How FQI works

Full Press Quarterback Impact is Full Press Coverage’s proprietary, play-level quarterback evaluation metric. It asks how well a quarterback played after accounting for modeled throw difficulty, expected yards after catch, opponent strength, positive-play consistency and damaging negative plays.

It describes performance in the selected season and scope. It is not a scouting grade, a forecast, or a claim to isolate quarterback talent perfectly. Read it alongside QBV/play, OQBV/play, playing time and the underlying game context.

FQI is standardized within the qualified quarterback pool for the selected season and scope, then placed on a scale centered at 50 with a 15-point standard-deviation multiplier and capped at 0–100. A higher score means a stronger performance relative to that pool.

A 75 FQI is not a 75% completion rate, win probability or percentile. The displayed component grades are separate 1–99 percentile profiles; averaging those grades will not reproduce FQI. Small score gaps should not be treated as precise proof of a meaningful difference in ability.

FQI measures rate and quality. Full Press Quarterback Value (FQV) measures cumulative QB value above the league QB-play baseline. A quarterback can have a high FQI in limited action while another produces more total FQV over a larger workload. Use both when discussing a full season.

Quarterback Value per play is the average modeled value credited to the quarterback on eligible QB plays, in expected-point units. It is the underlying value measure before opponent adjustment.

• Completed passes: air EPA plus expected YAC EPA.
• Incompletions, interceptions and sacks: QB EPA.
• Scrambles: QB EPA, with EPA as a fallback.
• Designed quarterback runs: EPA.

Kneels, spikes and explicit no-plays are excluded. EPA means expected points added: the change in expected scoring value associated with a play. A positive QBV/play means positive attributed value on average; the separate FQV measure compares total value with a league baseline.

A short completion can become a long gain because of receiver skill, blocking and missed tackles. FQI replaces actual YAC EPA with expected YAC EPA on completions, reducing automatic quarterback credit for extraordinary production after the catch. The difference between actual and expected YAC EPA is useful supporting context.

This remains a statistical attribution choice. Ball placement and timing can help create YAC, and the model cannot perfectly divide credit on every throw.

Opponent-adjusted QBV/play starts with raw QBV/play and subtracts the estimated defensive effect. It is the Value input in v1.4. Raw QBV/play stays visible so readers can see the value before schedule context.

The 50% Value weight makes attributed play value the largest part of the score. Accuracy adds 30%; Consistency+ and Avoidance+ each add 10%. These are the current v1.4 weights, established through historical testing.

Completion percentage over expectation compares actual completions with modeled completion probability on eligible passes. It measures accuracy relative to the difficulty captured by the source model, rather than treating every attempt as equally easy.

For example, 67% completions against a 62% expectation is +5 percentage points of CPOE. FQI uses raw CPOE for its 30% Accuracy input. The displayed Accuracy grade is a percentile profile, not the raw percentage-point value. CPOE is still influenced by the model’s available features and cannot fully separate throwing, receiving and protection.

Consistency+ measures the rate of positive-QBV plays above the rate expected for a quarterback’s OQBV/play level. It asks whether value arrived through a steadier supply of positive plays than that overall value would normally imply.

The “+” means above expectation. It is not simply completion percentage, raw success rate or week-to-week stability. Comparing success with expected success helps avoid rewarding the Value component twice.

A drive killer is an eligible play with QBV at or below −1.00: at least one expected point of negative attributed value. It is a value threshold, not a claim that the possession literally ended.

Avoidance+ measures avoidance of those plays above the level expected for the quarterback’s OQBV/play. It can capture costly sacks, interceptions and other damaging plays without treating every sack or turnover as equally costly. Higher is better.

The inputs are standardized using z-scores, which express distance from the comparison pool’s mean in standard-deviation units:

The shared modifier cap limits how far Consistency+ and Avoidance+ can move the score before final standardization. The ±0.50 modifier cap is separate from the ±0.12 QBV/play opponent cap.

Creation is a descriptive rushing/creation profile, not a fifth weighted component. Eligible quarterback runs already contribute to QBV.

The model compares how quarterbacks performed against a defense with their expected performance from other games. It estimates each defense’s effect while leaving the game being evaluated out of that defensive estimate. Sparse quarterback baselines move toward league average; sparse defensive estimates use a 250-play prior to reduce overreaction.

The defensive effect is capped at ±0.12 QBV/play per defense-game. A defense that suppresses value has a negative effect, so subtracting it raises OQBV/play. An easier defensive environment lowers it. The cap is in QBV/play units, not FQI points or a percentage bonus.

Illustrative example: raw QBV/play of +0.20 minus a defensive effect of −0.03 gives +0.23 OQBV/play. The adjustment is a modest schedule correction, not full protection or receiver attribution.

No. FQI v1.4 does not use leverage weighting in the headline score. Eligible plays still contribute under the normal play-value rules. The historical research did not find enough overall validation benefit to adopt the tested leverage-weighted versions.

Game-state information is displayed separately. These splits help readers judge the context of a performance without an additional garbage-time penalty or clutch bonus inside FQI.

• Competitive QBV/play: raw QBV/play when pre-play win probability is greater than 2% and less than 98%.
• Low-Leverage QBV/play: raw QBV/play at 2% or below, or 98% or above.
• Low-Leverage Share: the share of eligible QB plays in those extreme states.

The threshold uses modeled win probability, not a fixed point margin. It is an operational definition of low leverage, not a judgment about effort. Read each split with its sample size; a handful of plays can produce an extreme rate. Missing context should not be interpreted as zero.

Regular season uses regular-season plays; postseason uses playoff plays; combined uses both. Choose the matching scope in Historical or in the supported FQI stats controls. A playoff game belongs to its NFL season, even if it is played in the following calendar year.

Each scope has its own qualified pool and standardization. Combined FQI is recalculated from the combined play sample; it is not the arithmetic average of regular-season and postseason FQI. A postseason score describes performance relative to that postseason pool, which is much smaller and more selective.

• Historical regular season: 300+ dropbacks.
• Current early-season regular season: 75+ dropbacks.
• Historical postseason: 20+ postseason dropbacks.
• Historical combined: 300+ combined dropbacks.

The active current-season snapshot supplies its qualification rule; check the label on the selected leaderboard, especially for postseason and combined views. Qualification is an inclusion threshold, not a guarantee of statistical reliability. Dropbacks determine qualification; eligible designed QB runs also contribute to play value.

A quarterback missing from a table may not have reached its threshold or may not have a complete source record. That is not the same as receiving a zero FQI.

Use them as an early description, with caution about small samples. One interception, sack or explosive play has a larger effect over 75 dropbacks than over a full season. Postseason qualification is especially permissive so meaningful playoff appearances can be shown, but a 20-dropback sample is highly volatile.

Compare the same season type, check workload and context, and avoid treating adjacent ranks as a definitive talent order.

The current historical FQI dataset covers 2006–2025, with 2026 handled through current-season snapshots. The exact regular-season validation sample contains 601 qualified QB seasons. Historical postseason and combined datasets are separate and use their own thresholds.

Season-relative standardization helps compare how far a quarterback stood above his contemporaries. It does not fully equalize rule changes, schemes, supporting casts or data quality across eras. A higher historical FQI means a stronger relative profile in that season’s pool, not a proven answer to who would play better under identical conditions.

Traditional passer rating combines completion percentage, yards per attempt, touchdown rate and interception rate. It does not directly include quarterback rushing or sacks. NFL formula reference.

FQI uses attributed play value, expected YAC, opponent context and accuracy over expectation, plus consistency and avoidance modifiers. Two quarterbacks with similar passing box scores can therefore have different FQI profiles.

EPA/play summarizes how plays change expected scoring value. FQI starts from a QB attribution layer, replaces actual YAC EPA with expected YAC EPA on completions, adjusts the Value input for opponents and combines four standardized components.

EPA + CPOE approaches also combine value and accuracy. FQI shares those foundations, but its specific attribution, weighting, residual modifiers and scale are different. Correlation with another metric is expected; disagreement is a reason to inspect the plays and assumptions, not proof that either metric must be wrong.

Expected points and EPA explained.

No. ESPN’s QBR uses expected-point value, quarterback credit attribution and opponent adjustments. ESPN’s QBR methodology.

FQI is an independent Full Press Coverage metric with its own 50/30/10/10 architecture and explicit QBV/OQBV framework. Similar score ranges do not make the scores interchangeable. FQI is not produced, endorsed or licensed as an ESPN rating.

FQI is a play-by-play and statistical-model evaluation, not a human film grade. Film review can add information about decisions, protection failures, missed opportunities and responsibility that is absent from the FQI input data. Use disagreement as a starting point for closer review.

FQI v1.4 does not fully separate responsibility among quarterback, protection, receiver and scheme. Sacks flow through QB EPA; the score does not yet assign a verified responsibility percentage to each sack or pressure. Receiver drops, tipped interceptions, route errors, separation, time to pressure and time to throw are not fully resolved by the current attribution model.

Expected YAC and opponent adjustment address specific sources of context. They do not make the quarterback independent of his teammates. Richer charting and pressure/sack attribution remain research work; experimental QBV 2.0 ideas are not part of the published v1.4 formula.

The v1.4 production pipeline uses SportsDataverse model-scored play-by-play, with football data and supporting research from the public nflverse ecosystem. The availability of raw play-by-play does not guarantee complete modeled fields or comparable charting in every season.

FQI coverage reflects the inputs actually built and validated for this metric. Do not assume that recent tracking or charting detail exists throughout the historical sample. A blank or unavailable split is not evidence of zero production.

Source corrections, updated play classification, player matching or revised modeled inputs can change the underlying values. Current-season scores also evolve as new games arrive: the qualified pool, season baselines and opponent estimates can change even when a quarterback’s earlier plays do not.

A data refresh is different from a methodology revision. The v1.4 weights stay 50/30/10/10; any later change to the scoring method should be versioned and explained. Check the selected season, scope and snapshot information before comparing two saved results.

The v1.4 definition was tested across 601 qualified regular-season QB seasons from 2006–2025, including comparisons with other metrics, year-to-year stability, next-season EPA association and same-quarterback season ordering. The Validation tab on the rankings page summarizes the results.

These checks assess how the metric behaves; they do not prove causal attribution, guarantee future performance or establish every rank as correct. FQI remains a descriptive model with known limitations.