Gridiron · Game Center · 2026
The Splits Room
Where a team actually lives — the down-and-distance grid, the field, the scoreboard and the twelve situations, on every scrimmage snap rather than on dropbacks alone.
11,299 scrimmage plays from the first 7 weeks of the 2026 regular season, first through third down, read from both sidelines. 355 of them are quarterback scrambles, which the feed files as runs; they are counted here as the dropbacks they were. 256 fourth-down snaps are set aside beside 1,083 punts and kicks, because a fourth down that was run is the fifth a coach liked. 43 plays carry no down and are dropped rather than bucketed.
2026 is 7 weeks old
Only 7 of 18 regular-season weeks have been played, so the floors move with them: a team is ranked at 78 scrimmage snaps rather than 200, and a cell prints a rate at 10 rather than 25. 32 of the 32 teams that have taken a snap clear it. Everything here is 7 weeks of football, and the grid fills as the weeks land.
Every scrimmage play of the regular season, first through third down. Snaps this team ran. Expected points are its own, so higher is better.
Where a team actually lives — the down-and-distance grid, the field, the scoreboard and the twelve situations, on every scrimmage snap rather than on dropbacks alone. Acell under 25 snaps is shown but not ranked — it prints its number in grey, carries no colour and does not set the scale, because a rate off a handful of snaps is a story about the handful. This early in the season the WITHHOLDING floor moves with the weeks played, so only a cell under 10 snaps refuses a number outright and prints an em-dash.
Painted after the last game's play-by-play landed · the room repaints as each game's play-by-play lands, usually within two hours of the final, or the next morning when its release runs late
The ranking, and the control
All thirty-two on every snap
Ranked on expected points a play inside every snap, best first on this sideline. The bar is the departure from the LEAGUE's own line for the same cut, drawn from a true zero — not a rank, which has no zero and would paint a median team as a good one. Pick a row and every object below is redrawn for that team. Higher is better here.
A row under 25 snaps is shown but not ranked: it keeps its numerals and its place, loses the verdict colour, and does not set the scale the other rows are drawn against. A row with too little on file to print a rate at all — under 10 snaps in this cut — shows an em-dash and sinks to the bottom rather than ranking as a zero. The dropback share rides every row in numerals because it explains the number beside it and is never itself a verdict.
The signature object
SF — the down-and-distance grid
Twelve cells: three downs by four distance bands. The painted number is the DEPARTURE from the league's own line in the SAME cell — third and eleven is a losing down for everybody, so a raw rate says that twelve times and tells you nothing about the team standing in it. The raw rate rides underneath, and the snap count is in every cell. FOURTH DOWN IS DELIBERATELY ABSENT — only about a fifth of fourth downs are scrimmage plays and they are the ones a coach chose to run, so aggregating them prints a selection, not an execution. That question belongs to the Fourth Down Room, which asks it the right way round.
Thin sample: 4 qualified of 12 · 3 of 12 cells hold fewer than 10 snaps and print an em-dash rather than a rate. 5 of 12 cells are under 25 snaps, so they print their number in grey and are neither ranked nor coloured
A scramble is filed upstream as a run and is counted here as the dropback it was — 355 of them this season. Partitioned the naive way the designed run game gains about four hundredths of an expected point a carry it did not earn.
The axis the Drive Room takes out
SF by field position
Six bands from a team's own goal line to the opponent's. The Drive Room removes field position so possessions are comparable; this puts it back, because who you are inside the ten is its own question.
- Own 1-10+0.12−0.14 lg10 · 40.0% db
- Own 11-40+0.07−0.02 lg113 · 61.1% db
- Own 41 to opp 41+0.27−0.01 lg81 · 56.8% db
- Opp 40-21+0.15+0.02 lg65 · 53.8% db
- Opp 20-11+0.47+0.04 lg31 · 54.8% db
- Opp 10 in+0.35+0.08 lg32 · 56.3% db
The scoreboard
SF by score state
Seven states in this offense's own frame. Watch the dropback share rather than the value: that is where a play sheet stops being a preference and starts being an obligation.
- Down 15+—0.00 lg0 snaps
- Down 9-14—+0.06 lg0 snaps
- Down 1-8+0.15−0.01 lg56 · 55.4% db
- Tied+0.14−0.03 lg86 · 64.0% db
- Up 1-8+0.33−0.02 lg92 · 58.7% db
- Up 9-14+0.18+0.01 lg56 · 58.9% db
- Up 15++0.13+0.10 lg42 · 38.1% db
Against the league's own line for the same cut
SF — the situation bank
Twelve overlapping cuts. A snap belongs to every situation it satisfies, so these DO NOT add up to the season and nothing here ever sums them. Verdict: strongest on 3rd & 7+ at +0.68 against the league, weakest on 3rd & 1-3 at +0.03.
- Every snap+0.20+0.20 lg332 · 56.9% db
- Red zone+0.41+0.35 lg63 · 55.6% db
- Goal to go+0.45+0.42 lg26 · 50.0% db
- 3rd & 7++0.59+0.68 lg20 · 85.0% db
- 3rd & 1-3+0.07+0.03 lg28 · 50.0% db
- Two-minute+0.09+0.08 lg40 · 70.0% db
- No huddle——6 snaps
- Shotgun+0.28+0.27 lg157 · 80.3% db
- Under centre+0.13+0.15 lg175 · 36.0% db
- Backed up+0.20+0.22 lg40 · 52.5% db
- Trailing 9+——0 snaps
- Leading 9++0.16+0.12 lg98 · 50.0% db
Goal to go is not a subset of the red zone: the line to gain IS the goal line, which after a penalty can be the fourteen. Both are here and neither is derived from the other.