66 min read

Week 12, home against St. Louis, third quarter. We're up 13–10. Moose Odom's third-down run on 3rd-and-2

Chapter 28 — Analytics and Offensive Decision-Making: When the Numbers Should Override Your Gut — and When They Shouldn't

The Hook: Fourth-and-1 at Our 43

Week 12, home against St. Louis, third quarter. We're up 13–10. Moose Odom's third-down run on 3rd-and-2 at our 42 hit a wall of Stallions and fell forward, and the officials bring the chains out. The spot is a long foot short. Fourth-and-1 at our own 43.

The punt team is already standing up on the sideline. That's what punt teams do on fourth down at their own 43 — they've been doing it since they were nine years old.

Teo is in my ear from the booth, and on a fourth down the quiet window is still the quiet window: shell, box, one thing.

Teo: Bear. Eight. Mabry at three.

Bear front: the center and both guards covered by down linemen, so there's nobody to double-team and nobody standing in the A gaps — the gaps on either side of the center. Eight defenders in the box. Hollis Mabry, their Mike linebacker, stacked three yards behind the nose. Frank Ostrowski is telling us he knows exactly what fourth-and-1 looks like, and he has bought enough bodies to stop it.

On the other channel — the one Wade Harmon shares with Marisol Vega, our director of analytics — a decision is being made, and I can't hear it. Then Wade is on mine.

Wade: Marisol says go. She says it's not close. What's the call if we go?

That question is the most important sentence in this chapter, and you'll hear it again.

Lyle: Heavy, Check. If they bear it, it's the sneak. If they don't, duo. Darius knows.

There's a pause long enough for me to hear the crowd. Wade was a defensive coordinator for most of his career. Every defensive instinct he owns is telling him to punt, pin them inside the 20, and make Haugen and their wide-zone offense drive eighty yards against a defense that had been playing well.

Wade: Every bone I've got says punt it. Go.

We sent in Heavy — our word for 13 personnel, one back and three tight ends: Harlan, Brandt, and Sione Fale as the third tight end, plus Ruffin and Sutton. The Stallions stayed in their bear. Darius walked to the line, saw three down linemen covering Szabo and both guards, and said the code word for the sneak. Szabo drove the nose. Kendrick Moss and Pete Lorusso slammed their inside shoulders into the A gaps so nobody could knife through. Darius got under Szabo's right hip and pushed — Rome Pettaway's whole coaching point for the sneak is four words, ball over the line — and Mabry met him a yard and a half downfield, a half-count late. First down at our 44.

On the very next snap we lined up in an unbalanced line, faked outside zone, and booted Vance the other way. Brandt slipped into the flat for 14. Ostrowski had just watched us gain a yard against eight men, and his front bumped toward our overload to stop the run; the side he slid away from is where Brandt caught the ball.

Here is what I want you to notice. The chart didn't call the sneak. It doesn't know what a bear front is. It has never met Aaron Szabo. What it did was tell Wade, in one number, that the punt he wanted — the safe, normal, nobody-gets-fired punt — was actually the bigger risk. And then it handed the decision back to the only people who could answer the question it can't: what's the call?

Three weeks later, in Sacramento, on fourth-and-1 at our own 36, Wade asked me the same question. I didn't have a good answer that day, and he punted. Both decisions were right. This chapter is about why.

🏃 Fast Track: If you already know expected points and success rate, skim 28.2–28.3 and start at 28.4 (the run/pass selection trap), then read 28.6–28.7 closely — the fourth-down break-even number and the four questions that decide when a model should lose. The Project Checkpoint builds your dashboard and your fourth-down/two-point chart.

🔬 Deep Dive: Appendix H has the full expected-points ladder, success-rate variants, and fourth-down and two-point logic. Pair this chapter with Chapter 25 (a call is a bet on a distribution) and Chapter 26 (the situations these numbers live in).

📺 Film Watcher: This Sunday, on every fourth-and-short between the 40s, write down the call before the coach decides. Then ask the question the broadcast never asks: did the offense have a play it trusted against the front it saw?


28.1 What the Numbers Are For

Marisol Vega has a sentence she says to every new coach on our staff, usually in the first week of training camp, usually while they are telling her about a gut feeling: "I'm not here to make the decision. I'm here to make sure you know what you're deciding."

That's the whole job, and it's harder than it sounds. A football staff makes thousands of decisions a season — which plays go on the call sheet, which tendencies to break, when to go for it, when to go for two, which player to feature, which defender to attack. Every one of those decisions is a bet on a future nobody can see. Analytics, in our building, means one thing: replacing memory and feel with counted evidence wherever counted evidence exists, and being honest about where it doesn't.

Memory is a terrible analyst. You remember the 52-yard touchdown and forget the four times the same call was a checkdown. You remember the fourth down that failed on national television and forget the nine that converted on a Tuesday-night film session nobody watched. Coaches are not dumber than other people; they are people under pressure, and pressure makes recent, vivid, and painful outcomes loom larger than they deserve. Counting fixes that. It doesn't fix everything.

Here's what Marisol's department actually produces in a week. Notice that almost none of it is "advanced." Most of it is counting done honestly.

Day What Marisol produces Who uses it The question it answers
Sunday night Game dashboard: success rate, EPA per play, explosives, third down, red zone, sacks and their causes Lyle, Wade Did we play the way we think we played?
Monday Self-scout: what did we show over our last four games? ("kill the zeros," tells over ~70% on 10+ snaps) Lyle, Teo, Gus What does the next defense know about us?
Tuesday The slide: the opponent's five best defenders paired with the Outlaw who must block or beat each Game-plan meeting Where is our guy better than theirs?
Tuesday–Wednesday Call-sheet audit: efficiency of each box by situation, with sample sizes Lyle Which calls have earned their spot?
Thursday Fourth-down and two-point chart for this opponent and this kicker, printed for Wade Wade, Lyle Before game day: where are we a four-down team?
Game day Live chart on Wade's channel; the dashboard at halftime Wade, then Lyle Go, kick, or punt? Two or one?

Two of those rows came straight out of earlier chapters. "Kill the zeros" is Marisol's rule that every personnel group we use regularly must show every major play family at least a few times in any four-game window — a grouping with zero passes is a run tell, and a defense will find it. Her tell threshold — a look that predicts one play family about 70% of the time or more, on ten or more snaps — is the line between "a lean" and "a tell." Both are analytics. Neither involves a model. Both changed our offense more than any model did.

💡 Coach's point: Analytics is not a department that tells coaches what to do. It is a discipline that stops coaches from lying to themselves. If you only remember one sentence from this chapter, make it Marisol's: know what you're deciding.

Marisol's honesty rules

Marisol wins arguments in our building because she loses some on purpose. She will tell you when her number is too thin to use, and that is exactly why you believe her when she says it isn't. Her rules, which are now our rules:

  1. Counts before percentages. Write "9 of 14," never "64 percent." The count tells you how much to trust the rate.
  2. Every number carries its sample and its caveat. A number without a sample size is a rumor.
  3. Say what the number can't see. Every slide has a bottom line that begins "This doesn't know…"
  4. Separate the decision from the result. A good decision can lose. A bad decision can win. Grade the process, then look at the result to learn something about the inputs.
  5. When you don't know, say "I don't know." If more than about one snap in ten in a column is tagged "?", she won't give it a percentage at all.

The fifth rule saved us more than once. The fourth is the hardest one to live by on a Monday after a loss.

🔄 Check your understanding: A new assistant hands you a slide that reads "Opponent blitzes 71% on 3rd-and-long." Using Marisol's rules, name two things missing from that slide before you'd let it change your call sheet.

Answer

The count (71% of what — 5 of 7, or 50 of 70?) and the caveat (what the number can't see: which quarterbacks, which formations, which weeks, whether the defense has changed since). A third good answer: the definition — what counts as a "blitz" (five rushers? any linebacker rushing? a simulated pressure where four rush but one is a linebacker?). Definitions change the number.


28.2 Expected Points: The Value of a Down, a Distance, and a Spot

Every snap starts from a situation: down, distance, and field position (and, late in games, score and time). Some situations are worth more than others. First-and-10 at the opponent's 20 is obviously better than first-and-10 at your own 20. Expected points (EP) puts a number on "how much better."

Here's the idea in plain English. Take every play in many seasons of NFL football that started from, say, first-and-10 at the opponent's 20. Follow each one until somebody scores next — your touchdown, your field goal, their touchdown after your interception, a safety. Average those next scores, counting the opponent's points as negative. That average is the expected points of first-and-10 at the opponent's 20. Public models built from large play-by-play datasets — the open-source nflverse community publishes one; commercial services have their own — do this with more care, adjusting for down, distance, time, and more, but that is the core of it.

EXPECTED POINTS FOR 1st-AND-10, BY FIELD POSITION
(illustrative round numbers; every public model differs a little)
each # = 0.2 points

 1st-and-10 at   EP
 own  5         -0.3   <|
 own 20         +0.4    |##
 own 30         +1.0    |#####
 midfield       +2.0    |##########
 opp 35         +2.8    |##############
 opp 20         +3.8    |###################
 opp 10         +4.6    |#######################
 opp  2         +5.6    |############################

📊 Diagram (described): A horizontal bar chart of expected points for a first-and-10 at eight spots on the field. Pinned at your own 5, the value is slightly negative — the opponent is a little more likely to score next than you are. It climbs to about +0.4 at your own 20, +1.0 at your own 30, +2.0 at midfield, +2.8 at the opponent's 35, +3.8 at their 20, +4.6 at their 10, and about +5.6 at their 2. The bars grow steadily; the field is worth roughly a point for every fifteen to twenty yards through the middle and a bit more near the goal line.

Three things to see in that ladder. First, field position is worth points before anybody scores — moving from your 30 to midfield is worth about a point, the same as a made extra point. Second, the value is not linear at the ends: the last ten yards are expensive, because the field shrinks and the defense gets more help (that's the whole subject of red-zone football). Third, the numbers are approximate and model-dependent. I'm giving you round numbers so the arithmetic is easy. Marisol uses a full model; you should know the model exists and know its shape, not memorize decimals.

Down and distance matter as much as field position. Third-and-9 at your 38 is worth far less than first-and-10 at your 38, because a big share of the time it ends in a punt. That's why a 6-yard gain on third-and-5 and a 6-yard gain on third-and-10 are completely different plays.

EPA: what one snap was worth

Expected points added (EPA) is the change in expected points from before a play to after it. If the snap moves you from a situation worth +1.0 to one worth +2.2, it added +1.2. If it moves you from +0.6 to a turnover that gives the opponent a situation worth +1.8 for them (−1.8 for you), it cost you −2.4.

Let's run real snaps from our season through it, on the same illustrative scale. Marisol's model would give slightly different decimals; the shape is what matters.

Snap (fictional Outlaws season) Before After EPA (approx.) What it teaches
Game 1, Q4: "Inside PA, Gaines Post" — 1st-and-10 at the SAC 39, Gaines +35 ~+2.6 1st-and-goal at the SAC 4, ~+5.3 ~+2.7 One explosive is worth more than most whole drives
Game 1, Q4: inside zone, Ruffin +5 on 1st-and-10 at the AUS 49 ~+2.0 2nd-and-5 at the SAC 46, ~+2.1 ~+0.1 A "boring" 5 on first down barely moves the needle — but it keeps you on schedule
Game 2, Q3: 3rd-and-7 at the AUS 38, the 9-yard sack ~+0.7 4th-and-16 at the AUS 29 — a punt, ~−0.9 ~−1.6 Third-down sacks are drive-killers twice: lost yards and a lost down
Game 3, Q3: 3rd-and-7 at the AUS 41, Gaines hitch +10 vs. the mug ~+0.8 1st-and-10 at the SAC 49, ~+2.0 ~+1.2 A ten-yard hitch on third down is worth more than a twenty-yard gain on first
Game 3, Q4: 3rd-and-9 at the AUS 38, the interception ~+0.6 Sentinels' 1st-and-10 at the SAC 46, ~−1.8 for us ~−2.4 The cost of a turnover includes the possession you lose and the one you hand over

Read that table twice. The five-yard inside-zone run that started our tying drive in Game 1 was worth almost nothing by itself. The play-action post two snaps later — a pass that begins as a fake of that same run, from the same picture — was worth more than two and a half points. A pure EPA-per-play view of football would tell you the run was a waste of a snap. Hold that thought; it's the center of the biggest argument I've ever had with Marisol, and you'll see it in 28.4.

📈 By the numbers: The pattern in that table holds across the league in public data: third-down results and turnovers swing expected points far more than any individual early-down gain, and the biggest single-play swings belong to explosives and interceptions. That is why third-down conversion and turnover avoidance get their own lines on every dashboard, even though EPA already "contains" them.

⚠️ Common mistake: treating EPA as a grade for the player. In Game 3 the interception was −2.4 on the sheet, and it went on Vance's line. But Vance read two-high pre-snap and the Sentinels rotated to single-high at the snap — the same disguise they'd been waiting to use on third-and-long all season. EPA measures what happened to the scoreboard, not who was wrong. The fix: attach EPA to the call and the situation first, then assign responsibility with film, not with the number.

🔄 Check your understanding: On the illustrative ladder, roughly how many expected points is a 15-yard gain on first down from your own 30 worth? Why is a 15-yard completion on 3rd-and-14 from the same spot worth more?

Answer

From your 30 (about +1.0) to your 45 (between +1.0 and +2.0, about +1.7): roughly +0.7. The 15-yard completion on 3rd-and-14 is worth more because the before value of 3rd-and-14 is much lower (most of those snaps end in punts), and the after value is a fresh first down. EPA measures the change, so the worse the starting situation, the more a conversion is worth.


28.3 Success Rate, Staying on Schedule, and Explosives

EPA is powerful but noisy: one 60-yard touchdown can make a bad day look good. Success rate asks a simpler question about every snap: did this play keep us on schedule? The common definition, and the one we use:

Down A play is "successful" if it gains… Example
1st at least 40% of the yards to go 4+ yards on 1st-and-10
2nd at least 60% of the yards to go 3+ on 2nd-and-5; 6+ on 2nd-and-10
3rd and 4th 100% — a first down or touchdown 7 on 3rd-and-7

Some analysts instead define a successful play as any play with positive EPA. The two definitions mostly agree; when you read a number somewhere, check which one was used.

Grade the Game 1 tying sequence with it. Inside zone for 5 on first-and-10: success (5 is more than 4). Inside zone for 7 on second-and-5: success. Play-action post for 35: success. Inside zone for the 4-yard touchdown: success. Four snaps, four successes, and only one of them was an explosive. That is what a good drive looks like on a success-rate sheet — not a parade of big plays, but a series of snaps that never let the defense choose the down and distance.

Why schedule matters: the third-down cliff

Here is the reason success rate matters so much on early downs. It's the conversion table from the four-game sample of our own offense that the Sentinels charted before Week 6, and that I quote to myself every time I'm tempted to "take a shot" on second-and-5:

THIRD-DOWN CONVERSIONS BY DISTANCE -- Outlaws, four-game sample
(55 third downs; each # = one conversion, each . = one failure)

 3rd-and-1 to 3    10 of 14   ##########....
 3rd-and-4 to 6    13 of 26   #############.............
 3rd-and-7 to 9     2 of 9    ##.......
 3rd-and-10+        1 of 6    #.....

📊 Diagram (described): A dot chart of 55 third downs from our first four games, split by distance. On third-and-1 to 3 we converted 10 of 14 — mostly hash marks with a few dots. On third-and-4 to 6, 13 of 26 — half and half. On third-and-7 to 9, only 2 of 9, and on third-and-10 or longer, 1 of 6 — rows that are almost all dots. The chart shows a cliff: past six yards to go, our conversion rate collapses.

That's a small sample (Marisol would remind you it's 55 snaps), but the season told the same story: we converted about two-thirds of our third-and-shorts, about a third of third-and-7-to-10, and fewer than one in five third-and-11-or-longer. Every NFL offense lives on some version of this cliff.

🔍 Why this works: Distance changes what the defense is allowed to do. On third-and-2, the defense must defend the run, the quick game, and the sneak, so it can't play deep and can't rush recklessly. On third-and-9, it can drop eight, rush four, sit at the sticks, and dare you to complete something in front of it — or it can bring a disguised pressure because it knows you must hold the ball long enough for a 9-yard route. Every yard of distance is a yard of freedom handed to the defensive coordinator. Success rate on early downs is really a measure of how much freedom you gave Russ Kimbrough on third down.

That's why "staying on schedule" isn't a cliché in our building. It's arithmetic. A first-down play that gains four is a success not because four yards is impressive but because second-and-6 keeps the whole call sheet open.

Explosives: the other engine

Success rate can lie in the other direction. An offense can grind out a high success rate with four-yard runs and five-yard completions and still score 13 points, because it needs twelve snaps to go seventy yards and something always goes wrong in twelve snaps — a holding call, a tipped ball, a third-and-long. Explosive plays — in this book, a run of 12+ yards or a pass of 20+ — compress the field.

We had 71 explosives in 17 games, a little over four a game. Marisol's season split is the one I show young coaches: of our drives that included at least one explosive, a little under half ended in a touchdown; of drives without one, about one in eight did. The explosive isn't a bonus; for most offenses it's the difference between a field goal and a touchdown.

So the dashboard needs both engines. Success rate measures whether you can stay on schedule. Explosive rate measures whether you can skip the schedule. A good offense has both; a good defensive coordinator will try to take away one and dare you to live on the other. Portland's Cover 2 made us live on success rate — they let us complete everything short and tackled it. Sacramento's two-high, rotating shell was designed to take away our explosives while keeping enough people in the box that our early-down success rate dropped too.

⚠️ Common mistake: the explosive addiction. Curtis Bale, who worked with me on Walt Dupree's staff, led his conference in explosive plays in his first year as a coordinator. The next season defenses played deep, the explosives dried up, and instead of living on schedule he kept calling for the big one — abandoning the run when it got tight, asking a pocket quarterback to hold the ball against defenses built to wait. The sacks piled up and he was fired. Explosives are the most valuable plays in football and the ones defenses adjust to fastest. The fix: build your offense to survive on success rate, and let the explosives come from the plays the success rate sets up.

🔄 Check your understanding: Grade these three first-down plays by success rate (40% rule): Ruffin +3 on 1st-and-10; Powers +9 on 1st-and-15 after a penalty; a 2-yard completion on 1st-and-goal from the 4.

Answer

Ruffin +3 on 1st-and-10: unsuccessful (needed 4). Powers +9 on 1st-and-15: successful (needed 6). The 2-yard completion on 1st-and-goal from the 4: successful (needed 40% of 4, which is 1.6 yards). Success rate scales with the distance — which is exactly why it beats raw yards per play as a measure of staying on schedule.


28.4 Run, Pass, and the Selection Trap

No argument in modern football has produced more heat than run versus pass. Here's the honest version.

What the public data says

Over many seasons of public play-by-play data, most analyses have found that on early downs, passes produce more expected points per play than runs, on average, and that most NFL teams have historically run more often on early downs than those averages alone would justify. Analysts track this with early-down pass rate — the share of first- and second-down snaps (usually outside garbage time and two-minute situations) that are passes — and with pass rate over expected (PROE): how often a team passes compared with how often a model predicts an average team would pass in the same down, distance, field position, score, and time. A team with a positive PROE throws more than the situation predicts; a negative PROE means it runs more. In recent seasons the league as a whole has drifted toward throwing more on early downs, and toward running more efficiently when it does run — lighter boxes, more motion, more runs called against two-high shells.

Our offense was a little above zero in PROE on early downs for the season by Marisol's estimate — a few percentage points more pass-heavy than an average team in our situations. In the four-game sample the Sentinels charted, our 187 first- and second-down snaps broke down as called runs 37%, RPOs (run-pass options: a run with a pass attached that the quarterback chooses after the snap) 20%, play-action 21%, straight dropbacks and quick game 17%, and screens 5%.

That last sentence already shows the first problem with the debate. Is an RPO a run or a pass? If you classify it by what happened — handoff equals run, throw equals pass — then every RPO where the quarterback correctly threw because the conflict defender triggered becomes a "pass" with a great result, and every RPO where he correctly handed off into a box that had just lost a defender becomes a "run" with a great result. Both look efficient, and the credit goes to the wrong family. Marisol codes RPOs as their own family for exactly that reason. How you classify a play changes what you learn from it.

The selection trap

The second problem is bigger, and it has a name: selection effects. Coaches don't call plays at random. We run when we're ahead late, when it's third-and-1, when it's fourth-and-goal at the 2 against a goal-line front, when the box is light. We throw when we're behind, when it's third-and-9, when the box is loaded. So when you compare "all runs" with "all passes," you aren't comparing two plays in the same situations. You're comparing two different sets of situations.

Here's a clean example from our own season. Our play-action passes averaged about 8.6 yards per dropback; our straight dropbacks averaged about 6.2. That's a big gap, and the easy conclusion is "play-action is worth almost two and a half yards a throw." But about four of every five of our play-action calls came on first or second down, while barely half of our straight dropbacks did — the rest were third downs, where the defense is waiting at the sticks and everyone's yards per attempt drops. When Marisol compared first and second down only, the gap shrank: about 8.6 against about 7.0. Play-action was still better. It was not two and a half yards better. Most of the difference had been the downs we chose to call it on.

🔍 Why this works (the selection trap, plainly): A number describes the plays you chose in the situations you chose them. Before you conclude that play A is better than play B, ask: would I have called play B in the same spots? If the answer is no, the comparison is partly a measure of your own call sheet. The fix is to compare like with like — same down, same distance band, same box count, same score state — and accept that the sample gets small when you do.

"Establish the run" — the honest ledger

So where does that leave the old coaching phrase "establish the run" — the idea that you must run early, and often, so the defense will respect it later and your play-action will work?

The honest answer has two halves. Some public analysts have argued, using league-wide data, that play-action's effectiveness does not depend much on how often or how well a team has been running in that game. That challenge is real, and coaches who say "we have to run it 30 times so play-action works" haven't met it. On the other side: a run game buys things that a per-play average doesn't count. It keeps defenses out of their pass-only structures (light boxes, extra defensive backs, two-deep shells on every down). It gives you a low-variance play when you're protecting a lead. It wins short yardage. And it puts specific defenders in the habit of reacting to a specific picture.

That last one is the piece I'll fight for, and it's how I won an argument with Marisol.

The argument I won: grade the family

Tuesday morning of Game 1 week, the 7 a.m. game-plan meeting. Our rule for disagreements is simple: make your case on tape, make it in one sentence, make it in reps, and then the coordinator decides — and the loser's dissent gets written in the margin of the plan, in ink, so we can look at it later.

Marisol had a dissent.

Marisol: Plain Inside from Gun Doubles on first down. Five games: it's slightly negative per play and a little under half successful. The passes from the same picture are well into the positive. I'd cut the plain handoffs on first down by a third and throw.

Lyle: You're grading the parent without its children.

Marisol: Say that in one sentence.

Lyle: Inside Slant, Inside Glance, and Inside PA only work because the plain Inside looks exactly like them, so grade all four as one family, not one at a time.

Marisol: That's a hypothesis. My number is a fact.

Wade: Then test the hypothesis. Lyle keeps his runs. Teo tags every play-action and RPO with the run it came from. Marisol charts the family. We look at it at the bye.

Marisol: And my dissent goes in the margin.

It went in the margin. Five days later, against Sacramento, we ran inside zone twice from that exact picture in the fourth quarter — +5 and +7 — and watched Tre Holloway, their strong safety, come downhill harder each time. On the third snap we faked it and Gaines ran the post behind him for 35 yards. The fourth snap was the same inside zone, and Holloway hesitated, and Ruffin walked in from the 4. If you'd graded those four snaps one at a time, two of them were "low-value runs." Graded as a family, they were a touchdown drive.

At the bye, Marisol brought the ledger.

THE INSIDE FAMILY FROM GUN DOUBLES -- 1st and 2nd down, Weeks 1-8
(Marisol's bye-week ledger; EPA per play on her model; each # = 0.05)

 Member                   Calls   Successful   EPA/play
 Inside (plain handoff)     58     26 of 58     -0.02   |
 Inside Slant / Glance      31     18 of 31     +0.19   |####
 Inside PA (same look)      22     12 of 22     +0.41   |########
 -----------------------------------------------------------
 The family                111     56 of 111    +0.12   |##

📊 Diagram (described): A small table-and-bar chart of the "Inside family" — every first- and second-down call from the same Gun Doubles picture over our first eight games. Plain inside-zone handoffs: 58 calls, 26 successful, about zero expected points per play. The RPO members (Slant and Glance): 31 calls, 18 successful, about +0.19 per play. Play-action from the identical look: 22 calls, 12 successful, about +0.41 per play. The whole family together: 111 calls, 56 successful, about +0.12 per play — solidly positive, even with the plain runs included.

And then she said the thing that makes her the best analyst I've worked with.

Marisol: You win the call sheet. You don't win the argument. This shows the family is good and the runs aren't dragging it down much. It does not prove the runs made the passes better. Teams that are good at play-action may just be good.

Lyle: Teo, show her the Holloway card.

Teo's card from Game 1 week tracked Holloway's first step against our inside zone from that picture: he triggered downhill on 15 of 21 snaps on Sacramento's film, and four for four in our fourth-quarter sequence. That card was the evidence Marisol's ledger couldn't contain — not "the run sets up the pass" as a general law, but "this run sets up this pass against this safety, and here are his feet."

Marisol: That's better evidence than my table. It's also twenty-five snaps of one safety.

Lyle: Then I'll keep checking his feet.

That's where we left it, and where I'd leave it with you. The run doesn't earn play-action by volume. It earns it by picture and by the specific defender who is taught to stop it. You can chart that defender's feet. If you can't show me the defender, Marisol's average wins.

🛡️ The defense's view: Russ Kimbrough's answer, which I didn't understand until we talked in the spring, was to give his rotator a different job in later games. By the time we saw them again, the rotator-side safety's first steps against our run looked far more mixed, and in Game 2 his big-nickel safety had one job against our two-tight-end runs: don't be interested in the run. A defense that reads your family ledger answers it by refusing to let one defender be the hinge. That's why the family has to be re-checked every week — the evidence is an opponent's habit, and opponents change habits on purpose.

🔄 Check your understanding: Marisol's ledger shows the Inside family at about +0.12 per play. Give one reason this does not prove the plain runs improved the play-action passes, and name the kind of evidence that would come closer.

Answer

Selection and team quality: the play-action calls might be efficient because of where we called them (early downs, favorable boxes) or because our passing game is good, regardless of the runs. Closer evidence is defender-level: does the conflict defender (here, Holloway) change his first step or depth after seeing the run, and do the play-action results improve when he does? Teo's trigger card is that kind of evidence — specific, countable, and checkable on film each week.


28.5 The Argument I Lost: The Reset Run

I promised you one I lost. I'm glad I lost it, and I'll tell you why at the end.

Every play caller has a comfort call — the play he reaches for when the last one went badly. Mine, for my first three seasons in Austin, was what I called the reset run. First-down pass falls incomplete; it's second-and-10; I call inside zone. The logic in my head went like this: get four or five yards, settle Darius down, keep their rush from teeing off, and set up third-and-medium, which was our best third down. In the four-game sample, we'd been in three-by-one on 22 of our 26 third-and-4-to-6 snaps, called dig or flood on 14 of them, and converted 8 of those 14. Third-and-medium was home.

Marisol's bye-week slide was titled, in her flat way, 2nd-and-8+: what happens next.

Weeks 1–8, 2nd-and-8 or longer (fictional) Snaps Successful (6+ on 2nd-and-10) Average 3rd-down distance after
Designed runs (not RPOs) 25 7 of 25 about 3rd-and-7
Everything else (dropbacks, PA, RPO throws, screens) 38 17 of 38 about 3rd-and-5
All 2nd-and-8+ 63 24 of 63

Marisol: You call the run on second-and-long about two times in five. It reaches the 60% line about one time in four. When it doesn't, you're in third-and-7 — and you know what our third-and-7 looks like.

Lyle: It's not about the run. It's about what the run does to the rush on third down.

Marisol: Then show me the third downs. After the reset run, our third-down conversion is a little worse than after a second-down pass, not better. Their pass rush doesn't seem to have noticed it was supposed to be slowed down.

Lyle: Third-and-medium is our best third down.

Marisol: It is. It's still a third down. The best third down we have is worse than the worst second down we have.

That was the sentence. I looked at Wade, who has a way of saying nothing that settles arguments.

Wade: When you call it, what are you hoping for?

Lyle: Five yards.

Marisol: You got six or more seven times in twenty-five.

Wade: Then you're hoping for something you get about a quarter of the time, to set up something we convert about a third of the time. I'd call that a punt with extra steps.

I lost. Here's what we changed. The reset run didn't disappear; it became half of a Kill call — two plays called in the huddle, the quarterback picking at the line by look. "Inside, Kill, Slants," or "Inside, Kill, Smash." If the box was light — six or fewer, our rule since Week 2 of "seven or fewer, run it; eight, take the throw" tightened by a man for second-and-long — Darius kept the run. Otherwise the pass was live. The run became something the defense had to earn, not something I needed.

2nd-AND-8 OR LONGER, BEFORE AND AFTER THE BYE (fictional, our offense)

                         Runs called          Successful snaps
 Weeks 1-8   (63 snaps)  25  ##########       24 of 63  ##########
 Weeks 10-18 (68 snaps)  14  ######           30 of 68  ############

 each # (runs) = about 2.5 calls; each # (success) = about 2.5 snaps

📊 Diagram (described): Two rows compare our second-and-long snaps before and after the Week 9 bye. Before the bye we called 25 runs in 63 second-and-long snaps and were successful on 24 of the 63. After the bye we called 14 runs in 68 snaps and were successful on 30 of the 68. The bar for runs shrinks by nearly half; the bar for successful snaps grows a little.

Marisol's caveat went on the slide too: the second half of the schedule had different defenses, Vance was better in November than September, and 68 snaps is not many. "It moved the right way" is all she'd sign. But you can see the new rule on film. In Week 10 at Portland, on 2nd-and-8 at their 42, we threw Drive — the dig over the Will — and Gaines caught it for 15. In Week 12, on 2nd-and-8 at the Stallions' 44, the zip motion told us Cover 3 and Gaines caught a hitch for 7. And on one Kill call that same day, on 2nd-and-8 at their 40, the box looked light, Darius kept the run, a safety crept late, and Ruffin lost 2. The rule isn't magic. It's just better than my habit.

Why am I glad I lost? Because when I was honest with myself, I wasn't calling the reset run for the offense. I was calling it for me. It was the play that let me breathe after an incompletion. Marisol's table didn't know anything about my nerves, but it described them perfectly: a call with a low success rate, made most often right after a failure, with no evidence that it helped the next snap. That's what tilt looks like when you count it.

⚠️ Common mistake: the comfort call. A coordinator's comfort call is made to manage the play caller's emotions, not the defense. You can find yours: sort your self-scout by "the play I call most often right after a negative play," then check its success rate against the same down and distance when it didn't follow a negative play. If the comfort call is worse — and it usually is, because the defense knows you're rattled too — turn it into half of a Kill call so the defense has to earn it.

🔄 Check your understanding: Wade called the reset run "a punt with extra steps." Using the numbers on Marisol's slide, explain in two sentences what he meant.

Answer

The run kept the drive on schedule only about a quarter of the time (7 of 25), and the rest of the time it produced a third-and-7-ish down that our offense converted only about a third of the time. So the call mostly led to a punt two snaps later — it "spent" a down without improving the odds, the way a deliberate punt would, just more slowly.


28.6 Fourth Down: The Break-Even Number

🧩 Productive struggle: Before you read on, decide the hook for yourself. Week 12, third quarter, you lead 13–10. Fourth-and-1 at your own 43. Use the illustrative ladder from 28.2. A punt nets about 40 yards and leaves the Stallions with first-and-10 near their own 17 (worth about +0.2 to them). If you go and convert, you'll have first-and-10 around your own 45 (about +1.7 to you). If you fail, they get first-and-10 at your 43 (about +2.3 to them). League-wide, fourth-and-1 converts somewhere around two-thirds to three-quarters of the time. Go or punt? And how often would you have to convert for "go" to be right? Work it out on paper before you look below.

Here's the arithmetic Marisol's chart did, in plain English.

FOURTH-AND-1, AUS 43, UP 13-10, THIRD QUARTER (illustrative EP)

 PUNT --> Stallions 1st-and-10 near their 17 ............. -0.2

 GO ----> convert (about 7 in 10): our 1st-and-10, ~AUS 45  +1.7
     \--> fail    (about 3 in 10): their ball at our 43 ... -2.3

          GO overall = (0.7 x 1.7) + (0.3 x -2.3) = about +0.5

 Difference: GO is worth about 0.7 points more than PUNT.
 Break-even: GO wins whenever we convert more than ~53 in 100.

📊 Diagram (described): A decision tree with two branches. The punt branch leads to one outcome — the Stallions starting near their own 17 — worth about −0.2 points to us. The go branch splits: about seven times in ten we convert and have a first down near our 45, worth about +1.7; about three times in ten we fail and the Stallions take over at our 43, worth about −2.3 to us. Blending the two go outcomes gives about +0.5, roughly 0.7 points better than punting. The bottom line states the break-even: going is the better choice whenever our chance to convert is above about 53%.

That last line is the most useful number in fourth-down football: the break-even conversion rate — the chance of converting at which going and kicking are worth exactly the same. You find it by asking, "what conversion chance makes the blended value of going equal the value of punting?" Here, the answer is about 53%. If you believe you'll convert more than about half the time, go.

And now the reason analytics changed fourth down in the NFL: for fourth-and-short across most of the field, the break-even is well below the rate at which teams actually convert. The punt feels safe because its outcome is certain and nobody gets blamed for it. But "certain" isn't the same as "valuable." A 40-yard punt from midfield-ish buys you about two points of field position for the other team's offense to overcome, and gives up roughly a seventy percent chance at a first down that is worth more than that.

📈 By the numbers: In recent seasons NFL teams have gone for it on fourth-and-short — especially between the 40s — noticeably more often than they did a decade or two earlier, and public fourth-down models are a big part of why. Those models (the nflverse community publishes one; several broadcasts now show others live) compare go, punt, and field goal using expected points early in games and win probability — the chance of winning from this score, time, and situation — late in games, when the scoreboard matters more than the average next score. League-wide, fourth-and-1 has generally converted somewhere around two-thirds to three-quarters of the time, with quarterback sneaks at the high end. Treat every one of those as a magnitude, not a precise figure.

The call that made the number true

The model assumes you have a fourth-and-1 play that converts about as often as the league's. That assumption is the offensive coordinator's responsibility, and it's why Wade's question is the right one. Here's the call we had for him in Week 12.

PRE-SNAP: 4th-and-1, AUS 43 -- "Heavy, Under Wing Right, Check"
          vs. the Stallions' short-yardage bear (8 in the box)

                            F   <- Carrow (#28), single high
   C                                                     C
                  S         M               $   <- SS walked down
              E       T     N     T       E
                U   O   O   O   O   O   Y
  X                         Q              B
                            R

📊 Diagram (described): Austin is under center in Heavy personnel: Brandt (U) attached on the left end of the line, Harlan (Y) attached on the right, Fale (B) a step off the line as a wing outside Harlan, Sutton (X) split wide left a yard off the ball, Ruffin directly behind Vance. St. Louis shows a bear front: a nose tackle head-up on the center and two tackles on the outside shoulders of both guards, so the center and both guards are all covered; ends outside each tight end; the Mike (Mabry) stacked three yards behind the nose; the Sam over the left side; the strong safety walked down outside the right end. Eight defenders are in the box, with one deep safety and two corners.

Our Check menu that week had three plays, all from the same picture: the sneak, duo, and play-action to Harlan. (A 3-technique is a defensive lineman on the outside shoulder of a guard; a nose head-up on the center is a 0-technique. Some systems number the tight-end alignments differently, but nobody argues about those two.) Vance's rule was one sentence: a bear gets the sneak; anything else gets duo; if the safety walks down and both linebackers creep inside the tight ends, take the play-action. Against a bear, duo is in trouble, because duo depends on double-teams at the point of attack and a bear puts a man on every interior blocker. But a bear also leaves nobody in the A gaps until the Mike gets there — and a sneak is a race to the A gap that the quarterback starts with the ball already in his hands.

POST-SNAP: the sneak -- C drives the nose; both guards fit their
           3-techs inside-out; Vance follows Szabo's right hip

                              M   <- Mabry fills from 3 yds, late
                              :
                    T    N    v    T
                    X    X         X
                   LG    C         RG
                             /
                         Q__/   right A gap: +1 1/2, first down

  Tackles and tight ends: man on or inside, nobody climbs.

📊 Diagram (described): At the snap, Szabo drives the nose straight back. Moss and Lorusso each block the tackle on their outside shoulder with their inside shoulder first, so neither defender can slant into an A gap. The tackles and tight ends block the man on them or inside them and stay on the line. Vance takes the snap and pushes up through the right A gap, between Szabo and Lorusso. Mabry comes down from three yards deep to fill the gap but arrives a half-count late, meeting Vance a yard and a half past the line — a first down.

Our own fourth-and-1 number that season was 9 of 12. Marisol would tell you twelve is a small sample; I'd tell you nine of them came from plays we'd repped for exactly that front. Both are true. The model tells you the price of the decision. The call sheet decides whether you're paying league-average or better.

🩺 Player safety: The sneak is a quarterback run into a pile. Count it. We budget Vance's designed runs for the season (about three a game in the regular season, more in the playoffs only with Wade's sign-off), and Rome counts sneaks against that budget in his head even when the stat sheet doesn't. Teach the sneaking quarterback to go low and forward with the ball tucked in both hands, and never to leap the pile. Pushing techniques behind the quarterback on sneaks have been debated by the league in recent seasons; check your current rulebook before you install any of them.

Four-down territory changes third down

Here's the part of fourth-down analytics that almost nobody outside a building sees. The fourth-down decision isn't made on fourth down. It's made before third down.

On Thursday, Marisol prints Wade a one-page chart for that opponent and that week's weather: for each distance (1, 2, 3, 4–5, 6+) and each field zone, it says GO, LEAN GO, TOSS-UP, or KICK, adjusted for Evan Dorsey's range (about 52 yards in normal weather — our "range line") and for our own conversion rates. On game day, when we cross into a zone where that chart says we'll go on fourth-and-short, Wade's decision reaches me before the third-down call — sometimes Teo relays it as his "one thing," sometimes Wade says it himself on my channel: "Four-down."

That changes my third-down call. If I know I'll have fourth down too, third down isn't one snap to gain the distance. It's two snaps. What that buys depends on the distance. On third-and-6, it means I call the play most likely to gain half of it safely, because a four-yard gain sets up a fourth-and-2 we'll go for and a sack is the one result that kills the drive (Chapter 26 works that case). On third-and-1 or 2, it means I can call something with a bigger payoff — the play-action shot we keep for short yardage — because its likely failure is an incompletion that still leaves a fourth-and-short we'll go for. In the four-game sample the Sentinels charted, we'd taken a play-action shot on 6 of 17 second-and-1-to-3 snaps. "Four-down" turns third-and-short into the same kind of free down.

🛡️ The defense's view: Russ told me in the spring that his scouting report said I was as aggressive on fourth down as anybody in our division. I told him to take that up with Wade. Then he told me what he did with it: "When you're four-down, I'm not defending third down. I'm defending two snaps for a yard apiece, and I'm saving my best short-yardage call for fourth." A defense that knows you're a four-down team can play third-and-short more conservatively — keeping its safety deep against the shot — and spend its best stunt on the down you can't afford to lose. The offense's counter is to make third down the shot down and fourth down the play you trust most.

Our season on fourth down

We went for it on fourth down 26 times and converted 15.

Distance (fictional, full season) Attempts Converted Notes
4th-and-1 12 9 Sneaks and duo; the Week 12 sneak at our 43
4th-and-2 to 3 7 4 Includes the Week 2 4th-and-2 at the STL 37, incomplete
4th-and-4 to 6 3 1 All passes
4th-and-7 or longer 4 1 All late, trailing — the chart and the scoreboard agreed
Total 26 15 11 of the 26 came between the 40s

The chart and Wade didn't always agree. Marisol's chart said GO or LEAN GO on 35 fourth downs; Wade went on 24 of them. He went twice when the chart said kick — both in fourth quarters, both converted. And eleven times he kicked or punted when the chart said go, including the fourth-and-goal at the Sacramento 3 in Game 1 and the fourth-and-1 at our own 36 in Game 2.

Was Wade wrong eleven times? Marisol's honest answer is "on average, probably a few of those cost us a little." Her more important answer is that Wade could name a reason for almost every one of them — which is the subject of the next section.

🔄 Check your understanding: Why does Wade say "Four-down" before the third-down call rather than waiting for fourth down? Give one way it changes the third-down call and one way a smart defense can answer.

Answer

Because knowing you'll go on fourth turns third-and-short into one of two downs to gain the distance, which raises the value of a higher-payoff third-down call (a play-action shot, a deeper concept) — a failure still leaves a fourth down you intend to play. A smart defense that knows you're a four-down team can play third down conservatively (keep the deep safety home against the shot, avoid a risky pressure) and save its best short-yardage call for fourth down.


28.7 When the Numbers Shouldn't Win

Every fourth-down model makes assumptions, and you should be able to say them out loud:

  1. You have a league-average play for this distance.
  2. You have a league-average kicker and punter (good charts adjust for yours — Marisol's adjusts for Dorsey).
  3. You're facing a league-average defense in this situation.
  4. Your players are healthy and fresh.
  5. Nothing has happened in this game that changes the odds.

Most of the time those assumptions are close enough. When they're not, the gap is information the model doesn't have — and the head coach and coordinator are the only people who do. But "the model doesn't know everything" is also the oldest excuse in football for doing what you were going to do anyway. So you need a test that separates the two.

"Name it, then price it"

Our rule is four words long, and it applies to every override of every number in the building: name it, then price it.

Name it means your objection must be a specific, observable thing the model can't see: their goal-line front has stopped three different looks in a row; our left tackle just left the game; Rollins has beaten Ben on four of his last five rushes; our best short-yardage call is being saved. "I've got a bad feeling" is not a name. "They've been good on defense today" is not a name — the model already assumes a defense.

Price it means you translate the named thing into a change in the conversion chance and run the arithmetic again. If the break-even is 53% and you think the injury drops your chance from 70% to 60%, you still go. If you honestly think it drops to 45%, you kick. Most named objections, priced honestly, don't flip the decision — they flip the call.

Wade's question — what's the call if we go? — is the same rule in six words. It forces the coordinator to name the play and, by naming it, price it.

Run the two Sacramento decisions through it.

Game 1, fourth-and-goal at the Sacramento 3, up 7–0. Marisol's chart said go, close to the line. Wade named two things: the Sentinels had just stopped three different looks from inside the 6, and our best goal-line call — Jumbo Power Left — was being saved for a moment that might matter more. The first is evidence about this front; the second is a real strategic asset. Priced honestly, it was a close decision moved slightly toward kicking. We kicked. Nine weeks later, that saved card scored from the 2 in Game 2. You can argue about it — I would have gone — but it passes the test.

Game 2, fourth-and-1 at our 36, early fourth quarter, down 20–13. The chart said go, "not by a little." Wade asked the question. My honest answer: a sneak if they show a bear, duo if they don't — and our runs against their big nickel (a fifth defensive back who is really a safety, playing near the line) had been bad all day, Rollins was beating Achterberg, and we'd thrown incomplete on a third-and-1 in the first half. I named it; I couldn't price it above break-even, because I didn't trust either play against that front, that day. Wade punted. The Sentinels kicked a field goal to go up 23–13, and we scored a touchdown on our next possession. Chapter 14 draws the plays I didn't trust.

The lesson isn't "the chart was wrong." The lesson is that the chart's first assumption — a league-average fourth-and-1 play — wasn't true for us that afternoon, and that's my fault, not the model's. The fix was to build a short-yardage package we trusted against big nickel, so the next time the chart said go, its assumption would be true.

Two kinds of gut

There are two very different things that coaches call "gut," and only one of them deserves a vote.

The first is expert pattern recognition: Gus Ferrante seeing on the sideline that the backside end is chasing flat down the line, or Teo seeing Holloway's feet trigger before the guard's second step. That kind of gut is data the model doesn't have, collected by trained eyes, and it can be named and priced.

The second is fear of blame. A punt that leads to a loss is "the defense didn't get a stop." A failed fourth down that leads to a loss is "the coach's decision." Nobody gets fired for a punt. That asymmetry is real, and it pushes every human being toward the safe-looking choice even when the safe-looking choice is worse. That kind of gut is exactly the bias fourth-down models were built to correct.

The test is simple: can you name it? Pattern recognition can always say what it saw. Fear can only say how it feels.

💡 Coach's point: "Name it, then price it." If you can name what the model can't see, translate it into a probability and re-run the decision. If you can't name it, the number wins.

📋 Try it — Make the call. (A hypothetical situation for practice; not one of our games.) Next season, fourth quarter, tied, 6:30 left. Fourth-and-2 at the opponent's 38. Ben Achterberg left with an ankle injury two snaps ago; Rodney Kalu has moved to left tackle, and the opponent's best edge rusher, aligned over him, has beaten him on both rushes since. A field goal from here would be about 55 yards — beyond Dorsey's range. A punt would pin them around their 10. On the illustrative scale: punt ≈ +0.1 for you; convert ≈ +2.8; fail ≈ −1.2. Marisol's chart says GO. Your fourth-and-2 menu: (a) Slants with five-man protection; (b) Duo Left, behind Kalu; (c) "Eleven, Gun Trips Right, Sprint Right, Snag"; (d) Inside Read. Wade asks, "What's the call if we go?" Do you go? And what's the call?

Answer

Name it: a backup tackle losing to their best rusher on the left side. Price it: first find the break-even. Going is worth p × 2.8 + (1 − p) × (−1.2); punting is worth about +0.1. Setting them equal: 4.0p − 1.2 = 0.1, so p ≈ 0.33. You only need to convert about one time in three. Even if the injury knocks your usual fourth-and-2 chance from the high fifties down into the mid-forties, you're well above break-even. Go.

The injury changes the call, not the decision. Duo Left asks Kalu to win at the point of attack against the man who's beating him — throw it out. Slants with five-man protection leaves Kalu alone on a speed rusher with the ball out in about 1.1 seconds; possible, but it puts the snap in his hands. Inside Read can read the edge rusher instead of blocking him, which is a real answer if your QB-carry budget allows it. The best answer is Sprint Right, Snag: the pocket moves away from Kalu, the back kicks out the right edge, and Vance reads the flat defender on the side with your healthy tackle. Tell Wade: "Go. Sprint Right, Snag. It doesn't need Rodney to win." That sentence is the whole skill.

Process over outcome: the one that failed

Two weeks before Week 12, at Portland, Wade went for it on fourth-and-1 at the Portland 18, late in the third quarter, with us up 17–10. A 35-yard field goal was nearly automatic; Marisol's chart said go anyway, because a touchdown there is worth so much more than three points and a failure leaves Portland on its own 18. We checked to duo against their short-yardage front, Dion Pratt slanted into the A gap across our right guard's face, and Ruffin was stopped for no gain. Portland took over and drove for a field goal. We won anyway, 27–13, and the Monday review graded the decision right, the call reasonable, and the execution lost the rep — in that order. Case Study 1 walks through it snap by snap, because a failed fourth down you still believe in is the best teacher this chapter has.

Remember what Chapter 25 calls the threshold idea of play calling: a call is a bet on a distribution, not a promise of an outcome. A 70% play fails three times in ten. If you grade fourth-down decisions by whether they worked, you'll teach yourself to punt — because the failures are memorable and the punts are invisible.

🔄 Check your understanding: In Game 1, Wade overrode the chart and kicked. In the Make-the-call drill above, you kept the chart's decision but changed the play. Using "name it, then price it," explain why those two outcomes are consistent with the same rule.

Answer

In both cases the objection was named (a front that had stopped three looks and a saved goal-line card; a backup tackle losing to the edge rusher). In Game 1, priced honestly, the named factors pulled a close decision just past the line toward kicking. In the drill, the break-even was so low (about one in three) that even a substantial drop in the conversion chance left "go" clearly better — so the named factor changed the call instead of the decision. Same rule; different prices.


28.8 Two-Point Decisions and the Chart

The two-point decision is fourth down's little brother: one snap, one yard line, and an arithmetic comparison against a kick.

In the NFL, a two-point try is snapped from the 2-yard line, and the extra-point kick has been snapped from the 15 since 2015. Over many seasons, NFL two-point tries have converted somewhere around half the time, and extra points since the move to the 15 have generally been made somewhere in the low to mid nineties percent. Two points times roughly a coin flip is about one point; one point times a little over nine in ten is a little under one point. On average, the two choices are worth about the same. That's why the two-point decision is driven almost entirely by the score — what the margin will be after the try, and what the opponent then needs.

The logic is easiest late in the game. Ask: after the try, what does the other team need to tie or win — and does one point more or less change the number of scores? Here's the chart we teach, written from the offense's side, using the margin after your touchdown and before the try:

Margin before the try Kick makes it Two makes it Late-game lean Why
Down 2 Down 1 Tied Go A kick still loses
Down 5 Down 4 Down 3 Go Down 3, a field goal ties
Down 8 (you were down 14) Down 7 Down 6 Go — and debated Convert and your next TD + PAT wins; fail and your next TD + two ties, the same as kicking twice would
Down 10 Down 9 Down 8 Go Down 8 is one possession; down 9 is two
Down 1 Tied Up 1 Kick, usually Go only if you'd rather finish it now (last possession, worse team in overtime)
Up 1 Up 2 Up 3 Go Up 3, their field goal only ties
Up 5 Up 6 Up 7 Go Up 7, their TD + PAT only ties
Up 12 Up 13 Up 14 Go Up 14, two TDs + two PATs only tie

The "down 8" row is the one analysts have pushed hardest in recent seasons, and a growing number of NFL coaches now go for two there; others still kick and play for overtime. It's debated because it depends on your conversion rate and on how much you value knowing, early, whether you'll need two again. Every other row is close to settled logic. Early in a game, kick unless your two-point play converts well above the league's rough coin flip, or the opponent's kicking game makes the extra point unusually risky.

Our season: three two-point tries, two conversions, none of them against Sacramento. Our base play is the sprint run-pass from Chapter 14 — "Eleven, Gun Trips Right, Sprint Right, Snag" — which threatens the whole width of the goal line and makes one flat defender choose. It lives on the Green panel of the wristband next to the goal-line runs, and it's on the call sheet every week whether we expect to use it or not, because a two-point decision you haven't prepared a play for is just a hope.

🏫 Any level: Everything in this chapter tilts toward going for it as you move down the levels. In high school, field goals and extra points are far less reliable — many programs don't have a kicker who makes 35-yarders — and punts are shorter and more often shanked or blocked, so the value of kicking falls and the break-even for going drops. College and high school snap the two-point try from the 3-yard line, not the 2, which nudges the two-point rate down; check your code. A few high-school coaches are widely known for almost never punting at all — Kevin Kelley at Pulaski Academy in Arkansas is the best-known example — and while that's an extreme, the arithmetic behind it is the same arithmetic in this chapter, applied to a level where kicking is worth less. You don't need a model to start: have a manager chart success rate (the 40/60/100 rule) by hand on the sideline, and after six games you'll know your own fourth-and-short conversion rate better than any league table can tell you.

🔄 Check your understanding: You score a touchdown with three minutes left to go up 5. Why do most charts say go for two, and what's the worst case if you fail?

Answer

Up 7 (if you convert) means the opponent's touchdown plus extra point only ties; up 6 (if you kick) means their touchdown plus extra point beats you. Failing leaves you up 5 — the opponent still needs a touchdown to win, exactly as they would if you'd kicked to go up 6. So the downside of failing is small and the upside is large.


28.9 Small Samples, Noise, and Talking to Coaches and Players

How big is big enough?

Football's samples are tiny. A team runs roughly sixty to seventy offensive snaps a game, split across dozens of situations. By the time you slice to "third-and-7 from two-by-two against a two-high shell," you might have nine snaps in a season. Here's the rough rule of thumb Marisol gives every coach who brings her a percentage:

Snaps in the sample A success rate can wander, by luck alone, roughly…
20 about 20 points either way
100 about 10 points either way
400 about 5 points either way

So a call that's "60% successful" on 20 snaps could honestly be a 40% call that got lucky, or an 80% call that didn't. That's why her tell threshold has two parts — about 70% and ten or more snaps — and why Marisol, looking at our 11-of-17 man-coverage number on third down, said: "It's sixty-five percent. My line for a tell is seventy percent on ten or more snaps. It's not a tell yet. It's a lean."

The flip side of small samples is regression to the mean: unusually good or bad numbers on small samples tend to drift back toward normal as more plays arrive. Inside Slant averaged a little over six yards a call through Week 14 and then gained about 13 yards on 4 calls in Game 2. Was the play broken? No — Sacramento had changed what Jordan Ames did, which is a different problem (the next one). But even without an adjustment, four calls is too few to tell a broken play from a bad afternoon.

The opponent reads the data too

Here's the thing that makes football analytics different from weather forecasting: the thing you're measuring is measuring you back. A defensive tendency isn't a coin with a fixed weight. It's a choice made by a coordinator who knows you're counting.

Before Game 3, Teo's page said the Sentinels had stayed two-high on third-and-7 or longer against two-by-two 29 straight times since their bye. Twenty-nine for twenty-nine is a beautiful number. It's also exactly the number a smart defensive coordinator would let you collect if he were saving one rotation for the biggest snap of the season. The thirtieth was the interception. "You let us get good at the wrong test," I told Russ in the spring, and I meant it as a compliment.

Two rules came out of that, and they're both on my whiteboard now. Perfect streaks get a question mark, not a bold font — a tendency with no exceptions is either a law of nature or a trap, and defenses aren't laws of nature. And in red, at the top: "Never let a tendency outlive its evidence." When a streak gets long, the question isn't "how sure are we?" It's "what would it cost us if it broke on the snap that matters most?" That's why our offseason project is the middle-read tag — a post-snap answer built into every third-and-long call, so the play survives if the tendency doesn't. Chapter 23 shows how the scouting page itself changed.

📋 Try it — Whose fault? (A teaching scenario built on our real Week 6 numbers; the staff decisions in it are hypothetical.) Teo's Week 6 report showed that on 26 third-and-7-or-longer snaps, the Sentinels brought pressure on 15. Suppose a staff saw only the headline — "pressure on 15 of 26 long third downs, a bit over half" — and responded by calling eight-man max protection with two-man routes on every third-and-long against Sacramento. Against Vance, the Sentinels sat in two-high coverage and rushed four on most of those snaps; the two-man routes ran into seven droppers, and the offense went 1 for 5 on third-and-long before halftime. Whose fault — the coverage, the protection, or the number?

Answer

The number — or rather, how it was cut. The 15 of 26 was real, but it hid its most important split: Sacramento pressured 8 of 9 long third downs against rookie quarterbacks and only 7 of 17 against veterans. Vance is a veteran with a reputation for punishing the blitz. Against him, the relevant base rate was well under half, not over half — so the max-protection plan bought protection against pressure that mostly wasn't coming and gave up three receivers to do it. The fix has two parts: split every tendency by the condition that matters most for your team (here, quarterback experience), and report counts, not percentages, so "7 of 17" and "8 of 9" can't hide inside "15 of 26." This is a classic aggregation trap: a combined number that describes no actual opponent.

Talking to coaches

Marisol's slides look the same every week, and the sameness is the point. Coaches stop reading a slide the moment they have to figure out how to read it.

+--------------------------------------------------------------+
| THE NUMBER    2nd-and-8+ designed runs: 7 of 25 successful   |
| THE SAMPLE    Weeks 1-8, all opponents, 25 calls             |
| THIS DOESN'T  Box counts; what happened on the snap before;  |
|  KNOW...      whether third-down rushes actually slowed      |
| WHAT I'D DO   Make the run half of a Kill call: light box    |
|               keeps it, anything else throws                 |
+--------------------------------------------------------------+

📊 Diagram (described): A boxed slide template with four labeled rows. The first states one number as a count ("7 of 25 successful"). The second gives the sample — which weeks, which opponents, how many calls. The third, "This doesn't know…," lists what the number can't see. The fourth, "What I'd do," gives a single concrete recommendation. Every slide Marisol makes follows this four-line shape.

One number per slide. The count, not the percentage. The blind spot stated by the analyst before a coach can find it. And a recommendation in football words — not "reduce designed-run share on 2nd-and-long by 40%" but "make the run half of a Kill call." The recommendation is what gets argued about. That's good. That's the meeting doing its job.

Talking to players

Players don't get EPA. They get film and one sentence — and the sentence is in their language, not ours.

Marisol's count of Ben Achterberg's pass sets showed his oversets bunched after an edge rusher's second speed rush in a series. Ben never saw that table. Gus showed him three clips — the second speed rush, then the overset, then the inside spin — and said one sentence: "The third rush is the one. Set for the spin." Darius doesn't get a hot-throw location chart; Rome gives him "the ball goes where the pressure left." Rules with names travel; percentages don't. Our whole scramble system is four numbered rules with names, and when Vance came off the field after the interception, he didn't say "I misread the rotation." He said, "Rule four, Coach." That's analytics, delivered. Somebody counted, a coach turned the count into a rule, and a player could say the rule back under the worst pressure of his season.

🎬 Film room: When you watch a broadcast that shows a fourth-down model's recommendation on screen, pause and chart three things the graphic can't show you: the defense's front (count the A-gap defenders), the offense's personnel (is it the team's short-yardage group, or whoever was on the field?), and the play call. Then watch the result. Over a season, you'll see that most failed fourth downs fail against a front the offense didn't have an answer for — which is the coordinator's problem, not the model's.

🔄 Check your understanding: Why does a 29-for-29 defensive tendency deserve less trust in a playoff game than a 23-for-29 tendency might, even though 29 for 29 looks stronger?

Answer

Because the defense is a thinking opponent who knows you're counting. A tendency with no exceptions may be a deliberate setup — a coordinator letting you build confidence in a look he plans to break on a critical snap. A tendency with exceptions shows you what the exceptions look like (situation, alignment, disguise), so you can build an answer for them. A perfect streak tells you nothing about how it breaks. That's why perfect streaks get a question mark and every third-and-long call needs a post-snap answer.


Spaced Review

Answer each before you read the reply.

1. From Chapter 25: The fourth-and-1 at Portland failed, and we still graded the decision "right." What is the threshold idea that justifies grading it that way, and what would happen to a staff that graded fourth downs by result?

A play call is a bet on a distribution, not a promise of an outcome. A 70% play fails three times in ten, and those failures are part of the bet you chose. A staff that grades by result will remember the failed goes and forget the invisible cost of punts, and slowly teach itself to kick when it shouldn't.

2. From Chapters 7 and 9: What do you count to apply the rule "seven or fewer, run it; eight, take the throw," and what evidence created the rule?

You count the defenders in the box — the area roughly inside the tight ends and within about five yards of the line, with an apex nickel counting as "half." The rule was born the Monday after Week 2, when Marisol's slide showed our inside runs into eight-man boxes at St. Louis had gained almost nothing; by season's end her numbers said the same thing louder — plain Inside was successful a little more than half the time against six-man boxes, a bit over four times in ten against seven, and a handful of carries averaging under two yards against eight. That's analytics in its simplest, most useful form: a count turned into a rule a quarterback can apply in three seconds.

3. From Chapter 23: Why does every perfect streak on our scouting page now carry a question mark?

Because a streak with no exceptions tells you nothing about how it breaks, and the opponent knows you're counting. The Sentinels' two-high-on-third-and-long streak was 29 for 29 until the snap that ended our season. A question mark forces a post-snap answer into the plan.


Project Checkpoint: Offensive Metrics Dashboard + Fourth-Down / Two-Point Decision Chart

Last chapter you built your Motion & Shift Menu — the diagnostic motions that make the defense show its coverage and the mismatch motions that move your best player where you want him. Now you'll build the two pages that tell you whether any of it is working, and the chart your head coach takes onto the field.

Part 1 — The Offensive Metrics Dashboard (one page). Ten lines, each with four columns: Metric, This week (as a count), Sample, and This doesn't know…. Required lines:

  1. Success rate — all snaps, and early downs only
  2. EPA per play if you have a model; yards per play and points per drive if you don't
  3. Explosives (runs 12+, passes 20+), and points on drives that had one vs. drives that didn't
  4. Third down by distance band (short / medium / long / long-long), as counts
  5. Red-zone trips and touchdowns
  6. Turnovers and sacks — with each sack bucketed by cause (lost the rep, held the ball, count, coverage)
  7. Early-down mix: runs, RPOs, play-action, dropbacks, screens
  8. Self-scout flags: any look at or above your tell threshold (about 70% on 10+ snaps) and any "zeros"
  9. Your comfort call's success rate, after a negative play vs. otherwise
  10. Fourth downs: attempts, conversions, and agreement with your chart

Part 2 — The Fourth-Down / Two-Point Chart (one page, laminated for the head coach). A grid of field zones by distance with GO / LEAN GO / TOSS-UP / KICK, adjusted for your kicker's range and your own short-yardage rate, plus the two-point table from 28.8. Here's a filled-in teaching sample:

SAMPLE FOURTH-DOWN CHART -- early and middle of games, normal weather
kicker range line ~opp 35.  GO / LG = lean go / TU = toss-up / K = kick

 Field zone            4-1   4-2   4-3   4-4/5   4-6+
 own 1-20              TU    K     K     K       K
 own 21-39             GO    TU    K     K       K
 own 40 to opp 40      GO    GO    LG    TU      K
 opp 39-31 (no-man's)  GO    GO    GO    LG      TU
 opp 30-11 (FG range)  GO    LG    TU    K       K
 opp 10-1 (goal to go) GO    GO    LG    TU      K

 Late in games: trailing moves every box toward GO; leading with a
 good defense moves the own-territory boxes toward K.

📊 Diagram (described): A six-row, five-column grid. Rows are field zones from your own goal line to the opponent's; columns are fourth-down distances from 1 to 6 or more. Fourth-and-1 says GO everywhere except deep in your own territory, where it's a toss-up. Going becomes more aggressive in the opponent's 31–39 — too far for a reliable field goal, too close for a useful punt. Inside field-goal range, the chart leans toward going only on short distances, and goal-to-go fourth downs from the 1 or 2 say GO. A note at the bottom says late-game score and time shift the boxes.

Worked line (pro level): Own 40 to opp 40, 4th-and-1: GO. Call: "Heavy, Under Wing Right, Check" — bear = sneak, otherwise duo, safety down and both backers inside = PA to the Y. Break-even about 53%; our season rate 9 of 12. Every GO box should name the call that makes it true.

Levels. High school: hand-chart success rate with a manager; replace EPA with points per drive; shift the whole chart toward GO if your kicker's range is under 30 yards; two-point from the 3. College: two-point from the 3; adjust for wider hashes on field goals from the far hash. Pro: use a public or in-house model and adjust for your kicker, your weather, and your short-yardage rate. Film-room: chart one team for four games from broadcast film — success rate, third downs by band, and every fourth-down decision against your chart — and write one "This doesn't know…" line per game.

Next: Chapter 29 asks you to build your Wide Zone & Boot Family — a "Shanahan Borrow List" of what you adopt, adapt, or reject from the system built on making everything look the same.


Chapter Summary

The metrics, and what each one can't see:

Metric What it measures Best use Blind spot
Expected points (EP) The value of a down, distance, and spot, from many seasons of results Comparing situations; fourth-down and two-point math Assumes league-average teams and plays
EPA Change in EP on one snap Valuing calls, families, and situations Noisy; one big play swamps a sample; credits the call and the player together
Success rate (40/60/100) Did the play keep you on schedule? Early-down efficiency; stable over small samples Ignores explosives; a 60-yard TD counts the same as a 10-yard gain on 3rd-and-9
Explosive rate (12+ run, 20+ pass) Ability to skip the schedule Predicting touchdowns vs. field goals The first thing defenses adjust to; volatile
Third down by distance The cliff past six yards Judging early-down calls by the third down they leave Small samples per band
Early-down pass rate / PROE How pass-heavy you are vs. the situation League comparisons; self-scout Selection: you choose when to run
Tendency rate (≥ ~70% on 10+) What a look tells the defense Self-scout and scouting Opponents change on purpose; perfect streaks may be traps

When should the number override your gut? — four questions, in order:

  1. Is the sample big enough? Counts first. Under about 20 snaps, it's a lean, not a fact.
  2. Is this the situation the model assumes? League-average play, kicker, defense, healthy players.
  3. Can you name what it can't see — and price it? Turn the named thing into a change in probability and re-run the comparison. Most of the time it changes the call, not the decision.
  4. Is your objection about the play, or about the blame? If you can't name it, the number wins.

Fourth down in one paragraph: find the break-even conversion rate (go is worth more whenever your chance beats it); for fourth-and-short across most of the field, it's well below what teams convert. Decide before third down ("Four-down"), which turns third-and-short into a shot down. Make the model's first assumption true by having a fourth-and-short call you trust against the front you'll see — and answer the head coach's question before he asks it: what's the call if we go?


What's Next

The best evidence in this chapter wasn't a model — it was Holloway's feet: one defender taught to stop a run, and a pass that looked exactly like it. Chapter 29 is about the system built entirely on that idea. Wide zone, the boot, and the play-action off both are the Shanahan tree's answer to the question Marisol asked me: what is a run worth if you only count the run?