Find the mismatch.
Build the edge.
Compare the trenches, expose situational weaknesses, adjust your weights, and test your matchup ideas with repeatable research workflows.
Historical opponent filters
Only previous games are filtered. The matchup being predicted is never filtered by its outcome.
The same rules apply to advanced rushing/passing, trenches, third downs, possessions, penalties, turnovers, SOS, and opponent adjustments. Filtering can reduce the available sample.
FBS game slate
Select a game to load its teams and retrieve available season-to-date statistics.
Saturday Slate Analyzer
See moneyline-style win frequency, both sides of the spread, and over/under percentages beside available market lines. Analyze every scheduled FBS matchup using season-to-date CFBD metrics through the prior week, emphasizing rushing lanes and the trenches. Model v1.4 accounts for expected run/pass play frequency before rating success-rate and trench mismatches; v1.3 remains available as a historical baseline. The inherited v1.0 margin error scale remains provisional; v1.4 home-field changes are not recalibrated. Totals point means remain independent of sportsbook lines. None of these probabilities demonstrate betting profitability.
No slate analyses run yet. All opponent-strength data and advanced statistics end in the previous week. v1.4 home-field, box-score sack rates, play-call exposure, trench adjustments and opponent normalization are research-only and not validated. Injuries and verified odds timestamps remain unavailable. Odds never enter forecasts.
Model Backtesting Lab
Replay completed FBS games using previous-week inputs. v1.0's underlying coefficients were trained on 2025 weeks 5–6, so eligible evaluation starts at week 7 of 2025, or in 2026. Archived lines are for grading only. No red-zone metrics.
v1.7+ restores v1.7 drive-based scoring and applies small additional corrections for penalties and expected turnovers calculated from completed prior FBS games. Backtesting compares v1.7+ directly against the original v1.7 engine on identical pregame snapshots. Market lines never change the forecast. Missing new statistics mean no extra adjustment.
Defaults to weeks 7–8 (later than the calibration training window). Backtests now request up to 8 additional previously completed box-score weeks (cached individually) for genuine sack/dropback rates. Repeated seasons may consume your free CFBD quota; run short periods first. Weights are taken from the Matchup Lab when you start.
Choose the season and weeks, then run your first backtest.
How confident predictions actually performed
| Model winner chance | Games | Avg model chance | Actual winner hit rate |
|---|---|---|---|
| Run a backtest to build the calibration table. | |||
Completed games and picks
| Week | Matchup | Final | Predicted winner | Winner grade | ATS selection | ATS grade | Total selection | Total grade |
|---|---|---|---|---|---|---|---|---|
| No historical games have been evaluated. | ||||||||
Matchup analysis lab
Run a customized, 10,000-trial Monte Carlo demonstration.
Matchup builder
Name the teams, then compare each offense with the opposing defense.
Enter season-to-date rates from the same source and period for both teams. Injuries and power ratings below are discretionary point adjustments, not automatically verified facts.
Model weights
Balanced categories · tempo separated from efficiency.
Original eight v1.7 categories preserve their relative shares; penalties (4%) and turnovers (6%) are small scoring corrections. Ten sliders total 100%. Tempo remains separate; it is not a scoring-efficiency weight. These coefficients are experimental.
Classic box-score sandbox weights (separate model)
These 5 sliders apply only to the manual box-score sandbox. They do not change full-slate calculations.
10,000 simulated game scores · adjustable offensive / defensive metrics · spread and total comparison · local saving.
Demo numbers are hypothetical. For a real matchup, select the game and fill in verified season statistics before running the model.
Simulation dashboard
Your analysis results will appear here.
Run the sample scenario to see projected scores, win frequencies, distribution of game margins, and the biggest contributing matchups.
Matchup advantage breakdown
Advanced matchup indicators: green favors Team A (away), pink favors Team B (home). These do not sum to the projected spread; drive efficiency and discipline also influence team scores.
Simulated scoring-margin distribution
Positive margins favor Team A. Based on the illustrative model.
Saved matchups
Stored locally in this browser — not synchronized to an account.
Model methodology
Transparent inputs, explicit assumptions, and room to test improvements.
Offense versus defensive holes
Run blocking and pass-front disruption face the opposing defense: line yards, stuff rates, front-seven havoc allowed/created, and sack rate per dropback where available.
Your preferred philosophy
Relative v1.7 weights preserved: rush 13.5%, pass 13.5%, run block 10.8%, pass front 10.8%, drives 12.6%, PPA 10.8%, third downs 9%, opponent strength 9%; plus penalties 4% and turnovers 6%. Tempo is modeled as possessions, not another efficiency percentage. The manual box-score sandbox keeps separate weights.
Opponent-aware, experimental
The advanced slate adjusts run blocking (line yards, stuff rate) and pass-front disruption (front-seven havoc and sack rates when available) against prior opponents, with separate run-blocking and pass-front categories; tempo is estimated outside the scoring-category percentage sliders. These balanced weights have NOT been learned from historical outcomes. Drive scoring remains experimental. The manual-input simulator is separate. This is not evidence of a betting edge; independent future-season validation is still required.