The Loop, Part 4: The Teaser Engine — Buying Points, Not Predictions

There is a second engine in the loop, and it breaks the loop's own rule. It never predicts who wins. It does not care who wins. It cares about one number — six — and two landmarks on the football field's geography, three and seven. This post is the one engine in the machine that is not forecasting, and it may be the most interesting one, because it bets on the price instead of the game.

What a teaser is

A teaser is a parlay of point-spread bets with a twist: you get to move each spread six points in your favor, and in exchange the payout drops hard. A normal two-team football parlay pays about 2.6 to 1. A two-team six-point teaser pays about 1.91 to 1 — you are buying points, and points have a price.

Buying points sounds like a consumer bargain. Usually it is a bad one, because the points you buy are priced fairly or worse. Unless the six points you buy cross a key number — a margin where football games actually finish. And that is the whole engine: it does not find better teams, it finds badly priced geography.

The geography: why 3 and 7

Football scores come in chunks. A field goal is three points; a touchdown with the extra point is seven. Decades of NFL results pile up at exactly those margins — games end decided by three or by seven far more often than by four, five, or six. A spread that sits just off a key number is a different product from a spread that sits just past it, and the market knows it, which is why spreads cluster there too.

The teaser engine, built on the key-number analysis from Stanford Wong's Sharp Sports Betting — the book that named this edge — hunts exactly two windows:

An underdog at +2.5 loses if it falls by a field goal — the most common margin in the sport. Teased six points to +8.5, that same loss wins. Both key numbers crossed with one purchase. A favorite at −8.5 wins by a touchdown and still loses its bet; teased to −2.5, it wins with a field goal to spare. The six points are worth the most, by a wide margin, in exactly these two windows. Teasing a team from +6.5 to +12.5 buys points past where games end — you paid for land nobody lives on.

The arithmetic of the price

The engine assumes the standard two-team payout of 1.91 decimal (breakeven 72.4% per leg, since both legs must win). Legs that cross both key numbers have historically landed in the mid-70s — the classic Wong analysis put them around 74–76% — so the margin between the historical rate and the breakeven is the entire profit, a couple of points wide. Thin. Real. Not built on knowing anything about the Jets.

This is also where the honesty goes on the record: the assumed payout is the engine's biggest open assumption, and it is flagged as such. Every breakeven on every teaser card derives from that one number, and until the actual local book's two-team teaser price is confirmed at the window, the cards carry the assumption in the open. A bet whose profitability depends on one unconfirmed price is exactly the kind of thing this series promises to tell you rather than bury. If the real payout is meaningfully worse than 1.91, the windows still exist but the margin shrinks or vanishes — the ledger will show it either way.

What the engine does each tick

Every tick it filters the day's scraped spreads through the two windows, pairs legs from the same sport into up to three cards, and writes the cards to a daily file. NFL legs grade automatically against final margins. CFL legs are a known gap, stated plainly: the league's public API publishes no per-game scores, so a teased spread cannot be graded yet, and CFL legs sit flagged pending_scores_source until a scores source exists. It will be a useful measurement in itself: CFL football clusters less at 3 and 7 — the rouge and the two-point convert break the NFL's geography — so the ledger should end up quantifying how much weaker the Canadian tease is, and we would rather measure that than assume it.

Injury flags ride on the legs as context, not modeling: a team missing its starting quarterback or carrying three doubtful-or-worse players gets a warning marker on the card, and nothing else. Flags argue; they do not adjust. Same rule as the moneyline card.

Why it earns its place

The loop's thesis about small sportsbook edges is that most of them are noise until calibration proves otherwise — a cold model's three-point edge on a moneyline is usually the model flattering itself. The teaser engine is the escape hatch from that problem, because it leans on a different assumption: not we know more than the market about the game, but the market's spread is efficient, and the teaser price was set without enough respect for the key numbers it crosses. The first assumption a young model fails. The second is a pricing error you can hunt while the model grows up.

Next: what happens when the same discipline — rules written down, then mechanically kept — is pointed at a trading desk instead of a sportsbook. Part 5 — The Gates: /articles/the-loop-part-5-the-gates

Cross the numbers, not the teams. And confirm the price before you count the profit.