The 4M model room

SHOW
THE WORK.

PIPELINE · 4M-CFBD-0.3

A useful model should be explainable, versioned and judged against results it did not already know. CollegeFootballData supplies the evidence layer; the calculation and editorial judgment remain ours.

The starting position

A prior that can move—but cannot panic.

The published 4M 138 remains the preseason baseline until the CFBD edition clears its first audit. Every refresh creates a new database snapshot; no historical edition is silently overwritten.

35% · base4M preseason football prior

The existing neutral-field rating prevents a single noisy opener from producing a fake transformation.

40% · ratingsCORE, SRS, SP+, FPI and Elo

Multiple opponent-adjusted and predictive views reduce dependence on one outside model.

25% · rosterTalent, recruiting, returners and portal impact

Player usage and position value distinguish proven transfers from raw transaction volume.

Game evidenceUp to 70% as the sample matures

CORE, adjusted EPA, success rate, explosiveness and special teams earn more influence as 2026 snaps accumulate.

Schedule layerGame-by-game win probabilities

Neutral-field rating differences, a 2.4-point home field and completed results produce expected regular-season wins.

Audit layerImmutable, timestamped snapshots

Inputs, coverage, model version and output are stored together so every public claim can be reconstructed.

The complete model

Data gets the first word—not the last.

The automated layer prices repeatable football. Availability, coaching, role changes and weather remain explicit editorial adjustments until they can be measured reliably. The provisional weights will be challenged through rolling historical backtests before they become a public betting engine.

Opponent-adjusted efficiency

EPA and success rate by pass/run, explosiveness, early downs, passing downs, finishing drives and havoc—adjusted for opponent quality.

Roster prior

Returning production, recruiting talent, transfer value, quarterback continuity and coaching-system continuity. Its influence decays as 2026 snaps accumulate.

Availability

Quarterback status, official injury reporting, depth-chart movement and high-value absences at offensive line, pass rush and coverage positions.

Game environment

Home field, travel, rest, altitude, pace, expected possessions and stadium-specific weather near kickoff.

Special teams

Field position, place-kicking, punting and return value—small samples stabilized toward the national mean.

Market calibration

Consensus open and close used as a benchmark and limited calibration feature, while preserving an independent football number.

Daily operating rhythm

Update the data daily. Move the rating only when the evidence earns it.

A timestamped edition keeps the public board honest and prevents silent revisions after a result.

Return to the 4M 138 ↗

05:00 ET
Reconcile
Final scores, play-by-play, advanced box scores, participation and corrected game records.
06:00 ET
Refresh
Opponent-adjusted team metrics, roster transactions, injuries, depth charts and schedule data.
Monday
Re-estimate
Run the weekly model after the full slate is final; preserve the previous version for an audit trail.
Game day
Price the matchup
Refresh market consensus and weather, then issue a final number after availability reports.

The public scorecard

Accuracy has more than one definition.

Spread MAE

Average absolute error between projected and actual scoring margin.

Calibration

Whether a stated 60% outcome occurs near 60% over a meaningful sample.

Log loss

Rewards accurate probabilities and punishes unjustified confidence.

Closing-line value

Whether published positions consistently beat the eventual consensus close.

ATS result

Tracked at the exact posted number and price, never a later favorable line.

Version stability

Every weekly release archived so results cannot be rewritten after kickoff.

Data stackCollegeFootballData Tier 3 ↗The Odds API ↗NOAA weather ↗SportsDataIO injuries ↗

SHOWTHE WORK.