Personnel-conditional NFL engine

The playcalling DNA of every staff in the NFL.

Built from per-play tracking data — personnel, formation, coverage shell. Every coordinator's tendencies by down and distance, the drift when staffs change, and an early research engine that projects how a new staff is likely to lean.

500K+ plays in the cube 85 coach DNA vectors Validated vs PFF
Coordinator DNA · SF — K. Shanahan2023 · OFF
Pass Run EPA/play
1st & 10
PASS 52RUN 48
+.18
2nd & long
PASS 71RUN 29
+.09
2nd & short
PASS 38RUN 62
+.21
3rd & short
PASS 44RUN 56
+.05
3rd & long
PASS 89RUN 11
−.04
Red zone
PASS 47RUN 53
+.12
38% under-center (above league) · 31% 21-personnel (vs 8% league avg). Generated from the cube in seconds — real plays, real personnel.
10
Seasons of per-play personnel
500K+
NFL plays in the cube
85
Coach DNA vectors
4,180
Player archetype-seasons
The engine

Three layers. One cube.

A descriptive fingerprint of every staff, a function that predicts what a new hire will do, and a simulator that runs the game ten thousand times.

01 · DESCRIPTIVE CUBE Shipping now

The gameplan fingerprint

Every team's identity across 20 dimensions: formation mix, personnel mix, situational pass rates, EPA by situation. Cross-season normalized, coach-aware drift over time.

20 dimensions · 32 teams · 10 seasons
02 · PREDICTIVE COMPOSITION Beta Q3 2026

What a new hire will call

f(new coach, current roster) → projected Year-1 fingerprint. Blends coach DNA, college lineage, roster archetypes and team-history anchor. Backtested on held-out staffs — honest about where it holds and where tendencies revert to the mean.

In development · validated against the baseline
03 · MONTE CARLO SIM Enterprise pipeline

Run the game 10,000 times

Sample per play from P(outcome | personnel, coverage, coach play-call). Output: win-probability and scoring distributions for calibration and scenario analysis — scout-grade, not a betting product.

Per-play conditional sampling
Coach & coordinator intelligence

Read a coordinator like a fingerprint.

The depth no one else ships. Self-scout your own tendencies, or strip an opponent's playbook down to down, distance and situation — then watch how it shifts the moment a staff changes.

  • A

    Tendencies by down & situation

    Pass/run split, formation and personnel leaning, and EPA for every situation bucket — 1st & 10 through 3rd & long, two-minute, red zone.

  • B

    Coaching-era drift

    The Roman→Monken change at BAL in one column: pistol collapsed 33%→7%, shotgun exploded to 78%. The coordinator drives the gameplan more than the head coach.

  • C

    College lineage priors for rookies

    Extrapolate a rookie's likely fit and a first-year coordinator's tendencies from their college background — before there's a single NFL snap to chart.

Coaching-era drift · BAL — OC change2016 → 2023
Roman eraMonken era · Δ
Shotgun
41%→ 78%
+37
Pistol
33%→ 7%
−26
Under-center
9%→ 31%
+22
Early-down pass
44%→ 58%
+14
One OC change, four structural shifts. The cube isolates the coordinator's signature from the roster and the head coach.
The Film Room

See it in the workspace.

Pick a staff from the rail, read their tendencies by situation, and switch the lens without leaving the screen.

GRIDIRONSCIENCE ⌕  Search staff, team, player… Share view

Kyle Shanahan · Coordinator DNA

SF 49ers · 2023 · Offense · 1,087 plays
All downs By quarter By personnel vs Front
Tendencies by situationEPA/play
Pass Run
1st & 10
PASS 52RUN 48
+.18
2nd & long
PASS 71RUN 29
+.09
2nd & short
PASS 38RUN 62
+.21
3rd & short
PASS 44RUN 56
+.05
3rd & long
PASS 89RUN 11
−.04
Red zone
PASS 47RUN 53
+.12
RENDER · sf_fingerprint.png SF Gameplan Fingerprint 2023
Gameplan fingerprint · 20 dimensions, league-normalized
38%
Under-center (above league)
31%
21-personnel (vs 8% avg)
+.46
EPA, early-down pass
+98
Identity distinctiveness
Generated from the cube · 500K+ plays · per-play personnel ● Live  nflverse + CFBD substrate
Look at what the cube sees

Generated from real plays — not mockups.

Each chart is produced directly from the cube in seconds. Real plays, real personnel, real coaching staffs.

RENDER · sf_fingerprint.pngSF Gameplan Fingerprint 2023
Teams · SF 2023

Gameplan Fingerprint

Shanahan's identity in one chart: 38% under-center, 31% 21-personnel, +0.46 EPA on early-down passes.

RENDER · bal_coaching_drift.pngBAL Coaching-Era Drift
Coaches · BAL 2016–23

Coaching-Era Drift

The Roman→Monken OC change in a single column. Pistol 33%→7%, shotgun to 78%.

RENDER · sf_star_players.pngSF Star Player Footprint
Players · SF

Irreplaceable Players

Personnel-conditioned EPA delta. Five players each shifting team EPA by +0.13 to +0.22.

RENDER · bal_defense.pngBAL Defensive Fingerprint
Defense · BAL 2023

Defensive Identity

Macdonald's match-quarters scheme: 39% Cover-1, −0.39 EPA allowed on 3rd-and-long.

RENDER · league_overview.pngLeague Overview Matrix
League · 32 teams 2023

Identity Distinctiveness

Every team's z-score deviation across 10 dimensions. Find the outliers in seconds.

RENDER · predictive_backtest.pngPredictive Backtest
Predictive · BAL 2023

Hold-out Backtest

BAL 2023 excluded from the build. Pass-rate predictions within 7 points. L2 error 28.0 — and dropping.

Validation, not assertion

Every metric is benchmarked. We publish what failed.

We publish the metrics that failed, too.

Separation didn't predict route grade. The in-sample betting edge was selection bias. We tested them, they didn't hold, and we say so. If a number doesn't track the standard, it doesn't ship.

Tracking metric vs PFFr — correlation
Interceptions
+.994
Sacks
+.947
Pass break-ups
+.931
Coverage (catch% allowed)
+.872
Kicker FG-over-expected
+.880
Our tracking metrics, correlated to PFF's charted equivalents.
The method

Sixteen years in the making.

The cube design converged in 2010. The data substrate finally caught up.

2010

The Cube

A multi-dimensional cube design in VBScript — 4,826 lines, three full iterations before the architecture converged. The technique worked; the substrate was the limit.

2026

The Substrate Caught Up

Per-play personnel, jersey numbers on the field, coverage shells, time-to-throw. CFBD for college tendencies and rookie backgrounds. A Dictionary-of-dictionaries became a Polars DataFrame on a 146 MB Parquet store.

Now

The Engine Ships

Descriptive renders for every team. A composition function that predicts coaching-hire outcomes from college lineage and roster archetypes. A Monte Carlo simulator on the roadmap.

Early access

Get early access to the gameplan engine.

We're validating the predictive engine across a growing set of coaching-hire backtests. Join the list and we'll notify you when the technical white paper drops and the descriptive cube opens for select teams.

We'll only email you about Gridiron Science releases — white paper, beta access, launches. Unsubscribe any time.