How to read the model

Model probability (model %)

How often the EdgeGoat model expects an outcome to happen.

The probability the EdgeGoat projection model assigns to one side of a line (over/under, a side, a total). It is read from the model's full predicted distribution for the stat — not from a single point estimate — and is calibrated against settled outcomes. Public pages show a display probability that is shrunk slightly toward the sportsbook's price when the two disagree strongly, so a single outlier never reads as a certainty.

Example: Model 58% on "over 24.5 points" means the model expects the over to land about 58 times in 100 comparable games.

Implied probability (book %)

The break-even win rate baked into the sportsbook price.

The win rate at which a bet at the quoted American odds breaks even. It includes the sportsbook's margin (the vig), so the implied probabilities of both sides of a market add up to more than 100%. EdgeGoat compares the model against this real, payable price — not a de-vigged fair price — because the real price is what a bettor actually gets.

Formula: Negative odds: |odds| ÷ (|odds| + 100). Positive odds: 100 ÷ (odds + 100).

Example: −110 → 52.4%; +150 → 40.0%.

Edge

Model probability minus implied probability.

The gap between how often the model expects an outcome and how often the price needs it to happen to break even. A positive edge at the real price is exactly a positive expected value; a negative edge means the price is worse than the model thinks the outcome deserves. Edge is a statement about a price, not a promise about a single result.

Formula: edge = model probability − implied probability

Example: Model 58% vs book 52.4% (−110) → edge +5.6 points.

Expected value (EV)

Average profit per unit staked if the model probability is right.

The average return per $1 staked over many repetitions, assuming the model probability is correct. EV above zero and edge above zero are the same statement at the real price.

Formula: EV = model probability × decimal odds − 1

Example: 58% at −110 (decimal 1.909) → EV ≈ +0.107, about 10.7 cents per dollar.

Correlation-aware parlay probability

Parlay hit probability that accounts for legs moving together.

The probability that every leg of a parlay wins, computed jointly instead of by multiplying the legs. Legs from the same game are correlated — a quarterback over on passing yards and his receiver over on receiving yards tend to win together — so EdgeGoat joins the legs with a copula over the model distributions. Multiplying correlated legs as if they were independent misstates the true hit rate, which is also how same-game-parlay prices are set.

Projection

The model's predicted distribution for a player or game stat.

A full predicted distribution (mean, spread and shape) for a stat such as points, passing yards, strikeouts or a game total, built from recent game logs, usage and minutes, opponent and pace context, rest and venue. Every probability on the site is read from this distribution against a specific line.

Graded hit rate

How often settled model edges actually won.

Candidates are frozen shortly before each game starts and graded after the final by the same settlement logic that grades user parlays. The graded hit rate is the share of those frozen candidates that won (pushes and voids excluded). It measures the model's record on the prices it flagged, not a promise about future games.