How VEMScore works

For every game we publish our own win probability next to what the sportsbook market says, lock it before the game starts, and grade it in public.

Probabilities, not picks. We do not tell you what to bet. We show how likely we think each outcome is, how that compares with the market, and how our forecasts have actually done.

The market number

Sportsbooks put a price on every outcome, and those prices add up to more than 100% because the book keeps a cut, called the vig. We collect prices from many books, take the vig out, and combine them into one consensus probability per outcome. That is the market number. It is our yardstick: it is hard to beat, and we say so.

The VEMScore number

Our number is built from the final scores of past games. Every team has an attack rating and a defence rating that move after each game, with bigger moves early in a season and smaller ones later. The main ingredients are those team ratings, rest, and home advantage. The model never sees sportsbook prices, so when it disagrees with the market, that is its own view and not a copy.

A model needs a few games before its numbers mean anything. Until each team has played 6 games, we show the market line only and say the model is warming up.

Each day the models are re-checked on finished games. A change only goes live if it beats the current setting on games it has not seen, and every change goes in the change log.

Spread and over/under

For most games we also show how likely a team is to cover the spread and how likely the total is to go over. Those numbers have been less reliable than the win probabilities so far, so they are shown at half strength, pulled halfway toward 50%. We will replace that with a fitted correction once a league has about 100 graded lines.

Locked before the game, graded after

Each forecast is locked before the game starts and cannot be edited or removed afterward. When the game ends, it is graded against the final score, win or lose, and it stays on the public Record page.

The grade is a score called log loss. For each game it measures how much probability a forecast gave to what actually happened, so a forecast that was confident and wrong is punished hard. Lower is better. For a two-team game, saying 50/50 every time scores 0.693. We score the market the same way, so the two are directly comparable.

When we say who is ahead

A few games prove nothing. We show percentages once a league has at least 5 graded games for the comparison, and we say who is ahead only after 100 graded games in that league. Before that the Record page says it is too early to say. The home page shows how many forecasts each league has so far.

What this is not

It is not betting advice, and nothing here is a recommendation to place a wager.
It does not promise to beat the sportsbooks. Beating the market consistently is very hard, and our early results are too small to say either way. The value is an honest, public comparison.
Back-tests and candidate models are never shown as public performance. Only locked forecasts are graded on the Record page.

Questions or corrections

Email support@vemscore.com. If we got something wrong, we would like to know, and we will fix it in the open.