Position-relative inputs
Quarterbacks, receivers, linemen, defenders, and specialists are judged with different statistical contracts. Raw fantasy points do not become an overall rating.
METHODOLOGY
Every grade needs
a receipt.
NFSMELL is a statistical rating system, not a film grade and not a reputation poll. The public pages expose the inputs, sample, position rank, and model version needed to challenge a result.
Model version: player-forecast-v24 Data updated: August 24, 2026 at 6:13 PM EDT
PLAYER ENGINE
How a player rating is built
Percentile scoring
Each qualifying metric is compared with the distribution for that position and season. Lower-is-better inputs, such as turnover or missed-tackle rate, are inverted before weighting.
Sample shrinkage
Small samples move toward a conservative position baseline. More qualifying plays allow observed performance to carry more of the score.
Forecast and resume
QB Index blends a validated QBR projection, opponent-adjusted current form, and a maturity-shrunk NFL resume. Non-QBs use position-specific production forecasts where validation supports them.
One public overall
The canonical player OVR is the number used on rankings, team rosters, and player pages. Current-team impact remains a separate internal roster-building input.
TEAM ENGINE
Roster strength becomes matchup strength
The team model weights the positions that influence offense, defense, and special teams instead of taking a flat average. Injury status can reduce a player's current-team contribution without rewriting his underlying ability rating.
Current roster strength is calibrated around the live league distribution. When opponent-adjusted team performance is available, it receives up to 30% of the final team grade based on confidence. The matchup model then adds home field, availability, rest, weather, and market context where those inputs are available.
Prediction quality is evaluated only from snapshots captured before kickoff. Records captured at or after kickoff are rejected to prevent future-data leakage.
DATA SOURCES
What feeds the model
nflverse play data
Play-level efficiency, success, usage, regular season, and separately weighted postseason evidence.
ESPN context
Total QBR, team news, schedule, standings, roster identity, and injury context when available.
FTN charting
Charted quarterback mistakes and context, including interception-worthy throws and QB-fault sacks when joined coverage is available.
Model receipts
Version IDs, component scores, qualifying samples, and source notes are retained with the finalized rating snapshot.