Appendix H — Defensive Analytics Reference

How to use this appendix

Every number this book uses to judge a defense is defined here once: what it counts, out of what, a worked example, the traps, and how many snaps you need before you lean on it. Chapter 34 teaches the reasoning; this is the lookup table beside the call sheet. Read H.1 first. H.3 holds the Sentinels' season numbers, H.4–H.8 are the metric cards, H.9 is the weekly dashboard, and H.10 is a one-page formula card.

Every Sentinels, Outlaws, Stallions, and Ospreys number is invented for teaching — Owen Kessler's fictional charting. The methods are real. League-wide statements are hedged on purpose: they move from season to season and depend on who did the counting.

💡 Coach's point: Owen's rule, the first week he worked for me: a defensive number with no definition and no sample size is an opinion with a decimal point. If a number in your building can't answer the questions in H.1, it doesn't go on the front page.


H.1 Number Discipline: Four Questions, One Ladder, One Rule

The four questions

# Question What it protects you from
1 What exactly did you count? Two services, two definitions, two "pressure rates" for the same defense
2 Out of what? (the denominator) "Eleven explosives in Cover 1" means nothing until you know Cover 1 was played on 108 dropbacks
3 What decision does it change? Numbers that are interesting but move nothing on Tuesday
4 How many? (the sample size) Building a call around a coincidence

The confidence ladder (Owen's)

Snaps or attempts behind the number Owen calls it What you may do with it
Under about 10 A story Show it in the film room. Don't build a call around it.
About 10 to 25 A lean Tilt a decision when other evidence agrees
About 25 or more A plan Build a call-sheet box around it, and keep checking

The ladder isn't a statistical test, and Owen says so every time he uses it. It's a shared language for how hard to lean. Two companion rules: counts before percentages (write "14 of 22," never a bare "64%") and "a tendency is a count plus a reason" — a count with a mechanism you can see on film is evidence; a count alone is a rumor.

The square-root rule of thumb

For a percentage built from a count n, a deliberately cautious margin of uncertainty is one divided by the square root of n.

Count (n) Square root 1 ÷ √n Treat the percentage as fuzzy by about…
9 3 0.33 33 points either way
16 4 0.25 25 points
25 5 0.20 20 points
36 6 0.17 17 points
52 about 7.2 0.14 14 points
100 10 0.10 10 points
400 20 0.05 5 points

Worked example. Vance against five or more rushers: 28 of 36, about 78%, fuzzy by about 17 points (about 61% to 95%). Against four with rotation: 31 of 52, about 60%, fuzzy by about 14 (about 46% to 74%). The ranges overlap, so completion rate alone doesn't prove the gap — but yards per attempt, TD/INT, dangerous balls, and the mechanism all point the same way, and several independent signals pointing one way is how a fuzzy sample becomes a plan. Owen: "I'd bet on the direction. I wouldn't bet the house on the size." (The rule overstates the margin far from 50%; it's built to keep a coach humble.)

Sample-size guide, by metric

How fast each number fills up for one NFL defense, using last season's Sentinels volumes (about 1,050 snaps over 17 games).

Metric Denominator per game (approx.) One game is… A lean after… A plan after…
EPA per play, success rate ~62 snaps (~35 dropbacks, ~27 runs) Noise for EPA; a rough read for success rate 2–3 games 5–6 games (and still opponent-dependent)
Pressure rate ~35 dropbacks A story by rush type 3–4 games 6+ games
Blitz rate ~35 dropbacks Exact (your own call log), but one game is one plan Read it by situation Read it by situation
Explosives allowed ~3 events A story Half a season of first-cause codes A season
Third-down rate, by bin ~13 third downs, 3–4 per bin A story per bin 4–6 games per bin A season per bin
Red-zone TD rate ~3 trips A story Half a season A season or more
Missed-tackle rate ~60 tackle attempts A rough read 3–4 games 8+ games
Takeaways ~1.6 Luck Never — chart dangerous balls instead Dangerous balls over a season

H.2 Where the Numbers Come From (public sources)

Know what a source is built from, because that decides what it can and can't tell you. Products and access change; check what each one currently offers before you build a process on it.

Source Built from What it gives a defensive coach Watch out for
nflverse / nflfastR The league's official play-by-play, cleaned and published free, with open-source expected-points and win-probability models added EPA, success rate, win probability, explosives, situational splits (down, distance, field zone) for every team Play-by-play doesn't record the coverage called or how many rushed. Some third-party charting fields have been distributed alongside it for certain seasons; check what's currently included
NFL Next Gen Stats Player-tracking chips in the shoulder pads, recording position and speed many times a second Time to throw, receiver separation, speed, and other tracking-derived measures The public portion is a subset; definitions are the league's; tracking tells you where players were, not what their assignment was
Pro Football Focus (PFF) Human charters on every play, plus player grades Pressures by player, coverage snaps, targets allowed, turnover-worthy plays Charting involves judgment; definitions are PFF's own; much of it is subscription
Sports Info Solutions (SIS) Human charting of every play Coverage scheme charting, pressure and blitz charting, and their own value metrics Definitions differ from PFF's; numbers won't match across services
FTN / DVOA Football Outsiders' DVOA (a success-based, opponent-adjusted efficiency system), continued at FTN after Football Outsiders wound down, plus FTN's own charting Opponent-adjusted team defense ratings across many seasons DVOA is a model with its own baselines; compare it with itself, not with raw EPA

📈 By the numbers: Compare within a source, never across sources. Most charting services put league-average pressure rate somewhere between the mid-20s and mid-30s percent, and most of that spread is definition, not football. A defense that is "30 percent" at one service and "38 percent" at another hasn't changed; the ruler has.

What free play-by-play can build: EPA, success rate, explosives, third-down bins, red-zone rate, points per drive, and situational tendencies. What needs charting (yours or a service's): coverage played, rusher counts, blitz and pressure rates, time to pressure, missed tackles, dangerous balls, busts, loafs, hats, rotation rate, and PC@snap.


H.3 The Sentinels' Season Profile

The season profile (last season, Owen Kessler's charting — fictional)

Metric Sentinels Use
Nickel personnel share ~70% of snaps Ch 3, 4
Two-high pre-snap shell ~62% of snaps Ch 6, 30
Post-snap rotation (shell changed after snap) ~38% of snaps Ch 6, 32
Blitz rate (5+ rushers) 27% Ch 18, 34
Pressure rate (Owen's charting definition) 34% Ch 19, 34
Third-down conversion rate allowed 36% Ch 26
Red zone touchdown rate allowed 48% Ch 27
Missed tackle rate (Owen's charting) 9% (it was 14% in Russ's first season) Ch 23, 35
Explosive plays allowed (runs 12+ yds, passes 20+ yds — book's definition) 52 over 17 games Ch 34
Takeaways 27 Ch 35

The counts underneath it

Every rate above comes from a count; here are the counts.

Item Count Rate
Defensive snaps about 1,050 = 590 dropbacks (incl. sacks, scrambles) + 460 designed runs about 62 a game
EPA per play allowed all snaps about −0.05 (dropbacks −0.02, runs −0.08)
Success rate allowed (EPA > 0) all snaps 40% (dropbacks 41%, runs 38%)
Blitzes (5+ rushers) 159 of about 590 dropbacks 27%
Pressures 200 of 590 34% (four-man 133 of 431 = 31%; 5+ 67 of 159 = 42%)
Third downs 78 converted of 218 36%
Red-zone trips 52: 25 TD, 21 FG, 6 no points 48% TD
Tackle attempts 91 missed of 1,012 9.0%
Explosives 52 = 18 runs + 34 passes about 5.0% of snaps (passes 5.8%, runs 3.9%)
Takeaways 27 = 17 INT + 10 fumble recoveries about 1.6 a game
Dangerous balls 34 thrown, 15 intercepted 44%
Communication busts 9 (3 became explosives; 7 of 9 on the road) about 1 every 2 games

Russ's first season, for comparison: success rate allowed about 43%, EPA about +0.03, 29 communication busts, a 14% missed-tackle rate, 15 takeaways. The two-number map in H.4.2 shows where each season sits.


H.4 Efficiency Metrics: How Much, How Often, How Big

H.4.1 Expected points (EP) and expected points added (EPA)

Definition Expected points is the average number of points the team with the ball goes on to score next — counting the other team scoring next as negative — from a given down, distance, and field position (most models also use time remaining), built from many seasons of real play-by-play. EPA is the change in expected points from before a play to after it.
Formula in words EPA of a play = expected points after the play − expected points before it. EPA per play allowed = add up the EPA of every snap you faced, divide by the number of snaps.
Defense's sign Negative is good for the defense.
Split it Always report dropbacks (passes, sacks, scrambles — they all began as passes) and designed runs separately.
Sentinels, last season Dropbacks −0.02 (590); designed runs −0.08 (460); all plays about −0.05 (1,050)

The idea traces to early-1970s operations-research work co-written by Virgil Carter, then an NFL quarterback, and was popularized in The Hidden Game of Football (Bob Carroll, Pete Palmer, and John Thorn).

The ladder — intuition first. The values below are illustrative and rounded: the general shape public models produce, not any model's exact output.

EXPECTED POINTS, 1st-AND-10, BY FIELD POSITION (ILLUSTRATIVE ONLY)
Each # = 0.2 expected points for the offense.

 Ball on      EP     0    1    2    3    4    5    6
 own  5      -0.5   <|
 own 15      +0.2    |#
 own 25      +0.8    |####
 own 40      +1.6    |########
 midfield    +2.2    |###########
 opp  35     +3.0    |###############
 opp  20     +4.0    |####################
 opp   5     +5.5    |############################

📊 Diagram (described): A horizontal bar chart of illustrative expected points for an offense with first-and-10 at eight spots on the field. At its own 5 the value is slightly negative, about minus half a point. At its own 15 it is about plus 0.2; own 25 about 0.8; own 40 about 1.6; midfield about 2.2; the opponent's 35 about 3.0; the opponent's 20 about 4.0; and the opponent's 5 about 5.5. The bars lengthen steadily toward the opponent's goal line, faster inside the 20.

Field position gains value smoothly, and down and distance matter as much as the spot: third-and-8 at midfield is worth far less than first-and-10 there.

Worked examples (illustrative values, from Ch 34). Same three steps every time: state the situation before, state it after, subtract.

Snap Before (EP) After (EP) EPA
Game 1, Q4: Ruffin runs 16 on 1st-and-10 at the Austin 30 +1.0 1st-and-10 at the Austin 46: +1.9 +0.9
Game 1, Q4: 52-yard TD on 3rd-and-8 at the Austin 48 about +1.0 (likely punt) TD, minus the kickoff: about +6.5 about +5.5
Same snap, if stopped about +1.0 Punt; we start near our 15: about −0.3 for Austin about −1.3
Game 2, Q3: Mug Cat Fire sack on 3rd-and-7 at the Austin 38 about +0.6 4th-and-16 at the Austin 29: about −0.9 about −1.5
Same snap, if incomplete about +0.6 4th-and-7, punt: about −0.4 about −1.0

One snap outweighed any five ordinary ones; the stop-versus-touchdown swing was almost seven points; the sack beat an incompletion by only about half a point.

💡 Coach's point: "On third down, a sack and an incompletion are cousins. A first down and a touchdown are strangers." That sentence falls straight out of the table, and it's most of the analytics case against Cover 0 on third-and-long.

What a −0.05 means. Over a 62-snap game, about three points of expected value taken away compared with an average-ish defense. In most recent seasons, by most public models, the average dropback has been worth slightly more than zero to the offense and the average designed run slightly less; treat exact league averages as moving targets.

What EPA can't see: opponent quality (use an opponent-adjusted system such as DVOA), garbage time (trim or flag it), the decision behind a single play (one snap grades the result), and assignments (EPA knew the 41-yard post happened; only the film knew it was a bust). Win probability (WP), its cousin, asks "how much closer to winning?" — the better lens late and close.

H.4.2 Success rate

Definition The share of plays the offense would call a "win." There are two common versions — know which one you're reading.
Yardage version A play is an offensive success if it gains at least 40% of the yards needed on 1st down, 60% on 2nd down, and 100% on 3rd or 4th down. (Some analysts use 50% on first down.)
EPA version A play is an offensive success if its EPA is greater than zero. Used by nflfastR and most modern public work.
Formula in words Successful plays ÷ plays faced
Sentinels, last season 40% (EPA version; dropbacks 41%, runs 38%). Russ's first season: about 43%
League context Most public definitions put league-average offensive success rate somewhere in the 40s.

Worked example — a Week 11 Stallions drive graded both ways (EPA values rounded and illustrative):

Snap Situation Result Yardage rule EPA version
1 1st-and-10, STL 25 Day runs for 3 Needed 4 → fail about −0.3 → fail
2 2nd-and-7, STL 28 Haugen completes for 6 Needed 4.2 → success about +0.3 → success
3 3rd-and-1, STL 34 Day runs for 2 Needed 1 → success about +0.6 → success
4 1st-and-10, STL 36 Haugen boot, incomplete Needed 4 → fail about −0.6 → fail

Two of four: 50% on those snaps, and both versions agree here. They disagree at the edges — a 3-yard run on 1st-and-10 near the goal line can be positive EPA without reaching 40% — which is fine as long as you never compare a number built one way with a number built the other.

Why you need success rate and EPA. EPA tells you how much; success rate tells you how often. Put them on two axes:

THE TWO-NUMBER MAP (success rate allowed vs. EPA per play allowed)

                     EPA per play allowed
               LOWER (good)              HIGHER (bad)
            +-------------------------+-------------------------+
  SUCCESS   |  "Bend, then hit"       |  "Death by a thousand"  |
  RATE      |  They move the chains   |  They win often AND     |
  HIGHER    |  but rarely hit big;    |  hit big. Broken.       |
  (bad)     |  holds in the red zone  |                         |
            +-------------------------+-------------------------+
  SUCCESS   |  "Smothering"           |  "Boom or bust"         |
  RATE      |  Win most snaps, give   |  Win most snaps, but    |
  LOWER     |  up little.             |  explosives and busts   |
  (good)    |  Sentinels, last season |  erase it. Sentinels,   |
            |  (40%, -0.05)           |  Russ's first season    |
            +-------------------------+-------------------------+

📊 Diagram (described): A two-by-two grid. Horizontal axis: EPA per play allowed, lower (good) on the left, higher (bad) on the right. Vertical axis: success rate allowed, higher (bad) on top, lower (good) on the bottom. Top left, "Bend, then hit": offenses move the chains but rarely hit big plays. Top right, "Death by a thousand": offenses win often and hit big. Bottom left, "Smothering": the defense wins most snaps and gives up little; the Sentinels last season sit here at 40 percent and minus 0.05. Bottom right, "Boom or bust": the defense wins most snaps but explosives and busts erase it; the Sentinels in Russ's first season sit here.

H.4.3 Explosive plays and explosive rate

Book's definition A run of 12+ yards or a pass of 20+ yards. Scrambles count as dropbacks, so a scramble needs 20 yards to be explosive.
Formula in words Explosive rate = explosive plays ÷ snaps faced. Split runs (÷ designed runs) and passes (÷ dropbacks).
Sentinels, last season 52 in 17 games (about 3 a game) = 18 runs + 34 passes. About 5.0% of about 1,050 snaps; passes 34 of 590 = 5.8%; runs 18 of 460 = 3.9%
Other definitions Analysts use 10 and 20, 15 and 25, or one threshold for everything. Compare only within one definition.

First-cause coding. Every explosive gets one code: the earliest thing that went wrong — the one that, if fixed, would most likely have prevented the play. Secondary codes are allowed; the first cause drives the fix.

First cause Runs (12+) Passes (20+) Total Fixed by
Run-fit error (wrong gap, squeeze, wrong key) 11 — 11 Fit rules, eyes, keys (Ch 5, 20, 21)
Sound fit, then a missed tackle 4 — 4 Tackling (Ch 23)
Missed tackle after a catch short of 20 — 6 6 Tackling, rally (Ch 23, 26)
Good blocking, sound defense 3 — 3 Nothing — credit the offense
Communication bust — 3 3 Checks and echo (Ch 8, 14)
Safety false trigger on a run fake — 5 5 Eye discipline (Ch 22)
Lost in man (beaten, no technique error by the call) — 9 9 Technique, matchups, leverage (Ch 15, 17)
Concept beat the call (the coverage's known weakness) — 8 8 Call selection, disguise (Ch 9–12, 25)
Scramble / rush-lane break — 3 3 Rush lanes, scramble rules (Ch 33)
Total 18 34 52

Only 17 of 52 (lost in man plus concept beat the call) were "the coverage is broken." The rest were fits, tackles, busts, eyes, and rush lanes. The 11 fit-error runs, by the first defender out of his gap: lineman 5, linebacker 4, defensive back 2.

Worked example — coding Anchor C (Game 1, 45-yard inside-zone TD): first cause, run-fit error by a lineman (Greer's squeeze); secondary code, missed tackle "angle" (Holloway). It moves explosives, EPA, and missed-tackle rate — and success rate no more than a 7-yard run would. For comparison, the Outlaws produced 23 explosives in 247 snaps, about 9.3% (Appendix G).

H.4.4 Points per drive and points per red-zone trip

Definition Points per drive: opponent points ÷ opponent drives (a touchdown counts 7 unless you chart the try; exclude kneel-downs). Points per red-zone trip: points ÷ possessions reaching our 20.
Why it's not on the dashboard It's the outcome the other numbers explain, and it depends on where drives start — our offense and special teams.

Worked example 1 — end-of-half drives. Owen counted opponent drives starting in the final 2:00 of a half over three seasons through last season (kneels excluded): 46 drives — 7 TD, 10 FG, 29 no points. Points: 7 × 7 + 10 × 3 = 49 + 30 = 79. Points per drive: 79 ÷ 46 ≈ 1.7. Drives with a defensive DPI, holding, or roughing flag: 9 drives, 4 TD; without one: 37 drives, 3 TD — the lesson of Chapter 28.

Worked example 2 — red-zone trips. Last season: 52 trips, 25 TD, 21 FG, 6 no points. Points: 25 × 7 + 21 × 3 = 175 + 63 = 238, or 4.6 per trip. Illustratively, a 60% TD rate would be about 5.2 per trip (about 270 points) and a 40% rate about 4.2 (about 216): roughly fifty points a season between a good and a bad red-zone defense.

H.4.5 Third-down conversion rate, by bin

Definition Third downs converted (first down or touchdown on that snap, including by penalty if you choose — write it down) ÷ third downs faced
Bins (this book) Short 1–3, medium 4–6, long 7–10, very long 11+
Rule Always bin it. An average distance to go hides everything.
Sentinels, last season 78 of 218 = 36%. Average distance to go on third down: 6.8 yards
SENTINELS THIRD-DOWN CONVERSIONS ALLOWED BY DISTANCE, LAST SEASON
Each # = 4 percentage points.

 3rd-and-1 to 3    33 of 58   ##############     57%
 3rd-and-4 to 6    26 of 54   ############       48%
 3rd-and-7 to 10   15 of 62   ######             24%
 3rd-and-11+        4 of 44   ##                  9%
                  ----------
 All               78 of 218                     36%

📊 Diagram (described): A horizontal bar chart of the Sentinels' third-down conversion rate allowed last season by distance. Third-and-1 to 3: 33 of 58, 57 percent, the longest bar. Third-and-4 to 6: 26 of 54, 48 percent. Third-and-7 to 10: 15 of 62, 24 percent. Third-and-11 or more: 4 of 44, 9 percent, the shortest bar. All third downs: 78 of 218, 36 percent.

Reading it. Two defenses with the same overall rate can need opposite fixes, and the biggest lever is usually first and second down: every yard you add moves the snap into a better bin. Of the 20 completions that converted on 3rd-and-medium last season, 11 were caught short of the sticks — why underneath defenders finish at the sticks plus one (Chapter 26).

H.4.6 Red-zone touchdown rate

Definition Touchdowns ÷ opponent possessions that reach our 20-yard line (count the trip once, however many snaps it takes)
Sentinels, last season 52 trips: 25 TD (48%), 21 FG, 6 no points (2 INT, 1 fumble, 3 fourth-down stops)
Split worth keeping Trips that reached goal-to-go: 29, with 19 TD (66%). Trips that never did: 23, with 6 TD (26%). Keep them out of goal-to-go and the rate falls by more than half.
Goal-line runs Inside our 5: 31 carries, 12 TD — 3 TD on 19 carries with first contact at or before the goal line, 9 on 12 with first contact in the end zone.
Trap Small samples — about three trips a game. One weekend can swing it 20 points. Read it over half a season, and look at the goal-to-go split before you change the package.

H.5 Pass-Rush Metrics: Blitz, Pressure, Time, Sacks

Three words get mashed together in every broadcast booth. Keep them apart: blitz rate describes your call (you control it completely); pressure rate describes a result (you influence it); time to pressure decides whether the result mattered.

H.5.1 Blitz rate

Owen's definition A blitz is five or more rushers committed by about one second after the snap. A Hug (green dog) counts as a rusher only if he crosses the line before the throw. A sim pressure (four rushers, from unexpected places) is not a blitz.
Formula in words Blitzes ÷ dropbacks
Sentinels, last season 159 of about 590 = 27%
How others differ Some services call any linebacker or defensive back rushing a "blitz," even in a four-man rush, so your sims show up as blitzes in their data. Some count only rushers beyond four.
Read it by situation The overall rate is a budget; the situational rate is your tendency. 3rd-and-7+: 43 of 106 (41%) last season; 22 of 38 (58%) through Week 6 (see H.8.4).

Blitzes by type, last season (Ch 18 and Ch 34):

Blitz type Dropbacks Unblocked rusher within 2.5 s Pressures Pressure rate
Five-man fire zone with a lineman dropping 46 19 22 48%
Five-man fire zone, no lineman drop 42 9 17 40%
Five-man man-free (Cover 1 pressure) 41 7 14 34%
Six or more (Thin, six-man Cover 1, Zero) 30 11 14 47%
Total 159 46 67 42%

The six-plus row is Zero 10, Thin 8, and six-man Cover 1 12. "Unblocked within 2.5 s" is Owen's charting measure of whether the protection math worked — the six-line ledger from Chapter 18 written as a count.

H.5.2 Pressure rate

Owen's definition A pressure is a sack, a QB hit, or a hurry — a rusher forces the quarterback off his launch point or into an early throw — happening before or at the release. Credit goes to the first rusher to cause it.
Formula in words Pressured dropbacks ÷ dropbacks. Always split by rush count.
Sentinels, last season 200 of 590 = 34%. Four-man rush (incl. sims): 133 of 431 = 31%. Five or more: 67 of 159 = 42%
League context Most charting services put league average somewhere between the mid-20s and mid-30s percent, depending heavily on definition

How others differ: some services count pressures that arrive after the throw, some require a hurry to visibly change the throw, and some credit every contributing rusher.

Worked example — should we have blitzed more? Blitzing added about 11 points of pressure rate (42% vs. 31%) but only a small EPA edge (−0.04 vs. −0.02). Inside it, fire zones paid and five-man man-free pressures didn't — about +0.08 EPA per dropback, good for the offense. The answer: not more — differently. Pressured dropbacks averaged about −0.40 EPA and clean ones about +0.17 by Owen's charting; a blitz works only if the pressure it adds is worth more than the coverage it removes.

Where pressure comes from (first rusher, last season):

First pressure from Pressured dropbacks (approx.) Opponent completion rate Yards per attempt
Edge about 110 about 53% about 6.2
Interior about 60 about 40% about 4.6
Second-level blitzer about 30 not split out not split out
Total about 200

Interior pressure did the most damage: a quarterback can step up away from an edge rusher, not away from a man in his face.

⚠️ Common mistake — rewarding pressures without lanes. Grade rushers on pressures alone and they'll run past the launch point (Goodhart's law). A pressure counts only if the rusher kept his lane; a lost contain cancels it.

H.5.3 Time to pressure versus time to throw

Definitions Time to pressure: seconds from the snap to the first pressure. Time to throw: seconds from the snap to the release. Report medians or averages and look at the distribution — averages hide the hot throw.
Sentinels, last season Median time to pressure: blitz 2.1 s, four-man 2.6 s
Opposing quarterbacks Oduya (Ospreys): average release about 2.4 s. Vance (Outlaws) vs. a four-man rush: about 2.9 s
How to build it by hand On end-zone film, start a stopwatch at the snap and stop at the release, 20 straight dropbacks. Repeat, stopping when the first rusher reaches the quarterback's original spot.
THE RACE: WHEN THE BALL COMES OUT vs. WHEN PRESSURE ARRIVES
Seconds after the snap. Owen's charting (fictional).

                   0.0  0.5  1.0  1.5  2.0  2.5  3.0  3.5
                   |    |    |    |    |    |    |    |
Hot throw                    [==]  1.0-1.3
Free A-gap mugger                  [==]  1.6-1.9
Free edge rusher                        [===]  2.1-2.5
3-step last read                         [==]  2.2-2.5
Free DB from 6-8                           [===]  2.4-2.8
Deep over / dagger                             [====]  2.8-3.3
                   - - - - - - - - - - - - - - - - - - - -
Our blitz pressure                      ^  median 2.1
Our 4-man pressure                           ^  median 2.6
Oduya release                              X  avg 2.4
Vance vs. 4-man                                 X  avg 2.9

📊 Diagram (described): A timeline from zero to three and a half seconds after the snap. The top rows are time windows from Owen's film study: a hot throw comes out between 1.0 and 1.3 seconds; a free A-gap mug rusher arrives between 1.6 and 1.9; a free edge rusher between 2.1 and 2.5; a three-step drop's last read between 2.2 and 2.5; a free defensive back blitzing from six to eight yards between 2.4 and 2.8; a deep over or dagger between 2.8 and 3.3. Below a dashed line, carets mark the Sentinels' median time to pressure, 2.1 seconds on blitzes and 2.6 on four-man rushes, and X marks show Oduya's average release at 2.4 seconds and Vance's against a four-man rush at 2.9.

Reading it. Oduya releases before a four-man rush usually arrives (our pressure rate against Oklahoma City last season was about 19%), so grade short completions and tackling instead. Vance holds the ball against four, but blitz him and the hot throw beats even a free A-gap rusher: the blitz buys him a shorter decision, not us a pressure.

H.5.4 Sack rate

Definition Sacks ÷ dropbacks (dropbacks = pass attempts + sacks + scrambles)
Worked example Vance in the four-game Outlaws sample: 124 attempts + 7 sacks + 14 scrambles = 145 dropbacks. Sack rate = 7 ÷ 145 ≈ 4.8%.
Why it's not on the dashboard Sacks are a minority of pressures, swing far more from season to season, and depend partly on the quarterback holding the ball. Pressure rate, or a tracking-based measure such as ESPN's pass rush win rate (built from Next Gen Stats tracking), is steadier. Many pressures with few sacks often means more sacks next year; few pressures with many sacks, fewer.

H.6 Coverage Metrics: Efficiency, Selection, Man vs. Zone, Turnovers

H.6.1 Coverage efficiency ("coverage EPA")

Definition For each coverage family, the EPA per dropback, success rate, explosive passes, and pressure rate on the dropbacks where that coverage was actually played after the snap (not the shell shown, not the call on the sheet)
Formula in words Add up the EPA of every dropback in that coverage; divide by that coverage's dropbacks. Same for success and explosives.

Sentinels, last season (every dropback assigned to the coverage played):

Coverage family (post-snap) Dropbacks EPA per dropback Success rate 20+ yard passes Pressure rate
Cover 3 family (Sky, Spin 3, Buzz, Cloud) 148 +0.03 44% 12 31%
Quarters and Six 143 −0.05 40% 6 31%
Cover 1 family (incl. man-free pressures) 108 +0.02 43% 11 34%
Fire zones (five- and six-man) 96 −0.09 37% 3 44%
Two (plain Cover 2) 43 +0.06 46% 2 28%
Two Man 40 −0.12 33% 0 33%
Zero 10 −0.15 30% 0 60%
Other (Two Pole, prevent, goal line) 2 — — 0 —
All 590 −0.02 41% 34 34%
EXPLOSIVE PASSES ALLOWED (20+ YDS) BY COVERAGE FAMILY, LAST SEASON
Owen's charting (fictional). Each # = one play. Dropbacks in ( ).

Cover 3 family  (148)  ############   12    8.1%
Cover 1 family  (108)  ###########    11   10.2%
Quarters / Six  (143)  ######          6    4.2%
Fire zones       (96)  ###             3    3.1%
Two (Cover 2)    (43)  ##              2    4.7%
Two Man          (40)                  0    0.0%
Zero             (10)                  0    0.0%
                       ---------------------------
                       34 on 590 dropbacks = 5.8%

📊 Diagram (described): A horizontal bar chart of the 34 explosive passes allowed last season, one mark per play, by coverage. Cover 3 family: 12 on 148 dropbacks, 8.1 percent. Cover 1 family: 11 on 108, 10.2 percent. Quarters and Six: 6 on 143, 4.2 percent. Fire zones: 3 on 96, 3.1 percent. Plain Cover 2: 2 on 43, 4.7 percent. Two Man and Zero: none. Total: 34 on 590 dropbacks, 5.8 percent.

The chart says where explosives happened, not why. The first-cause codes in H.4.3 say why.

H.6.2 The selection trap: compare inside a situation

Sort the season table by EPA and Two Man looks like the best coverage in the building. It isn't. Coverages aren't called at random: Two Man is a third-and-long call, and on third-and-long almost every defense looks good. Zero is a ten-snap story. Plain Two lives on first-and-10 and in the red zone. Comparing those rows is comparing a closer's earned-run average with a starter's.

The fix: compare within a bin. Every 3rd-and-7 to 10 last season:

3rd-and-7 to 10, last season Snaps Conversions Conversion rate
Two Man 18 6 33%
Cover 1 family 9 2 22%
Quarters and Six 10 3 30%
Fire zones 18 3 17%
Cover 3 family 3 1 — (a story)
Runs faced (draws) 4 0 —
All 3rd-and-7 to 10 62 15 24%

The coverage that topped the season table was the worst on 3rd-and-7 to 10; its season EPA was carried by 3rd-and-11+. The film showed why: Two Man's underneath defenders play trail technique, conceding the short throw — a punt on 3rd-and-12, a first down on 3rd-and-8. Two Man moved to the 3rd-and-11+ box. A coverage that looks great only because of where you call it isn't great; it's well-placed.

H.6.3 Man versus zone splits

Definition A quarterback's (or a defense's) results split by whether the post-snap coverage was man or zone
Classification trap Match coverages look like zone before the snap and play like man after it (Chapter 31); services classify them differently. Most services show NFL defenses in zone on a clear majority of dropbacks, with man varying a lot by team.

Worked example — Vance [4-GAME]:

Vance Attempts Completions Yards / att Scrambles on called passes
Against man (Cover 1, Two Man, Zero, man pressures) 41 30 (73%) 9.0 9 — for 81 yards and 6 first downs
Against zone (incl. fire zones) 83 53 (64%) 6.6 5 — for 22 yards
Total 124 83 (67%) 7.4 14

The mechanism: in man every defender's back is to the quarterback, so a scramble has nobody facing it; and man invites Petrakis's rub, stack, and mesh answers (11 of 17 on third down when a defense showed man).

Worked example — Oduya, a different quarterback, a different answer:

Oduya vs. Sentinels corners Comp–Att Yards / att Completions of 20+
Off coverage 41 of 52 (79%) 6.3 0
Press 17 of 29 (59%) 5.9 3

Off gave up completions and denied explosives; press did the opposite. On first-and-10, take the 6-yard completion; on 3rd-and-6, it's the whole game.

H.6.4 Turnover-worthy plays and the "dangerous ball"

PFF's turnover-worthy play (TWP) A throw its charters judge to have a high chance of being intercepted, or a ball-security mistake likely to cause a fumble, whether or not a turnover happened. A judgment stat — two charters won't always agree.
Owen's dangerous ball Any throw a defender had a realistic play on — into a defender's zone or lane, behind a receiver into a trailing defender, or late over the middle — whether it was caught, dropped, or picked
Formula in words Dangerous-ball rate = dangerous balls ÷ attempts. Interception conversion = interceptions on dangerous balls ÷ dangerous balls.
Sentinels, last season 34 dangerous balls; 15 intercepted (about 44%). The other 2 of the 17 interceptions came on tipped balls off good throws.

Worked example — Vance's dangerous balls [4-GAME]:

Vance Attempts Dangerous balls Interceptions
Vs. 5+ rushers 36 1 0
Vs. 4 rushers, post-snap rotation 52 6 3
Vs. 4 rushers, shell held 36 2 1
Total 124 9 4

Against pressure, one dangerous ball in 36: the blitz handed him a pre-snap answer. Against rotation, six in 52: what he saw before the snap wasn't what he got after it. Three interceptions is a story; six dangerous balls with a mechanism is a lean worth building a disguise around.

Owen's decision rule: a high dangerous-ball quarterback rewards aggression (robbers, traps, disguised rotation); a low one, like Oduya, punishes it — play sound, rally, tackle.

H.6.5 Takeaways: habit upstream, luck downstream

Habit: creating dangerous throws, getting hands on the ball, and attacking the ball on every tackle. Partly luck: catching the ball you touched, and whether a punch knocks it loose. Mostly luck: who recovers a loose ball — close to a coin flip in most public analyses, with little carryover from season to season.

Worked example — fumbles. Last season opponents fumbled 21 times (14 forced by a strip or punch — the habit), and we recovered 10, right where the coin usually lands. Owen projected "about half of whatever we force," not "10 again," so the staff coached forcing (Chapter 35).

Worked example — regression. Vic Maddox's defense (Chapter 34, rebuilt from public play-by-play) recovered 16 of 24 fumbles in Year 1 (about 67%), then 9 of 20 and 7 of 15. Takeaways fell 31 → 21 → 16 while success rate allowed barely moved (45%, 45%, 46%) and explosives climbed (64, 69, 77). The bounce came back to earth; the explosives were the real trend.

🩺 Player safety: "Attack the ball" is a tackling skill, not a license. The strip comes after the ballcarrier is wrapped — "first man stops, second man attacks the ball." Count strip attempts in practice only when the tackle was secure, and never reward leading with the helmet to dislodge the ball.


H.7 Tackling, Assignment, and Effort Metrics

The "cheap fixes" that moved the Sentinels from boom-or-bust to smothering. All come from your own film, graded the same way every week.

H.7.1 Missed-tackle rate

Definitions A tackle attempt is any time a defender has a realistic chance to make the tackle — he gets contact, or he's close enough that sound technique would have produced contact. A missed tackle is an attempt that fails: the runner breaks contact, slips past, or makes him whiff, and keeps moving forward.
Formula in words Missed tackles ÷ tackle attempts
Owen's charting rules Every defender who makes a legitimate try is charted made or missed, even if someone else finishes the play two yards later. An assisted tackle is two made attempts. Two coaches chart, and they must agree.
Sentinels First season 149 of 1,066 (14.0%); second season about 11%; last season 91 of 1,012 (9.0%); Game 1 at Austin 12%
Russ's rule of thumb Teams that miss 15% or more of their tackles tend to finish in the bottom half — a rule of thumb from experience, not a cited finding

Coded by type (first season → last season):

Miss type First season Last season
Angle (wrong pursuit or alley angle) 41 19
Lunge / no breakdown 37 22
No wrap 29 17
High / arm tackle 22 15
Runner won (sound technique, better athlete) 20 18
Total 149 91

Yards after a miss: 7.4 in the first season (about 1,100 in all), 6.8 last season (about 620). The coachable categories fell by more than half; "runner won," which measures the other team's talent, barely moved.

H.7.2 Busts and communication busts

Definitions A bust is a snap where at least one defender plays an assignment that doesn't match what the other ten are playing. A communication bust (Owen's narrower count) is a snap where two or more defenders played different calls. A coaching bust is one the staff caused (a call that didn't fit the formation, a check never installed).
Formula in words A count, not a rate — tallied by type: call didn't arrive, check didn't arrive, check made by the wrong player, check made late, check not echoed
Sentinels Communication busts: 29 in Russ's first season (11 became explosives); 9 last season (3 explosives; 7 of the 9 on the road). Coaching busts: 11 last season (Owen thinks the count is too low). Game 1 at Austin: 1 (the Q1 Lock/Quarters split, a 41-yard post)
Where it lives On the dashboard's watch line, not the front page — unless busts come back

Worked example — the Game 1 bust. Half the secondary played Lock and half played plain Quarters; Holloway executed Quarters correctly with the wrong call. Coded as a communication bust, not technique. The fix was systemic: "No echo, no check," and Royce alone owns coverage checks.

H.7.3 Loafs

Definition A loaf is a visible drop in effort before the whistle by a defender who could still affect the play, or whose job required him to keep running.
Grading Every player, every snap: +, L (loaf), or M (medical — never an effort grade). A second coach confirms every L from a cutup that hides the scoreboard. Report confirmed loafs per game.

The five categories:

Category What it looks like on film
Speed loss Visibly slows from full speed while the ball is alive
Late turn No turn-and-sprint within a step of the ball being thrown, caught behind him, or reversing field
Quit on the block Stops fighting a blocker and watches the ball
Quit early Pulls up before the whistle, assuming the play is over
Slow up Stays down, or gets up slowly, with the play alive and no sign of injury

Not a loaf: arriving late because the assignment held you (effort plus); being outrun by a faster player; a wrong angle at full speed (technique); the wrong gap (a bust); pulling up hurt (a trainer's question, never the loaf chart).

Sentinels loafs Total Per game
First season (charted retroactively) 212 about 12.5
Second season 104 about 6.1
Last season 46 about 2.7
This season, Weeks 1–6 14 about 2.3 (Game 1: 4 confirmed)

Caveat: Owen estimates a quarter to a third of the first-to-second-season drop may be grader change, not player change.

H.7.4 Hats at the whistle

Definition The number of our defenders within five yards of the ball carrier when the whistle blows, charted on every run and every completion caught within five yards of the line of scrimmage (the plays where everybody can get there)
Formula in words Total hats ÷ charted plays = average hats per play
Sentinels 3.9 (first season, retroactive) → 5.0 → 6.1 last season → 6.4 this season through Week 6 (Game 1 at Austin: 5.8)
Caveats Our own charting, not a public statistic. A missed-tackle week drags it down too. The count is not the grade: a defender held back by his assignment isn't a missing hat.

H.7.5 Ball attacks

Definition Strips, punches, and rakes at the ball after the wrap, counted per game
Sentinels About 7 per game in the season it was first charted → about 11 last season
Why it's charted It's the habit upstream of forced fumbles. Recoveries are near a coin flip; attacks aren't.

H.8 Disguise and Call-Sheet Metrics: Measuring Yourself

These numbers grade your predictability. They're the core of the self-scout (Chapter 25): your tendencies are the opponent's scouting report.

H.8.1 Rotation rate

Definition The share of snaps from a given pre-snap shell on which the shell changed after the snap (two-high to one-high, or the reverse)
Formula in words Snaps rotated from a look ÷ snaps that look was shown. Always compute it per look — a season-wide rate hides a look that never rotates.
Sentinels Last season, post-snap rotation on about 38% of all snaps. This season, from Big two-high on 1st-and-10 through Week 10: 31 of 58 (53%)
Target From any one pre-snap look on early downs, rotate about 40–60%, and never three in a row

Worked example — what rotation buys in the run game (last season, Big vs. 12 personnel):

Big vs. 12 personnel Two-high held Rotated
Designed runs 64 carries, 5.2 yards per carry 41 carries, 3.6 yards per carry
Play-action 21 attempts, 6.3 yards per attempt 17 attempts, 9.8 yards per attempt

Rotation took 1.6 yards a carry off the run game and gave 3.5 yards an attempt to play-action — the trade from Chapter 30. The 40–60% target keeps either picture from becoming a tendency. And rotation only disguises if the safeties start as twins (10–11 yards on run downs, 12–13 on passing downs): the self-scout caught Holloway at 10 yards or less rotating down on 17 of 19.

H.8.2 PC@snap (play-clock seconds at the snap)

Definition Seconds left on the play clock when the ball is snapped. Owen added it as a column to the charting sheet.
What it measures How long the quarterback needed at the line — a proxy for how much your look made him think, and how much time he had to check out of the call
Worked example Vance's average PC@snap on third down: Game 1 about 9 s; Game 2 against two-high and Home looks about 8 s; Game 2 on 11 snaps with a mug, walked-up nickel, or late walk about 3 s, twice with 1 second left. Anchor B was snapped with about 3 seconds left.
Caveat Owen: "a lean, not a plan." Offenses also milk the clock for reasons unrelated to you; compare like situations.

H.8.3 Look-to-outcome table

Definition For each pre-snap look, how often it ended in each rusher count (4 / 5 / 6). It answers: if an offense knows the look, does it know the rush?
Rule of thumb No third-down look should land on one rusher count much more than about 60% of the time.

Sentinels, this season through Week 15 (all downs):

Look Snaps 4 rushers 5 rushers 6 rushers
Mug 44 19 22 3
Home 312 271 39 2
Walk 17 11 6 0
Ghost 23 15 8 0

Home is honest on most downs by design; the rule bites on third down. Last season's 58 mug snaps landed on five rushers 33 times (57%), close to the line; this season's 44 split 19 / 22 / 3.

H.8.4 The self-scout rates (Owen's bye-week packet, Weeks 1–6)

Self-scout line Count
5+ rushers on 3rd-and-7+ 22 of 38 (58%) — last season 41%
5+ rushers on 3rd-and-7+, Q4, within 8 points 7 of 8 (the Week 6 zero was the 7th)
Four-man calls that showed two-high 15 of 16
Pressures from one-high or walked-down looks 21 of 22
Two-high shown on 3rd-and-7+ that stayed two-high 14 of 15
Mug look → pressure 6 of 6
Holloway at 10 yards or less → rotated down 17 of 19
Vo within 3 yards of the line inside the slot → blitzed 10 of 11
Home on 1st-and-10 vs. 11 personnel 2x2 61 of 97 (63%)

Each line gets a sort: Break (hands the offense a free answer — fix it), Bait (keep it and plan the counter; the call sheet's "HOLD" line), or Accept (predictable by design).


H.9 The Weekly Dashboard and the Decision Receipt

H.9.1 Eight numbers, and the decision each one drives

Definitions are in the cards above and in H.10.

# Number Decision it drives
1 EPA per play allowed, pass and run Where to spend practice time: pass game or run game
2 Success rate allowed Whether we're losing the every-snap fight or only the big ones
3 Explosives allowed, first-cause coded Which tool to fix: fit, tackle, bust, eyes, or call
4 Pressure rate by rush count, plus median time to pressure Rush plan and pressure package
5 Third-down rate allowed, by bin Third-down menus
6 Red-zone TD rate allowed Red-zone package
7 Takeaways and dangerous balls Takeaway drills; how aggressive the coverage plan can be
8 Missed-tackle rate Tackling periods

Not on the front page, on purpose: yards allowed, points allowed (the outcome the eight explain), and sacks. Watch line: communication busts. Example trigger: if explosives exceed 4 in a game, every one is first-cause coded before Monday, and the first fifteen minutes go to the most common code.

H.9.2 The Week 6 card

SENTINELS DEFENSE -- WEEKLY DASHBOARD (Owen Kessler; fictional)
WEEK 6 at AUSTIN: Outlaws 34, Sentinels 27   (1 game = a story)

 #  NUMBER                         GAME       LAST SEASON   TARGET
 -  ----------------------------   --------   -----------   -------
 1  EPA/play allowed               +0.16      -0.05         < 0.00
      pass | run                   +.24|+.01  -.02|-.08
 2  Success rate allowed           49%        40%           < 42%
 3  Explosives allowed (R12/P20)   5 (2R 3P)  52 (18R 34P)  <= 3/gm
 4  Pressure rate                  22%        34%           >= 32%
      4-man | 5+                   22%|22%    31%|42%
 5  3rd-down conversions allowed   8 of 14    36%           < 38%
 6  Red-zone TD rate allowed       2 of 3     48%           < 50%
 7  Takeaways | dangerous balls    0 | 1      27 | 34       --
 8  Missed-tackle rate             12%        9%            < 10%
 -------------------------------------------------------------------
 WATCH (not graded): comm. busts 1 (Lock/Quarters split, Q1)

📊 Diagram (described): A one-page dashboard for the Week 6 game at Austin, a 34–27 Outlaws win, labeled "one game equals a story," listing each number's game value, last season's value, and target. EPA per play allowed: plus 0.16 (plus 0.24 passing, plus 0.01 rushing) against minus 0.05; target below zero. Success rate: 49 percent against 40; target below 42. Explosives: 5 (2 runs, 3 passes) against 52; target 3 or fewer a game. Pressure rate: 22 percent on both four-man rushes and blitzes, against 34; target 32 or higher. Third downs: 8 of 14 against 36 percent. Red-zone touchdowns: 2 of 3 against 48 percent. Takeaways and dangerous balls: 0 and 1. Missed-tackle rate: 12 percent against 9. Watch line: one communication bust.

Reading it. The five explosives — Harlan's 22-yard dig, Gaines's 41-yard post, Ruffin's 45- and 16-yard runs, Gaines's 52-yard TD — include three coded to us: a bust, a fit error, and a call. Pressure was cold on blitzes too, which should have said blitzing wasn't the cure. (A 12-yard Vance scramble counted as a dropback, not an explosive.)

H.9.3 The decision receipt: grade the call and the result separately

Owen's receipt is one page per big call with two rows: what the play cost (the result) and what the call cost (the decision, as expected value). Resulting — judging a decision by its outcome, a poker term popularized by Annie Duke's Thinking in Bets — is how a bad call with a good outcome gets called again.

Worked example — the Week 6 Cover 0, per 100 third-and-8s near midfield. Every number here is illustrative, a teaching sketch of the shape, not league data:

Per 100 third-and-8s (illustrative) EPA each (to the offense) Four-man, two-high, rotation Six-man Cover 0
Ordinary stops (incompletion, short completion) −1.3 51 → −66.3 40 → −52.0
Sacks −1.7 6 → −10.2 14 → −23.8
Takeaways −3.5 3 → −10.5 4 → −14.0
Ordinary conversions +1.8 35 → +63.0 29 → +52.2
Gains of 20+ (not touchdowns) +3.5 4 → +14.0 7 → +24.5
Touchdowns +6.0 1 → +6.0 6 → +36.0
Total per 100 snaps −4.0 +22.9
Per snap about −0.04 about +0.23

Check the arithmetic. Four-man: stops 51 + 6 + 3 = 60, conversions 35 + 4 + 1 = 40. Zero: stops 58, conversions 42. Multiply each count by its value, add, divide by 100. The gap is about a quarter of a point per call against an ordinary quarterback, and wider against Vance's 28 of 36. Zero's extra sacks were worth about 13.6 points per hundred; its extra touchdowns cost 30. Add the game state — tied, 4:12 left, a stop was enough — and the receipt's bottom line reads: grade the decision, not the result. The decision was still wrong.


H.10 Formula Card

Metric In plain words Sentinels, last season
EPA per play Sum of (EP after − EP before) ÷ snaps −0.05 (−0.02 / −0.08)
Success rate Successful plays ÷ plays 40%
Explosive rate Explosives ÷ snaps 52 (≈5.0%)
Points per drive Points ÷ drives End-of-half drives: ≈1.7 (three seasons)
Third-down rate Conversions ÷ third downs 36% (78 of 218)
Red-zone TD rate TDs ÷ trips reaching our 20 48% (25 of 52); 4.6 points per trip
Blitz rate 5+ rusher snaps ÷ dropbacks 27% (159)
Pressure rate Pressured dropbacks ÷ dropbacks 34% (31% / 42%)
Time to pressure Snap → first pressure 2.1 s blitz / 2.6 s four-man
Sack rate Sacks ÷ (attempts + sacks + scrambles) Vance [4-GAME]: 7 of 145 ≈ 4.8%
Coverage EPA EPA ÷ dropbacks, by post-snap coverage Quarters/Six −0.05; Cover 3 +0.03
Man vs. zone Results split by post-snap coverage type Vance: 9.0 vs. 6.6 Y/A
Dangerous-ball rate Dangerous balls ÷ attempts 34 thrown, 15 picked
Takeaways INT + fumble recoveries 27 = 17 + 10
Missed-tackle rate Misses ÷ tackle attempts 9% (91 of 1,012)
Communication busts Count, by type 9 (3 explosives)
Loafs Confirmed L grades per game about 2.7 per game
Hats at the whistle Defenders within 5 yards of the ball carrier at the whistle ÷ charted plays 6.1
Rotation rate Snaps rotated ÷ snaps shown, per look about 38% of all snaps
PC@snap Play-clock seconds left at the snap Vance on 3rd down: ≈9 s (Game 1)

Before you let any number change a call: (1) Definition — same ruler? (2) Size — story, lean, or plan, and how fuzzy? (3) Situation — measured inside the situation you're deciding? (4) Mechanism — what's the reason on film? (5) Decision versus result — which are you grading? (6) Game state — does this call add variance you need, or variance you're paying for?