Methodology
How we know what we know
Every finding is labelled by where it came from, and every number shows how many plays and games stand behind it.
Each engagement
Scope
Your staff names the opponent, the phases that matter most and the questions you want answered.
Build
The engine assembles the evidence, charts any film you send, applies the sample-size rules and labels every claim. A person reviews it before delivery.
Deliver
The package reaches your staff ahead of game week, with the clips to watch numbered to your cut-ups.
Refine
Your feedback on what was useful shapes the next package.
Three evidence classes
Every finding carries one label. A finding takes the lowest label of anything it depends on.
- Observed
Observed
Read directly from recorded play-by-play, or counted from it by a fixed rule.
“Went for it on 4th-and-1 or 2 in 7 of 12 chances, 8 games.”
- Estimate
Estimated
Estimated by a documented model, such as the expected points of a starting field position. Always labelled as an estimate.
“About 0.5 expected points for 10 yards of field position near midfield.”
- Film read
Read from film
Read from film your staff authorizes, tied to the official play and graded high, medium or low confidence. Anything a call depends on is flagged for your staff to confirm.
Required for anything about alignment, assignment, coverage or technique.
Sample sizes, every time
The number of plays behind a claim is printed next to it. Always.
Count and games
Every number shows its count and the games behind it, for example “7 of 12, 8 games”.
15 plays, 3 games
A rate or a ranking needs at least 15 plays across at least 3 games.
Every play when it’s small
Anything smaller is listed play by play: every kick, every return, every 4th down, with no percentage and no ranking.
Outliers named
When one long play drives an average, we show the number without it, or drop the claim.
Penalties apart
Plays wiped out by a penalty are counted separately, not mixed into averages.
Provenance and review
Where every number comes from, and who checks it.
- DATA
Recorded play-by-play
Every FBS game this season, from CollegeFootballData.com, whose terms allow this use. Each report names its sources and the games it covers.
- FILM
Read by the engine, graded
Film your staff authorizes is read at the snap and each read is graded high, medium or low confidence. Results come from the official play, not the frame.
- MODEL
Estimates are labelled
Expected points and other model outputs are marked as estimates every time they appear, with the model and the drives it was fitted on named.
- REVIEW
Checked before delivery
Every package is reviewed by a person before it leaves us. Anything a call depends on that the frames can’t show clearly is flagged for your staff to confirm.
What we don’t claim
- No promises about accuracy, time saved or wins. We show the evidence and how much of it there is.
- No reads of every player on every play. The engine reports what the frames show and grades how clearly.
- No called plays or intentions. A play-by-play rush is reported as a rush, not as a called run.
Data standards
Permitted sources
Play-by-play comes from CollegeFootballData.com, whose terms allow this use. Film comes only from the program that owns it. Every report names its sources.
Confidential by default
Your staff’s questions, film and feedback stay with your program. We don’t share them, and we don’t name the programs we work with.
Nothing invented
If the data can’t support a claim, we don’t make it. Alignments, coverages and assignments are stated only when confirmed on film.