Chapter 34 — Key Takeaways: Analytics and Defensive Decision-Making

Key Takeaways

  1. A number needs a definition, a denominator, a decision, and a sample size. Owen's three questions — what did you count, out of what, what decision does it change — plus "how many?" Without them, a number is an opinion with a decimal point.
  2. Story, lean, plan. Under about 10 snaps is a story; about 10–25 is a lean; about 25+ is a plan. A rough margin for any percentage is one over the square root of the count (36 → about ±17 points). Rare events (interceptions, fumble recoveries) are the fuzziest numbers of all.
  3. A tendency is a count plus a reason. Several independent kinds of evidence pointing one way — counts, mechanism, protection math, expert judgment — turn fuzzy samples into a plan.
  4. EPA puts every snap in one currency. Expected points is the average next score from a situation; EPA is the change a play causes. For a defense, negative is good. Last season the Sentinels allowed about −0.05 per play (−0.02 per dropback, −0.08 per designed run).
  5. Success rate measures how often; EPA measures how much. Use both. The two-number map tells you what kind of defense you are; the Sentinels moved from "boom or bust" in Russ's first season to "smothering" by cutting busts, missed tackles, and fit errors.
  6. Blitz rate is your call; pressure rate is a result; time to pressure decides whether it mattered. Pressure that arrives after the release is a highlight, not a stop. Against Vance, the hot throw (1.0–1.3 s) beats even a free A-gap rusher (1.6–1.9 s).
  7. Pressure is a means. A blitz pays only if the pressure it adds is worth more than the coverage it removes. Fire zones paid; five-man man-free pressures didn't.
  8. Season coverage tables lie through selection. Compare coverages inside a situation bin. Two Man looked best on the season and was worst on 3rd-and-7 to 10, because trail technique concedes the throw that moves the chains.
  9. Explosives are where games are lost — and only a third of ours were "coverage." Code every explosive by first cause; the fix follows the code.
  10. Turnovers are part habit, part bounce. Coach the upstream numbers — dangerous balls created, forced fumbles, hands on balls. Expect recoveries to land near half.
  11. Grade the decision and the result separately. Owen's receipt: the Week 6 Cover 0 call cost about six points as a result and about a quarter of a point per call as a decision — and it was wrong either way. Mug Cat Fire was a good decision that would have stayed good even if it had failed.

Action Items (this week)

  • Write a one-line definition for every number you currently track. Cut any that don't change a decision.
  • Print the confidence ladder at the top of your tendency report and label every number story, lean, or plan.
  • Clock 20 dropbacks of your next opponent: release time versus first-rusher arrival. Decide whether pressure rate is even the right number against them.
  • Re-sort your own coverage results by situation bin (at least 3rd-and-4–6, 7–10, and 11+) before you trust any season-long ranking.
  • Code the first cause of every explosive you allowed in your last three games.
  • Write a pre-receipt for your riskiest call: ledger, outcomes, expected value, box, limit, kill rule.
  • Build the Chapter 34 Project Checkpoint: your eight-number dashboard with definitions, sample notes, targets, and if-then triggers.

Common Mistakes (and the fix)

Mistake What it looks like Fix
Comparing across rulers "We're 30 percent at one service, they're 38 at another — they rush better." Compare within a single source and definition.
The coverage leaderboard Sorting the season table by EPA and installing more of the "best" coverage. Compare inside situation bins; demand the mechanism on film.
Blitz-for-pressure "The rush is cold — bring more." Fix the rush with games and launch-point plans first; check time to throw and the protection ledger.
Pressure as a target Rushers chase credit and abandon lanes; scramble yards rise. A pressure counts only if the rusher kept his lane (Goodhart's law).
Blaming the coverage for every explosive A new coverage installed after a bad stretch of fit errors. First-cause codes before Monday's meeting.
Celebrating recoveries A takeaway spike treated as a new standard. Track forced fumbles and dangerous balls; expect recoveries near half.
Resulting A bad call kept because it worked once; a good call dropped after a bad bounce. Two grades for every big call: the film and the card.
Story-sized certainty A call-sheet box built on 7 snaps. Story, lean, plan — and a reason on film.

Decision Framework — before a number changes a call

Step Ask If the answer is no…
1. Definition Do I know exactly what was counted, out of what, on whose ruler? Don't use it yet.
2. Size Is it a story, a lean, or a plan? How fuzzy is it (1 ÷ √count)? A story goes to the film room, not the call sheet.
3. Situation Was it measured in the situation I'm deciding about? Re-bin it.
4. Mechanism Is there a reason on film — a protection rule, a release time, a technique? Treat it as a coincidence until you find one.
5. Decision vs. result Am I grading what I chose or what happened? Write both grades separately.
6. Game state Late and close, do I need variance or am I paying for it? Tied or ahead: prefer the lower-volatility call.

The eight dashboard numbers (Sentinels): EPA per play allowed (pass/run) · success rate allowed · explosives allowed with first-cause codes · pressure rate (four-man / five-plus) with time to pressure · third-down conversions by bin · red-zone TD rate · takeaways with dangerous balls · missed-tackle rate. Watch line: communication busts.