ATSAgainst the spread
—— win rate •— pushes
Week 8 · Canonical HOLD — Tuesday and Wednesday source checks are available. View game readiness, current context and lock windows →
FBS MODEL • WEEKLY COMMAND CENTER
Research onlyProject the game.
Audit the context.
Independent college football projections for score, spread, total and win probability—paired with market comparison, input quality and the context that can change a matchup.
Projection first
Context visible
Validation enforced
Average model total—Projected combined points
Average favorite margin—Absolute model spread
Highest win probability—Across loaded games
Market references—Games with comparison data
WEEKLY BOARD
Projection board
Scan the full slate, compare the independent model with a current public market reference, and inspect data readiness before drawing conclusions. The publication-time market snapshot is saved separately for later grading.
ModelIndependent projection
MarketCurrent comparison; publication snapshot is locked for grading
ValidationResearch and publication state stay separate
All projections are informational and are not official wager recommendations.
| Game | Model score | Model spread | Market spread | Spread difference | Model total | Market total | Win probability | Implied ML | Data | Actions |
|---|
No games match those filters.
TRANSPARENT PERFORMANCE
Projection performance
Tracks model projection positions against the market spread and total saved when the projection was published. ATS, Totals and Straight Up are graded separately. This is not a wager ledger.
SUStraight up
—— win rate
TrackedProjection results
—legacy fallback rows
TREND
ATS cumulative performance
WEEKLY
Weekly record
| Week | ATS | ATS % | SU | SU % |
|---|
BREAKDOWN
Conference audit
| Conference | ATS | ATS % | SU | SU % | Games |
|---|
EVIDENCE CENTER
Research hub
Availability, weather, market movement, efficiency, tempo, roster and situational context—organized around what can move a projection.
MODEL PIPELINE
How the projection is built
The process is designed to separate statistical projection, contextual review, market comparison and validation.
Build team strength
Team efficiency begins with EPA, success rate, explosiveness, havoc, scoring and drive-level production, with recency weighted deliberately.
Adjust for opponents
Production is normalized for schedule and opponent quality so identical box-score output is not treated equally across very different competition.
Project the matchup
Offense and defense are paired with pace, drive volume and matchup style to estimate team points, home spread and game total.
Add game context
Availability, depth, weather, travel, rest and venue conditions are audited separately. Material uncertainty can keep a matchup in review.
Compare with market
The independent projection is compared with a current public market reference. The difference creates a research direction; the publication-time market snapshot is saved separately for grading.
Validate prospectively
Spread and totals systems must earn promotion through out-of-sample results. A validation HOLD cannot be bypassed by one attractive number.
WHY THE GATES MATTER
Accuracy before action.
College football changes quickly. Injury information can be incomplete, depth roles can move, market prices can change, and historical relationships can weaken out of sample. The site therefore keeps projection, market difference, input quality and publication state separate.
READ THE BOARD
Glossary
Fast definitions for the numbers shown throughout the site.
Spread
A point handicap. A favorite has a negative number; an underdog has a positive number.
ATS
Against the spread. The model/market difference identifies the projection side, then the final score is graded against the market spread saved when the projection was published.
Moneyline
The projection board may show a model-implied moneyline. The record tracker also saves the publication-time market moneyline for reference, but Straight-Up W/L is graded only on whether the projected winner won.
Total
The model total is compared with the market total to create an Over or Under projection. The record is graded against the publication-time market total.
Push
A tie against the saved market spread or total. Pushes are separated from wins and losses.
Win probability
The model's estimated chance that a team wins outright.
Model / market difference
The numerical difference between the independent model and the current market reference. It identifies a projection direction; it is not automatically expected value or a wager instruction.
Data quality
A completeness/readiness measure for required inputs. High coverage does not guarantee predictive accuracy.
SUPPORT THE PROJECT
Keep the research free and public.
Support is optional. Every projection, record update and research page remains free regardless.
ABOUT CB MODEL PROJECTIONS
A transparent front end for the model.
CB Model Projections exists to make weekly college football projections easier to inspect. It puts the independent number, market comparison, research context, data quality and validation state in the same place without blurring the difference between them.
PROJECT SOURCE
Inspect the build
Questions, feedback or bugs can be reviewed through the public repository.
Open GitHub repository ↗