The value a single play adds relative to the expected scoring outcome of a given situation. A play that moves the offense from a 2.0 expected-point situation to a 3.5 expected-point situation has an EPA of +1.5. Positive EPA means the play helped the offense; negative means it helped the defense.

The average number of points a team is expected to score from a given down, distance, and field position, based on historical play-by-play outcomes. For example, a 1st-and-10 at the opponent's 20-yard line might have an EP of ~4.2 because teams in that situation historically score touchdowns more often than not.

A team's average EPA across all plays — the single best per-play efficiency measure. Higher is better for an offense. For a defense the sign depends on the surface (see Defensive EPA): team-rating and matchup surfaces report defensive performance (higher is better), while the Situational page reports EPA allowed (lower is better).

EPA/Play split by play type. EPA/Pass averages EPA over dropbacks (sacks and scrambles included); EPA/Rush over designed runs. Passing EPA is more stable and more predictive than rushing EPA, which is why our ratings lean on it. Offense: higher is better. Defense on team-rating surfaces: higher is better (defensive-performance convention — see Defensive EPA).

The percentage of plays with a positive EPA. A "successful" play is any play that increases the offense's expected points. Roughly analogous to "moving the chains" — consistently positive plays win games.

EPA/Play on early downs — 1st down, or 2nd-and-6-or-fewer. Staying efficient on early downs avoids obvious passing downs and is one of the most stable signals of team quality. Offense higher = better; defense higher = better on team-rating surfaces, lower = better on the Situational page (EPA allowed).

EPA computed within a specific game context rather than across all plays. Contexts: Red Zone (inside the 20), Third Down, Early Downs (1st, or 2nd-and-6-or-fewer), Passing Downs (2nd-and-7+, 3rd-and-5+, 4th-and-5+), First/Second Half, and No Garbage Time. On the Situational page, offense higher = better and defense lower (EPA allowed) = better; teams with fewer than 10 plays in a situation are excluded.

Low-leverage snaps in a decided game, where play-calling no longer reflects normal intent. We define it as 21+ point leads in the 4th quarter and 28+ in the 3rd. The "No Garbage Time" splits filter these out to strip stat-padding and reveal true efficiency.

Our core team-strength building block. Each team's offense and defense is rated by expected points added per drive, adjusted for opponent quality. Both off_drive_rating and def_drive_rating follow a HIGHER = better convention — an elite defense posts a large POSITIVE def rating (Ohio State ~+1.47 in 2025; worst FBS ~−1.54). These are the per-drive components that feed the Final Rating.

A single number ranking every FBS team's overall strength, expressed in points per game against an average opponent. It is the per-game expected-points rating (the opponent-adjusted offensive and defensive Drive Ratings scaled by how many drives the team's games actually contain) plus an opponent-adjusted special-teams component. Higher is better; roughly 0 is average FBS.

A team's offensive Drive Rating plus its defensive Drive Rating — one opponent-adjusted number for total team strength, on the same points-per-drive scale as its components. Both components follow the higher = better convention, so a large positive Net Rating means a team that wins on both sides of the ball. Closely related to the Final Rating, which is built from the same per-drive components.

How a team changes the length of a game, measured in possessions. Pace is an opponent-adjusted estimate of how many drives a team adds to or removes from a game relative to average — OFF Pace for when its offense has the ball, DEF Pace for when its defense is on the field. Positive = faster (more possessions than average); negative = slower. Pace describes style, not quality: a fast bad team and a slow good team are both possible, which is why Pace is shown alongside the ratings rather than inside them.

EPA adjusted for strength of schedule, so production is credited relative to the quality of the opponent faced. Our per-drive ridge model solves every team's offensive and defensive value simultaneously, which folds opponent quality directly into the ratings rather than treating it as a separate after-the-fact adjustment.

How tough a team's opponents have been. Rather than a standalone SOS number, opponent strength is baked directly into our ratings: the per-drive ridge model solves every team's offensive and defensive value at once, so credit scales with the quality of the defense or offense faced. See Opponent-Adjusted EPA.

Where a position group's recruiting talent sits nationally. We take each team's top signees by composite recruiting rating across the last four classes — top five per position group, top three for QB and TE, where rotations are shorter — average them, and rank that average against every FBS team's same position group in the same season. Shown as a percentile: 90 means the group out-recruited 90% of FBS. Blue-chips are 4- and 5-star signees. It is a roster-talent proxy — it doesn't see transfers, development, or attrition.

On game and slate matchup grids, the edge bar marks which side holds the statistical advantage on each row: a unit is compared through its national rank — the offense's rank on a stat against the opposing defense's rank on the allowed version — and the side with the larger league-relative advantage gets the edge, with near-ties shown as even. The summary pill counts how many rows each side leads. This is an analytical comparison of season-to-date profiles. It is not a betting edge, not a prediction, and not a recommendation — a team can lead most rows of a matchup grid and lose the game.

The gap between a team's rank by straight-up record and its rank in our per-drive, opponent-adjusted team-strength model, shown on the Reality Check tool. A large gap means the win-loss record and the underlying play disagree. The scatter plots both ranks against a perfect-agreement diagonal; each team's breakdown explains why the gap exists through strength of schedule, margin-of-victory pattern (Pythagorean expectation and close-game luck), turnover variance, and garbage-time scoring — plus a quality-peer benchmark (the average wins of all team-seasons with a model rank within ±10, 2017–2025) and the team's own ATS / over-under record. It is descriptive, not predictive: we tested whether divergence predicts against-the-spread outcomes and the no-lookahead result was null. A research lens for understanding what already happened, not a forecast or a betting signal.

The win total a team's points scored and allowed "should have" produced: expected wins = games played × PF^2.37 / (PF^2.37 + PA^2.37) — the classic points-based estimator with a commonly used exponent for college football. A team that won nine games on a Pythagorean expectation of 6.5 banked the close ones; a team that won six on an expectation of eight lost them. Reality Check uses the gap as one explanation for record-vs-model divergence — a description of the season that happened, not a projection of the next one.

The Archive's match score. Every team-season since 2017 is ranked within its own season on 44 statistical features — EPA splits, identity, situational performance, and more — and similarity between two profiles is a weighted distance on those ranks. Using ranks instead of raw values is what lets a 2018 profile and a 2025 profile compare fairly across eras. The displayed score converts that distance to an index where higher means more similar; treat it as relative — for ordering candidates within one search, not comparing across searches. Resemblance in past profiles describes how alike two teams were; it does not predict a game's outcome.

How well a defense prevents expected points. IMPORTANT — Fourth Down Data reports this two ways with OPPOSITE signs. On team-rating and matchup surfaces (EPA Rankings, the Game-on-Paper grids) it is a defensive-PERFORMANCE value, so HIGHER is better — an elite defense posts a high positive number (2025 median ~+0.17; Texas Tech +0.30, Ohio State +0.27). On the Situational EPA page it is raw EPA ALLOWED, so LOWER (more negative) is better. Same idea, flipped sign — always check which surface you're reading.

The defensive counterpart to success rate. On team-rating surfaces it is the share of plays the defense WINS (forces a negative-EPA outcome), so HIGHER is better (2025 median ~0.60; elite Ds ~0.64–0.67). On the Situational page it is the success rate ALLOWED, so LOWER is better.

Defensive disruption. Havoc plays = sacks + tackles for loss + interceptions + forced fumbles. Havoc Rate is the percentage of opponent plays disrupted this way; Havoc/Game is the per-game count. Higher is better — disruptive defenses manufacture negative-EPA plays and turnovers.

Two components of havoc, reported per game. A sack tackles the QB behind the line on a pass; a tackle for loss (TFL) is any tackle behind the line of scrimmage. Both inflict large negative EPA on the offense. Higher is better for a defense.

Average drive start, in yards from the opponent's goal line. For an offense, LOWER is better (drives begin closer to scoring). For a defense, HIGHER is better — opponents are pinned deep (2025: elite units force starts ~74 yards out vs ~70.6 median).

The share of available yards (line of scrimmage to the goal line) an offense actually gains. A drive starting at its own 25 has 75 available yards; gaining 30 is 40%. Offense: higher is better. Defense: the share ALLOWED, so LOWER is better (elite Ds ~0.27–0.31 vs ~0.45 median).

Average points scored per offensive possession. More tempo-aware than points per game — 28 points on 9 drives is more efficient than 28 on 13. Higher is better for an offense; lower allowed is better for a defense.

Success rate on third-down plays — the share of third downs that yield a positive-EPA result (usually a conversion). For an offense, higher is better. For a defense it is the conversion rate ALLOWED, so LOWER is better.

Average yards-to-go a team faces on third down (2025 FBS median ~7.0). For an offense, lower is better — short third downs mean efficient early downs. For a defense, HIGHER is better: forcing opponents into 3rd-and-long.

Share of red-zone trips (offense inside the opponent's 20) that end in a touchdown rather than a field goal or empty trip. Measures finishing — settling for field goals leaves points on the board. Higher is better.

The number of yards between the line of scrimmage and the opponent's end zone. A key input to the Expected Points model — a 1st-and-10 at the 20 is worth far more than a 1st-and-10 at the 80.

A play that swings the scoreboard, not just the chains: we define explosive by value, not yardage — any pass with an EPA above 1.0, or any rush above 0.8. A value threshold beats a fixed yardage cutoff because identical yardage isn't identical impact: a 25-yard completion on 3rd-and-30 barely matters, while a 15-yard touchdown run changes the game. Explosive Rate is the share of a team's plays that clear the bar; Explosives/Game and Explosives Allowed/Game are per-game counts. Offense: more is better. Defense: Explosives Allowed lower is better (elite Ds ~2–3/game vs ~4.2 median).

Takeaways minus giveaways (interceptions and fumbles gained minus lost). Turnovers are the most impactful play type — roughly −4 to −5 EPA each — so margin correlates strongly with winning, but it is also among the noisiest, least repeatable metrics year to year. Higher is better.

Total game time a team holds the ball. Widely cited but largely OVERRATED as a predictor — it's mostly a SYMPTOM of other things (run-heavy scheme, leading and milking the clock, converting third downs), not a cause of winning. A defense isn't "tired" from being on the field longer; the efficient team tends to control the ball because it's good, not the reverse. We surface it for context, not as a quality signal.

Total yards gained divided by total plays. A simple efficiency metric that measures how many yards a team gains on an average snap. Less sophisticated than EPA but useful as a quick reference.

The current down (1st through 4th) and yards needed for a first down. The primary situational context for every play. 3rd-and-long is a very different situation than 1st-and-10, and the EP model captures this.

A change of possession via interception or fumble. Turnovers are the most impactful plays in football, typically worth -4 to -5 EPA because they hand the opponent the ball in scoring position.

Any play that directly results in points: touchdowns (7 EP), field goals (3 EP), safeties (-2 EP for the offense). These plays set the EP_after value directly rather than requiring a model lookup.

The top level of NCAA Division I college football, comprising roughly 133 teams across conferences like the SEC, Big Ten, Big 12, and ACC. Our platform focuses on FBS teams and filters out FCS (lower-division) matchups.

Whether a team covered the point spread set by oddsmakers. If a team is -7 (favored by 7) and wins by 10, they covered ATS. If they win by 3, they did not cover. ATS records measure betting performance, not just wins and losses.

A team's result on the scoreboard — did it win the game — with the point spread ignored entirely. SU records are ordinary win-loss records. The contrast matters on betting surfaces: a 10-2 team that was favored by wide margins all year can easily be 4-8 against the spread while 10-2 straight up.

How far a team beat or missed the spread, in points: the final scoring margin plus the team's spread. A 7-point favorite (−7) that wins by 10 has an ATS Margin of +3 — it covered by 3. Negative means the team fell short of the number; zero is a push. The column turns the win/loss ATS result into a magnitude.

The expected margin of victory set by oddsmakers. A spread of -6.5 means the favored team is expected to win by about 7 points.

The total combined points oddsmakers expect both teams to score. Bettors wager on whether the actual total will be over or under this number. Also called the "total."

When a team beats the point spread. If a team is +3.5 (an underdog) and loses by only 2, they "covered" the spread. Covering is the fundamental metric of spread betting success.

A straight-up bet on which team will win, with no point spread involved. Odds are expressed as positive (underdog, e.g. +150 means $100 bet returns $150 profit) or negative (favorite, e.g. -200 means you must bet $200 to win $100).

A spread of exactly 0 — neither team is favored. The game is considered a toss-up by oddsmakers.