.

The Formula

The z-score measures how many standard deviations a player is above or below the league average for their position group that season. The multiplier (22) controls how spread out ratings are — it's tuned so that roughly 2 standard deviations above average yields a 94.

Per-Game Normalization

All counting stats (yards, touchdowns, sacks, tackles, etc.) are converted to per-game rates before z-scoring. Both the league baselines and the player values are divided by games played, so the comparison is always rate-vs-rate.

This means a QB who plays 6 games due to injury is rated on his per-game production, not his season totals. A 6-game QB averaging 280 yards per game rates the same as a 17-game QB averaging 280 yards per game.

Position Breakdowns

Each position has its own set of skills, each mapped to one or more real stats. Here's what drives every rating.

Pre-1999 Proxy Stats

Advanced metrics like EPA, CPOE, YAC, and target share weren't tracked before 1999. Rather than defaulting every affected skill to 50, the pipeline uses proxy stats — derived values from available counting data that approximate what the advanced stat measures.

Defensive stats, kicker stats, and punter stats don't exist before 1999 at all, so those positions remain at 50 for pre-1999 seasons.

Universal Attributes

Every player receives three universal ratings that apply across all positions.

Speed

Speed is derived from real NFL Combine 40-yard dash times when available (7,800+ players since 2000), with three fallback paths for players without combine data.

Age decay is applied on all paths: no penalty before age 27, then -1/year through 29, -2/year through 32, and -3/year beyond 32. A player who ran 4.4s (base 92) at the combine rates around 83 by age 32.

Stamina & Intelligence

Overall Rating

The overall rating is a weighted average of position skills and universal attributes. Each position has its own weight table emphasizing the most important skills.

For example, QB accuracy and intelligence are weighted heavily, while a RB's power and elusiveness carry more weight than receiving. Defensive linemen use separate weight tables for EDGE (pass-rush emphasis) and IDL (run-stuffing emphasis). The weights ensure that a player's overall number reflects what matters most for their role on the field.

XvO Factor

Beyond stat-based ratings, we measure each player's on-field impact using Regularized Adjusted Plus-Minus (RAPM) — a ridge regression technique adapted from basketball analytics — then normalize it to a 1–99 scale.

How It Works

For every scrimmage play from 2020–2024, we know all 22 players on the field and the play's Expected Points Added (EPA). A ridge regression estimates each player-season's marginal EPA contribution per snap, controlling for every other player on the field simultaneously.

The raw EPA/snap values are converted to XvO Factor using z-scores within each position group per season: 50 + (z × multiplier), clamped 1–99. Crucially, the multiplier varies by position — see Reliability-Scaled Spread below.

Reliability-Scaled Spread

Not all positions are equally measurable. RAPM reliably differentiates quarterbacks — the player who touches the ball every snap — but struggles to separate a running back's contribution from his offensive line's. To keep XvO Factor honest, we scale each position's spread by how much signal the model actually carries.

We measure this with split-half reliability: split all plays into odd-week and even-week halves, run the regression independently on each, and correlate the player estimates. High correlation means the model is measuring something real. Low correlation means the differences are mostly noise.

The formula is multiplier = 22 × reliability. A QB with reliability 0.56 gets a multiplier of 12.4, spreading scores across a wide range. A linebacker with reliability 0.17 gets a multiplier of 3.7, honestly compressing scores around 50. This means a QB XvO Factor of 85 and a LB XvO Factor of 60 both represent approximately the same confidence: “among the best we can measure at this position.”

Limitations

The ELITE Badge

The ELITE badge identifies players who were among the very best at their position in a given season. It draws from three independent sources, layered by era:

A player only needs to qualify through one of these paths to earn the badge. For 2020+ seasons, all three sources overlap and reinforce each other. For earlier eras, the badge relies on All-Pro selections and major awards.

Data & Limitations

All stats come from nflverse (nflfastR/nflreadpy), licensed under CC-BY 4.0. The dataset covers 1970–2025.

Team Ratings

Team ratings combine individual player evaluations with actual game results to produce four distinct measures per team-season.

Talent Rating (1–99)

A weighted composite of the 8 unit ratings (Pass Game, OL Pass, Run Game, OL Run, Pass Rush, Run Defense, Secondary, Special Teams). The raw weighted average is z-scored and mapped to 1–99. Weights shift across four strategic eras to reflect how the game was actually played:

Results Rating (1–99)

Derived from actual game outcomes, z-scored within each season. For 1999–2025, game data comes from nflverse. For 1978–1998, historical schedule data (scores and playoff results) is sourced from the devstopfix/nfl_results dataset (Public Domain). Pre-1978 seasons have no schedule data available.

Where play-by-play data is not available (1978–1998), the EPA per play component naturally drops out (all teams z-score to 0), so Results depends on point differential, win percentage, and playoff depth.

Playoff multiplier values: Missed playoffs (−0.5), Wild Card loss (0), Divisional loss (+0.3), Conf. Championship loss (+0.6), Super Bowl loss (+0.4), Super Bowl win (+2.5). The large gap between winning and losing the Super Bowl ensures that regular season dominance alone cannot produce the highest ratings.

Clutch Factor

Clutch = Results − Talent. Measures whether game outcomes exceeded roster quality. Displayed as a signed number with a label:

Playoff Fate

A separate measure of whether a team's postseason matched the expectations set by their regular season. A 14-2 team losing in the conference championship is a different kind of disappointment than a 9-7 team losing in the same round — Playoff Fate captures that distinction.

Teams are classified into a record tier based on their win-percentage rank within their season, then matched against their playoff result to produce a label. By ranking teams relative to their peers each year rather than using fixed win totals, this adapts naturally to different schedule lengths and competitive landscapes — a 12-4 record in a 14-game era is treated as elite when it leads the league.

Overall Rating (1–99)

The Clutch modifier is clamped at ±5 points so extreme over- or underperformance nudges the composite without dominating it.

Pre-1999 adjustments: For 1978–1998 seasons, Talent is measured from fewer differentiating units (4 of 8), making it less reliable. The Overall formula shifts weight toward Results: T × 0.30 + R × 0.65 + clutch. This ensures that great teams with unmeasured defensive talent (like the 1985 Bears) are properly rated through their dominant game results.

Pre-1978 (no schedule data): Overall regresses Talent toward league average: Talent × 0.70 + 50 × 0.30. This caps talent-only teams around 79, reflecting the inherent uncertainty of roster ratings without any game outcomes to validate them.

Draft Value Analysis

The Draft Value page answers a specific question: given a draft position and a player position, what does historical data say about the value of that pick? It analyzes every draft pick in rounds 1–3 from 2000–2023 (~2,300 players) across 11 positional groups.

First-Contract Production

For each drafted player, we sum their per-season XvO player ratings (1–99) over the rookie contract window:

An elite pick might total 350+ over the contract window; a bust might total under 150. Players with fewer seasons get their actual total — no imputation. Busts, injuries, and early exits are real outcomes reflected in the data.

Contract Cost

Two eras require different cost models, both normalized to a 0–100 scale:

Surplus Value & Grading

Positive surplus means the player outproduced their draft slot cost; negative means they underperformed relative to the investment. The surplus is mapped to a letter grade:

Outcome Classification

Each pick is classified into one of three outcomes based on first-contract production:

Elite Impact Multiplier

Measures how much having a top-rated player at a position correlates with team win improvement. For each position, we compare the average win-percentage residual (actual wins minus expected wins from talent rating) for teams with an elite player at that position versus teams without. Granular positions are resolved from the database using depth chart data (e.g., DE vs DT, T vs G) rather than the generic DL/OL categories in roster files, giving each positional group its own distinct signal.

Positional Groups

11 groups, granular where sample size supports it, merged where it doesn't:

Draft Tiers

Six tiers aligned to CBA-era rookie wage scale cost cliffs:

Data Scope & Limitations