The CFB Key Numbers: Why 3 and 7 Matter More Than You Think (and Why We Cut the Rest)

Analysis · by Rick's Picks Analytics

If you've ever heard a sharp bettor talk about "buying through 3" or "the line is stuck on 7," they're talking about key numbers -- the integer margins where football games actually finish. Get the spread right relative to those numbers and you're betting smart. Get them wrong and you might as well flip a coin.

We pulled the empirical margin distribution from 5,649 FBS-vs-FBS games (2015-2022 train) and the picture is dramatic.

The Top 10 Margins in College Football

From most common to least:

(One bookkeeping note: our data also clips every margin of 28 or more into a single "28+" blowout bucket, which covers about 22% of games. It shows up later in this post, but it isn't a margin games "land on," so it's excluded from this table.)

A 3-point game happens almost 10% of the time. A 7-point game almost 9%. Together, those two margins account for ~18% of all FBS games. That's why the books fight so hard to set the line ON those numbers -- when the spread sits at -3, the game has the highest probability of pushing, which is great for the book.

Train vs Test Stability

We validated the distribution on 2,313 unseen test games (2023-2025), including the full 2025 season as fresh holdout. The top-10 margins overlap 90% between train and test -- and 90% between train and 2025 alone. A 3-point game happened 10.7% of the time in test and 11.4% in 2025; if anything, the spike got sharper. That's exactly the kind of stability you want before treating a finding as a real, exploitable feature of how college football is played.

Compare that to most "edges" we test -- a hot-streaks angle that shows 55% in train collapses to 50% in test. Key numbers don't collapse because they're not a market quirk; they're a property of how points are scored (touchdowns, field goals, two-point conversions, safeties).

We Trimmed Our Own List

Here's the part most pick services would never publish. Our original key-number set was 3, 7, 14, 17, 21, 24, and 28 -- the classic "touchdown arithmetic" list every sharp bettor recites. When we re-validated against the completed 2025 season, we made ourselves re-test every number individually, with Bonferroni correction for multiple comparisons. The list didn't survive intact:

• 3 and 7 are bulletproof. Both pass Bonferroni at absurd significance (p on the order of 10^-95 and 10^-68) with the largest effect sizes in the distribution (Cohen's h = 0.26 and 0.22), and both replicate at full strength in 2025.
• 14 is real but marginal. It passes Bonferroni (p = 0.0066), but the effect size is tiny (h = 0.048). It earns a spot at reduced weight, nothing more.
• 17, 21, and 24 are artifacts. None of them ever passed Bonferroni in train (corrected p = 1.0). Worse: in 2025-only data, margins of 21 and 24 occurred BELOW the uniform baseline (ratios of 0.75 and 0.83). They looked like key numbers because they're multiples of 7 and the eye wants to see them -- but the games don't land there any more often than chance.
• 28 isn't a key number at all in our data -- it's the clipped "28 or more" blowout bucket. It matters for one specific thing (see the half-point math below), but it's not a line-sitting number.

So the engine's list shrank from seven numbers to two-and-a-half: 3, 7, and 14 at half weight. Cutting your own list is uncomfortable. It's also the whole point of the methodology -- the same rules that reject other people's folklore have to reject ours.

What This Means for Betting Spread

The Walters principle: a half-point bought through a key number is worth more than a half-point bought elsewhere. Specifically:

• Moving a line from -3 to -3.5 is valuable -- you're now safe from the most common margin (3) and won't push
• Moving from -3.5 to -4 is near-useless -- you've crossed no key number
• Moving from -6.5 to -7 is costly -- the line is now stuck on 7, the second most common margin

Our key-number coefficients table comes from this exact empirical distribution. When Rick's Picks evaluates a game, the algorithm looks at where the closing line sits and adjusts pick confidence accordingly:

• Line sits exactly on 3 or 7 (or 14, at half weight): -5 confidence (the spread is "sticky" and small edges won't move it)
• Line sits within 0.5 of one of those numbers: +3 confidence (the line is "unstable" and live betting could flip you onto the right side)

This isn't a big-money edge by itself -- it's a context modifier that prevents us from saying "high confidence!" on picks where the line is structurally hard to beat.

The Half-Point Buy Math

Some books offer half-point buys for additional juice (usually -120 instead of -110). For an integer k, the expected value of buying through it is:

EV = P(|margin| = k) × 1.00 - (1 - P(|margin| = k)) × 0.20

For the EV to be positive, P(|margin| = k) must exceed 16.7%. In our CFB data, only margin 28 cleared that bar at -120 juice -- and remember, our "28" is really the clipped "28 or more" blowout bucket (about 22% of train games, 19% in test), which is why it clears. Not 3. Not 7. This held again in the 2025 holdout. Buying half-points across CFB key numbers is generally NOT profitable at -120, despite what intuition might suggest. The books know the distribution and price it correctly.

This is one of those "well-known sharp angles" that mostly doesn't work in practice -- another methodology win, even when the answer is "no."

The Honest Takeaway

Key numbers are real -- but there are fewer of them than the folklore says. The CFB margin distribution is stable across years, 3 and 7 dominate it, and the rest of the classic list didn't survive honest testing. Knowing the key numbers doesn't print money -- it just gives you better context for how confident to be on a pick, and it correctly tells you when buying through a key number is and isn't worth the juice.

That's how Rick's Picks uses it: as a quiet, transparent confidence modifier you can see in the breakdown of every pick.

Rick's Picks publishes statistical analysis of college football for informational and entertainment purposes. Nothing here is betting advice. 21+. If gambling is affecting you or someone you know, call 1-800-GAMBLER.