The Film Room for modern football analysis
Interactive, ground-truth NFL scouting and game-planning — built on public tracking data and validated against the industry's gold-standard grades.
What it is
Gridiron Science is an NFL analytics engine and interactive product — a modern Python rebuild of a cube-based game-plan-inference engine first validated in 2010, now running on real, current data. It turns raw play-by-play and player-tracking into scout-grade context: who a team is, how a coordinator calls plays, which matchups break a game open, and how a season projects.
Founded and built by a former player (safety) who designed the original engine — football intuition first, statistics in service of it.
How it works
- OLAP cube over every play (~1M feature slices) + a layered predictive engine: rolling team strength & strength-of-schedule, coordinator play-call tendencies, coverage DNA, injury impact, game-script, a Monte-Carlo drive simulator, and a full-season projector (win totals, playoff odds).
- Data lane: public NFL play-by-play (nflverse), NFL Big Data Bowl player-tracking frames, and PFF grades as a validation benchmark — every number traceable to its source.
- Built in Python (polars / DuckDB); deployed as a live web product plus a read-only REST + LLM (MCP) API so the numbers can power chat assistants and downstream tools.
Validated, not hand-wavy
Player ratings are derived independently from tracking geometry, then benchmarked against PFF's same-stat grades. Representative agreement (Pearson r):
+0.99
Defender interceptions
+0.87
Coverage catch-rate allowed
+0.64
Receiver contested-catch
And when something doesn't work, we say so. Raw separation turned out not to proxy route-running craft — we reported the negative result rather than dress it up. That discipline is the product's backbone.
The product — The Film Room
- Formation Gameplan ★ — formation-vs-front "wars," offense tendencies + defensive response by down/distance, drawn as a matchup picture
- Matchup & Gameplan — A-vs-B breakdowns, run-gap tendencies, pass-zone heatmaps
- Scheme DNA — per-coordinator formation tendencies, coverage mix, blitz/pressure identity
- Player Ratings — validated tracking-vs-PFF axes with per-game trends & situational splits
- Head Coaches — game-management, environment "pull" vs league, round-adjusted rookie development
- Game Logs — weekly box scores + smoothed rate trends
- Season Projector — division standings, playoff odds, and a scouting verdict per team
- API + plugin — query the engine from any LLM assistant
★ Flagship — the Formation Gameplan. Pick any offense vs any defense and the formation war is drawn as a picture — the offensive formation facing the defensive front across the line — with each formation's tendencies (pass/run, direction, depth, ball-carriers), the defense's response (front mix, coverage, blitz/man), a per-formation matchup EDGE, and a player-tracking blitz-direction overlay (who rushes, from where). It's the game-plan cut-down the founder built in 2010 — now live on NFL data.
Honest positioning
This is a calibration- and scouting-grade engine — context, projection, and player evaluation — not a betting tipster. A fully held-out backtest put it at coin-flip against closing lines, and we publish that rather than hide it. The value is rigorous, transparent football insight that holds up to scrutiny.
Where it's going
The per-matchup Formation War now ships with a tracking-fed blitz-direction overlay. Next: multi-season formation data, in-season convergence as live games accumulate, and deeper coordinator scheme modeling — each shipped only after it validates.