About This Model
How to read what's on this site.
Reading a game tile
A model line is not a pick. Only games with a bold pick line cleared the confidence bar. Every other line is raw opinion — context, not a recommendation.
A big model/market gap means real value or a wrong model — and on a non-flagged game, "wrong" is the safer assumption. Most disagreements aren't edge; that's what the filter is for.
Team scores are just (total ± margin) / 2, split from the two numbers the model actually predicts. Roughly a ±9-point error bar each — don't read precision into them.
Stake sizes
A unit is whatever you decide it is. A common convention is 1 unit = 1% of your bankroll, which puts about 10% in play in a typical week. The 25-unit ceiling is enforced: a heavy Saturday gets every stake scaled down proportionally to fit — it binds on 3 of 71 backtested weeks, the heaviest of which would otherwise have run 36.5u.
On spreads, size is the line. A spread pick is staked 3u when the line sits between 2 and 8 points and 1u when it's bigger. Across five backtested years the 3u tier hit 62.3% against the 1u tier's 54.9% — a 7-point gap with a ±6.0 margin on it, so read it as a lean, not a law.
On totals, size is confidence — and confidence doesn't predict hit rate. Totals are still sized 1–3u from the model's own probability, and there is no measurable relationship between that size and how often they win. Treat every number here as a starting point to size down from.
Backtested performance (2021–2025, out-of-sample)
| Season | Spread bets | Win % | ROI | Total bets | Win % | ROI |
|---|---|---|---|---|---|---|
| 2021 | 65 | 52.3% | −0.5% | 3 | 66.7% | +36.4% |
| 2022 | 84 | 57.8% | +13.4% | 63 | 59.7% | +21.5% |
| 2023 | 51 | 72.0% | +41.0% | 6 | 50.0% | +14.5% |
| 2024 | 43 | 53.5% | +1.7% | 58 | 58.6% | +13.3% |
| 2025 | 26 | 53.8% | +14.5% | 28 | 67.9% | +33.2% |
| Pooled | 269 | 58.1% | +14.4% | 158 | 60.5% | +20.7% |
Walk-forward: each season was graded by a model fit only on seasons before it. Watch the bet counts — a season with single digits says nothing, and the pooled figure rests on 427 bets, not thousands. This is backtest, not live results.
What drives the numbers
Three models: one predicts margin, one predicts total, and a third decides spread bets directly from "did the home team cover," with the market's line as an input.
| Factor | Score | Total | Spread bet |
|---|---|---|---|
| Team strength (Elo, talent, returning production) | 71% | 7% | 19% |
| Per-play efficiency (opponent-adjusted) | 9% | 47% | 13% |
| Trench play (run blocking / run stopping) | 1% | 12% | 37% |
| Luck regression (close games, fumble bounces) | — | — | 17% |
| Schedule & timing (rest, travel, kickoff slot) | — | — | 9% |
| Pace & explosiveness | — | 25% | — |
| Market signals (moneyline / spread) | 19% | 4% | 6% |
| Other power rating (SP+) | — | 5% | — |
| Home field / altitude | <1% | <1% | <1% |
The betting model leans on trench play and luck regression, not Elo or market odds — a book already prices those efficiently, so agreeing with them finds no edge.
Badges
Context only. No badge changes any number or filters any pick — the model never sees them.
🔄 QB change A starting QB changed within the last 2 games.
⛈️ Severe weather Severe wind/rain/snow forecast. The market prices forecasts faster than this model does, so treat a big gap here with extra skepticism.
⚔️ Rivalry One of ~45 hand-picked matchups. For fun.
🔥 Hostile environment A 7pm ET or later kickoff at one of ~25 venues with an atmosphere reputation, never neutral-site.
💤 Body clock The away team is crossing 2+ time zones.
★7 Rank on our own power ratings: the margin the model projects against an average FBS team on a neutral field, recomputed weekly. Subtract two teams and you get the spread it would project between them. Full top 25.
Game detail pages
Efficiency, success rate and explosiveness drive the predictions. The explosive-run split, pass-success split and havoc rate are shown for interest only. Ranks are out of every FBS team with data that season; trench and havoc splits show "—" until real current-season plays exist, usually week 2 or 3. Last 5 results are graded against each game's real closing line after the fact.
Bet responsibly
We are all here to have fun. There is no get-rich-quick scheme. Betting on sports should always be viewed as recreational and should be fun. Never wager more than you are willing to lose.
If you are going to bet on sports, have a plan. Create a budget, track your wins and losses, and know when it's time to take a break.
Let's have a great season and remember: sports betting should be strictly for fun, win or lose. If it becomes more than that, please get the help you need from the resources available.
Where to get help
- National Problem Gambling Helpline — call or text 1-800-GAMBLER. Free, confidential, available 24/7.
- Helpline chat — ncpgambling.org/chat.
- National Council on Problem Gambling — ncpgambling.org, including a self-assessment and a state-by-state directory of treatment.
- Gamblers Anonymous — gamblersanonymous.org, for in-person and online meetings.
Most sportsbooks also offer deposit limits, time-outs and self-exclusion in their own account settings. They are worth setting up before you need them.