Expected Points Added, averaged over every dropback (pass attempt or sack) that has a value on file. Positive means the play typically moved the team closer to scoring more than an average play in that situation would; negative means it cost expected points.

Inputs: Each play's EPA (from CFBD's own model for most FBS/FCS plays, or this project's own trained model for a smaller share of other games) averaged over the QB's dropbacks. The dropback count with an EPA value on file (not every play has one) is always shown alongside the rate, and boards require at least 20 such plays before a QB qualifies -- a thin sample was found live to distort rankings otherwise.

The same EPA/Dropback figure, but plays in a blown-out, non-competitive score/clock situation count for less (35% of a normal play) instead of being averaged in at full weight.

Inputs: Same EPA values as above, reweighted using this project's own garbage-time definition (a 7-tier rule based on score margin, quarter and clock).

The share of dropbacks that were 'successful' by a standard down-based yardstick, regardless of whether EPA is available for that play.

Inputs: Gained at least 50% of the yards needed on 1st down, 70% on 2nd, or the full distance (or a touchdown) on 3rd/4th -- computed from down, distance, yards gained and touchdown, all structured fields.

Passing Success Rate, with garbage-time plays down-weighted the same way as GT-Adjusted EPA/Dropback above.

Inputs: Same down-based success rule, reweighted by the garbage-time definition.

How often the QB's pass attempts on 3rd or 4th down actually picked up the first down (or scored).

Inputs: Pass attempts where down is 3 or 4, divided into how many were marked as a first down or touchdown by the play-by-play itself.

How a player performed specifically on 3rd down plays this season -- success rate, touchdown rate, and the underlying counting stats, isolated from every other down.

Inputs: Every play on file where down = 3 and this player was involved (passer, rusher, receiver, or a tackle/sack/INT/fumble recovery on defense).

The same idea as 3rd Down Splits, isolated to 4th down plays -- the highest-leverage down in football.

Inputs: Every play on file where down = 4 and this player was involved.

How a player performed once his team's offense (or the opponent's, for a defender) reached inside the 20-yard line.

Inputs: Every play on file within 20 yards of the goal line (`distance_to_goal <= 20`) this player was involved in. COVERAGE CAVEAT: true field position is only recorded for about a third of plays leaguewide, concentrated in FBS/FCS games -- a player whose games come from a source without it will show thin or missing red-zone splits even if he played plenty of red-zone snaps. This is a gap in the underlying play-by-play data, not a sign he didn't play there.

How a player performed in the last two minutes of either half -- the clock-management, high-pressure stretch at the end of the 2nd and 4th quarters.

Inputs: Every play on file in the 2nd or 4th quarter with 2:00 or less on the clock this player was involved in. Same field-position/clock coverage caveat as Red Zone Splits above -- roughly a third of plays leaguewide carry a usable clock reading, concentrated in FBS/FCS games.

A 7-tier rule (bigger lead required as the clock runs down) that flags plays happening in a game that is no longer competitive -- e.g. a 30+ point lead right after halftime, or a 12+ point lead inside the final 3 minutes.

Inputs: Quarter, game clock, and the score of both teams at the time of the play. Garbage-time plays are not thrown out -- they count for 35% of a normal play in every "garbage-time-adjusted" figure, so a player who pads stats once a game is decided shows up as such rather than being invisible.