PFF Run Blocking Metrics Explained vs NFL's AI Model
NFL Next Gen Stats spent this offseason building a transformer-based AI model that identifies the run concept and every individual blocking matchup on a play, rather than just recording how the run turned out, according to NFL.com. Three days later, PFF released its own new run-blocking numbers, positively and negatively graded play rates, built from PFF's play-by-play grades, according to PFF. If you're looking for PFF run blocking metrics explained in plain language, the distinction is simple: the NFL model infers assignments, while PFF reports grade-based rates.
Falcons guard Chris Lindstrom posted the top overall PFF positively graded play rate among qualifying linemen at 23.6%, while 49ers tackle Trent Williams posted the lowest PFF negatively graded play rate at 6.8%, according to PFF. Those figures come from PFF's grading system. The NFL's model works from something different: a transformer architecture built to interpret the spatial and temporal relationships among players throughout a play, according to NFL.com.
What the NFL's new AI model actually tracks
Next Gen Stats worked with the AWS ProServe team this offseason on what it called its most expansive effort yet to quantify the run game, aiming to identify the run concept and each individual blocking matchup on a given play. The resulting run-blocking matchup model identifies each offensive player's blocking assignment, the type of block executed, a pull, down block, or crack block among others, and exactly when each engagement begins and ends, according to NFL.com.
The model runs on a transformer architecture, the same broad approach behind large language models, applied here to interpret the spatial and temporal relationships among players throughout a play. A companion run-scheme classifier assigns one of 16 intended-design labels per play, sorts each into man, zone, or gap schemes, tags concepts like read option and split zone, and compares the intended run gap to the gap the ball carrier actually hits, according to NFL.com.
Separate models track each defender's individual gap responsibility. That frame-level detail lets Next Gen Stats measure how often a blocker gets double-teamed and how long it takes a defender to shed a block. Next Gen Stats said the goal is to statistically credit players in the trenches who have historically gone unmeasured, according to NFL.com.
Why assignment data can explain a stuffed run
Next Gen Stats frames the value of this data with a simple scenario: four blockers can execute their assignments correctly, but one lost matchup can still blow up the play. The reverse holds too, four good blocks don't guarantee a good outcome, and neither does one bad one: a skilled running back can escape immediate penetration and turn a busted blocking assignment into a positive gain, meaning the box score alone doesn't show who actually won or lost at the point of attack, according to NFL.com.
The model can also identify defenders who disrupt a run without ever making the tackle. A player who blows up a blocking scheme two gaps away gets no credit in a traditional stat line but shows up in the assignment data, according to NFL.com.
Consider a zone run stuffed for no gain. A guard could have lost his down block, a linebacker could have filled his assigned gap correctly, or the runner could have missed the designed lane entirely. Yards per carry alone would not distinguish those explanations; the NFL says its assignment and timing data are intended to provide that context.
PFF run blocking metrics explained: what the rates mean
PFF grades every player on every play on a scale from -2.0 to +2.0 in 0.5-point increments, with 0 representing expected performance for that specific play. The new PFF positively graded play rate and PFF negatively graded play rate convert those grades into percentages, how often a player's snap lands above that baseline versus below it, according to PFF. PFF said the positive rate measures impact, how often a player delivers a block above expectation that helps the play, while a low negative rate reflects stability and how consistently a player executes his assignment without a costly mistake.
PFF also reported that Lindstrom ranked third among offensive linemen in 2025 run-blocking grade, at 91.7. On zone runs specifically, Colts guard Quenton Nelson led all qualifying linemen at 29.4%, finishing 1.2 percentage points ahead of both Lindstrom and Chiefs center Creed Humphrey. Bears tackle Darnell Wright led gap-concept runs at 24.1%, 1.8 points ahead of the next-closest lineman, Quinn Meinerz, according to PFF.
The PFF run defender play rates flip the lens to the other side of the ball. Cowboys defensive tackle Quinnen Williams, who joined Dallas in a midseason trade, posted a 30.3% positive-play rate against the run, finishing 3.7 percentage points ahead of the next-closest interior defender. Seahawks edge rusher DeMarcus Lawrence led all edge defenders at 23.3%, Titans linebacker Cedric Gray topped his position at 18.6%, and Colts cornerback Kenny Moore II led corners at 7.0%, according to PFF.
On the negative side, PFF listed Pittsburgh defensive tackle Cameron Heyward at a 9.1% negatively graded play rate and Cleveland edge defender Myles Garrett at 2.5%, according to PFF.
Comparing the two systems
The NFL model and PFF's rates measure different things by design, not just different topics. Next Gen Stats infers who was assigned to block whom, what type of block was used, and when that engagement occurred, built to explain why a run succeeded or failed. PFF's rate reports how often a play built from its grading process landed above or below its expected-performance baseline, according to PFF and NFL.com. Neither linked announcement reports run-specific validation figures, so there's no published number yet showing how well either system predicts actual rushing outcomes.
A useful comparison point is ESPN's update to its separate pass-block and pass-rush win rates, a metric family covering the passing game rather than the run game. ESPN sports data scientist Brian Burke wrote that reliability correlations for the updated pass metrics rose from 0.53 to 0.59 year over year, and correlation with EPA per play rose from 0.24 to 0.34, according to ESPN. Those figures show what a published reliability check looks like for a blocking metric, even though they apply to pass protection rather than the run-blocking numbers covered here.
The NFL Analytics Textbook notes that block win-rate methodologies differ by provider in their thresholds and matchup definitions, and that one-on-one framing can hide help from teammates. That caution was written about win-rate systems specifically, but it applies just as well to comparing any two metrics built on different processes, including PFF's grades and the NFL's tracking-based model, since neither shares the same underlying methodology. The same source states that run-block win rate "does not equal rushing success," worth remembering here too: both the NFL's assignment data and PFF's grade-based rates are provider-defined descriptions of a snap, not proven substitutes for yards per carry or team rushing output.
What to remember next
Two providers put out run-blocking numbers within days of each other this month, and neither is a finished scouting report. Before leaning on either one, check who produced it, a human grading team or a tracking-based model, whether it's scheme-specific the way PFF's zone and gap splits are, and whether the provider has published validation data the way ESPN has for its updated pass-block figures.
As of this week, neither the NFL's run model nor PFF's run rates come with that kind of published check. That's reason enough to treat both as descriptive tools rather than final verdicts on who's actually winning at the line.