Twelve seasons of game lines. Two seasons of every player prop posted. Every angle the internet swears by, graded at real prices. Almost all of it dies. Three things don't.
Every August the same folklore comes back: home dogs, revenge games, bye-week rest, wind unders, fade-the-public. Some of it comes from real studies. Most of it comes from someone's memorable Saturday. We put all of it — the academic findings, the tout-sheet systems, the barstool rules — through one harness: real historical prices, flat one-unit stakes, results split by season so nothing gets to hide inside one lucky year.
The headline is uncomfortable for anyone selling picks: the college game-line market is close to unbeatable, and the prop market is beatable in exactly three places. Here is the full accounting.
Break-even at standard −110 juice is 52.38%. Here is how the famous angles actually did across 12,150 games from 2014–2025:
| Angle (as claimed on the internet) | n | Hit | ROI | Seasons + |
|---|---|---|---|---|
| Under, all week-1 games | 910 | 55.5% | +5.9% | 7/11 |
| Fade a team off a 38+ pt spread aberration (VSiN system) | 352 | 54.6% | +4.1% | 4/10 |
| Road dog, total ≤45, conference game | 542 | 54.1% | +3.2% | 4/8 |
| Road dog, total ≤50 ("covers 53%") | 2,134 | 52.4% | +0.0% | 8/12 |
| Back team off a bye vs normal rest | 2,097 | 50.8% | −3.0% | 7/12 |
| All home dogs | 4,203 | 50.2% | −4.1% | 1/12 |
| Revenge: back the team that lost last meeting | 6,664 | 51.5% | −1.7% | 4/11 |
| Under, wind ≥13 mph ("strong under") | 331 | 49.9% | −4.8% | 0/2 |
| Fade 3-game ATS cover streaks (overreaction) | — | 51.0% | −2.7% | 4/12 |
| Back all home teams | 12,150 | 48.1% | −5.9% | 0/12 |
Read the road-dog row twice, because it is the most instructive line in the table. The internet claim — road dogs in totals of 50 or less cover 53% — is true. We measured 52.4% over 2,134 games, positive in 8 of 12 seasons. And it returns exactly +0.0%, because 52.4% is what break-even looks like. Most famous angles are like this: real patterns priced to the penny. The claim survives; the profit doesn't.
The blowout-aberration fade — the one genuine tout-sheet system that graded out positive — confirmed VSiN's published direction (they claim teams off a 38-point spread miss go 44.2% next game; fading them, we measured 54.6%). But 4 of 10 seasons positive means it has losing years, and 352 games over twelve seasons is roughly one bet a week in season.
Week-1 unders is the one we didn't find on a tout sheet: openers went under 55.5% across 910 games, positive in 7 of 11 seasons. The mechanism is plausible — new offenses are installed before new defenses are exploited, and totals are set with no current-season data. We'd call it promising rather than proven: after testing dozens of angles, one is expected to look this good by luck alone.
Wind over 13 mph is an automatic under. Home dogs bark. Revenge is a motivator. Rest wins.
Wind unders hit 49.9% — below a coin flip. Home dogs were profitable in one season out of twelve. Revenge teams and anti-revenge teams both lose. Teams off a bye cover 50.8%, which is a loss after juice. The market prices all of it.
Two popular systems we could not test, and we'd rather say so than fake it: betting-splits fades (fading 75%+ public sides) require handle data that no public archive keeps, and turnover-differential systems need box-score joins we haven't built. Everything else on the standard lists is in the table above or died quietly beneath it.
Player props are a different animal. Books post fewer of them, staff them more thinly, and set them off season-long usage in a sport where usage changes in a fortnight. We bought every closing line posted on college players in 2024 and 2025 — 25,357 graded — and tested at best-available prices across DraftKings, FanDuel, BetMGM and ESPN Bet.
Three prop findings cleared the vig and replicated in both seasons:
| Strategy | n | Hit | ROI | 2024 | 2025 |
|---|---|---|---|---|---|
| Fade the new QB's own props after a QB change | 160 | 73.1% | +32.0% | +45.8% | +14.2% |
| Fade RBs after a QB change | 147 | 66.0% | +22.7% | +20.8% | +27.2% |
| Fade all props after a QB change | 550 | 65.1% | +14.4% | +15.5% | +12.3% |
| Under, all pass-TD props (no filter at all) | 1,452 | 56.7% | +5.3% | +4.5% | +6.1% |
| Under, high pass-yard lines (top quartile) | 291 | 56.0% | +5.6% | +3.3% | +7.8% |
| Under, low rush-yard lines (bottom quartile) | 491 | 57.2% | +7.8% | +2.3% | +13.3% |
Why the quarterback change works: a book prices a receiver at 62.5 receiving yards off a season of usage with his starter. The starter gets hurt or benched. Three things collapse at once — the backup delivers the ball worse, the offense turns conservative, and the new quarterback has different favorite reads. The book's model still sees the season average, so the number barely moves. Everyone attached to that offense is now over-lined, and the new quarterback himself is over-lined worst of all.
Why the blanket unders work: recreational bettors take overs — you bet a player to do something, not to sit on the bench. Books shade lines up to balance that action, and in college the shade is heavy enough that an entire market, passing touchdowns, goes under 56.7% of the time with no filter whatsoever. The posted line is not a forecast. It is a price that clears action.
And the honest half: the famous stuff failed here too. Favorite-longshot bias is enormous in the anytime-TD market (the biggest single signal we measured) and still unbettable — backing longshots returned −16%, because the mispricing is smaller than the hold on those prices. Thin markets, single-book games, MACtion, late-night kicks: none of it was softer. We tested obscurity six ways and it lost six ways.
The most useful thing the prop market gives a fantasy manager is not a bet at all. At any point in the season, take the market's number on a player and compare it to what he has actually produced. That gap prices his rest of season:
| Market vs his production | n | ROS change (PPG) | 2024 | 2025 |
|---|---|---|---|---|
| Market far below production (q1) | 1,668 | −3.30 | −3.03 | −3.55 |
| q2 | 1,667 | −1.03 | −0.98 | −1.10 |
| q3 | 1,668 | −0.15 | −0.08 | −0.20 |
| q4 | 1,667 | +0.75 | +0.79 | +0.71 |
| Market at or above production (q5) | 1,667 | +1.70 | +1.85 | +1.54 |
Perfectly monotonic, nearly identical in both seasons, and it holds at every position. This is not just mean reversion dressed up — controlling for the player's own scoring level, the market's view still adds real predictive signal (+0.18 partial correlation). When a player is outscoring his market number by a wide margin, the market is telling you it doesn't believe the level, and it is usually right: those players decline by 3.3 points per game the rest of the way. Sell high on q1. Buy q5.
Spreads and totals: efficient. Twelve straight losing seasons for blind home teams is what a priced market looks like, and the folklore angles are either wrong (wind, revenge, rest) or true-and-worthless (road dogs in low totals, at exactly break-even). The prop market is where inattention lives: a quarterback change is worth 14–32% fading everything attached to it, whole markets lean under from public over-shading, and the market's player-level number reprices rest-of-season production more honestly than the box score does.
Data: CollegeFootballData betting lines and box scores 2014–2025 (12,150 games with lines and scores); The Odds API historical closing player props 2024–2025 (25,357 graded at DraftKings, FanDuel, BetMGM, ESPN Bet). Spreads and totals graded at −110; props at best available posted price. Weather verdicts rest on the ~3,000 games with joined station data, concentrated in recent seasons. All strategies reported per season; nothing here is betting advice.