TELEGRAPHED

We call the play before the snap.

Pass or run, predicted from what can be known before the ball moves — then checked against what actually happened, on seasons the model never trained on.

New every Tuesday
How predictable is your offense?
Every NFL offense ranked by how often we call its plays before the snap.
See all 32 rankings →

One real drive

Chance of passChance of runCalled correctly

Right more often than the best public model

—
Calls correct, all plays
—
Public benchmark (nflfastR), same plays
—
Of the benchmark's misses, the share we get right instead

Every season, out of sample

Each season is predicted by a model trained only on the seasons before it.

TelegraphedPublic benchmark

Where the calls are hardest

Neutral game state: 1st and 2nd down, neither team running away with it.

TelegraphedPublic benchmark

When it's sure, it's almost never wrong

Hit rate by confidence

Bar length is the share of all plays that fall in each band.

Probabilities you can price

When it says 70% pass, teams pass about 70% of the time.

2026, as it happens

Telegraphed, livePublic benchmark

Every offense has a tell

How often each offense's play call was predicted correctly in 2025, against the public benchmark.

TelegraphedPublic benchmark

How it's tested

Every number on this page is out of sample. For each season, the model is trained only on earlier seasons, then asked to call every snap of the new one. Nothing about a play is used unless it is knowable before the snap.

Pass includes dropbacks that end in a sack or scramble. Kneel-downs, spikes and two-point tries are left out. The public benchmark is nflfastR's expected pass model, the standard open-source reference.

The live 2026 model runs on data available during the season; the historical model uses additional pre-snap data that is published after each season.