Research path
Find a player
Start with a season, position group and workload threshold, then compare production within the same source category.
Open find a player →College football reading room / Field guideStart with the evidence
How to use the football desk
Learn the game
behind the box score.
A stat becomes useful when its denominator, source and question stay attached. This guide moves from raw college football records to a matchup preparation workflow without turning a small sample into a scouting verdict.
01 / Choose a starting point
One desk, several questions.
Research path
Read a matchup
Open the next slate to see the persisted score, margin, win probability, uncertainty and source clock together.
Open read a matchup →Research path
Study team efficiency
Compare play-weighted offensive and defensive rates with opponent filters and game-level evidence.
Open study team efficiency →Research path
Audit the source
Search every retained source row, including fields that are not yet mapped into a ranking or model feature.
Open audit the source →Research path
Read personnel context
Compare source-listed rosters, recruiting commitments, team talent and returning production before asking a matchup question.
Open read personnel context →02 / Metric dictionary
Translate the number into a decision.
Football concept
Expected points added (EPA)
What it is. EPA measures the change in expected scoring value created by a play relative to the game state before it.
How to use it. Use total EPA to describe accumulated production and EPA per play to add a workload context. Passing, rushing and receiving are separate categories and should not be added together.
Open the evidence →Football concept
Success rate
What it is. The share of recorded plays that produce a positive expected-points change under the source model.
How to use it. Pair it with EPA, yards and play count. A high rate on a tiny sample is a lead for review, not a complete player evaluation.
Open the evidence →Football concept
Yards per play
What it is. Recorded yards divided by the plays in one source category.
How to use it. Use it as a descriptive explosiveness measure. It does not account for field position, down, score state or opponent strength by itself.
Open the evidence →Football concept
Opponent-adjusted team efficiency
What it is. Silvermine's ridge model estimates team effects from retained game records while controlling for opponent and venue context.
How to use it. Read offensive and defensive rates with the number of games and plays behind them. These ratings describe the archive; the forecast model is a separate artifact.
Open the evidence →Football concept
Forecast margin and probability
What it is. The current model stores a projected score, home margin, total and calibrated home-win probability for each eligible upcoming game.
How to use it. Check the model ID, creation clock and interval before using the estimate. Probability is calibrated on a historical holdout and is not a promise about a future result.
Open the evidence →Football concept
Source coverage
What it is. Coverage is the set of records a release actually supplies, not an assumption that every roster, snap or statistic exists.
How to use it. Inspect source receipts, missing fields and excluded placeholders before drawing a personnel conclusion. Missing data stays missing.
Open the evidence →The football archive preserves publisher fields and source labels. A field that is absent from a release is unavailable; it is never silently replaced with zero or inferred from a player name.
03 / Build a matchup question
Read in a repeatable order.
- Confirm the game. Check the teams, scheduled kickoff, venue and whether the source marks the time as final.
- Read the forecast context. Keep the score, margin, probability and interval together with the model ID and creation clock.
- Find the unit lever. Use team efficiency and player production to choose a passing, rushing, receiving or defensive question for film.
- Audit the evidence. Open the source row and coverage desk before treating a missing player ID, event record or market quote as a conclusion.
04 / Keep the model honest
Evidence has a boundary.
The production forecast uses historical team records, venue and model calibration. It does not know a late injury, a depth-chart change, weather or an eligibility decision unless a future release adds and validates that evidence.
The player board ranks offensive categories by recorded EPA. The defensive and specialist notebook keeps name-attributed events separate when the publisher supplies no stable athlete ID.