Track Record
Every call the model made, graded — and shown next to the trivial rule it was competing with.
A win rate on its own says nothing. Backing the market favourite in every game wins most of them; taking the underdog on a run line covers more than half by itself. So every record here sits next to the simplest rule that would have made the same kind of call for free, and the comparison is the number that matters. On most of these markets our model has not beaten it.
| Sport | Moneyline | Spread / line | Total | Accuracy |
|---|---|---|---|---|
| College Football → | 259-71 78.5%-4.8 vs rule | 169-189 47.2%-6.4 vs rule | — | — |
| NFL → | 34-15 69.4%+6.1 vs rule | 23-26 46.9%-6.1 vs rule | — | — |
| MLB → | 311-236 56.9%-4.2 vs rule | 315-242 56.6%-0.2 vs rule | 275-268-23 50.6%-1.3 vs rule | 3.47 runsmarket 3.39 |
| NHLreturns October | — | |||
| NBAreturns October | — | |||
| College Basketballreturns November | — | |||
“vs rule” is how far the model is above or below the best trivial rule for that market — backing the favourite, taking the underdog, or taking the under, whichever of them did better over the same games. Each market covers a slightly different set of games, because each needs a different pre-game price to have existed; the per-sport pages show every denominator. These are regular-season games. Postseason results are counted separately and shown on each sport’s own page, because averaging the two would misstate both.
Retired betting products
1233-1126
52.3% · 2359 settled picks
These were betting picks: two products that named a side and a price, and were scored win or lose. That is a different thing from the records above, which measure what the models projected. This one is kept because it is a real record of what was published — but it is not a measure of the models running today.
Best Bets + Model Likes
1170-1075
still settling occasionally — last 2026-10-03
+EV v2
63-51
still settling occasionally — last 2026-10-03
What it was made of
How these numbers are calculated
- Every market separately. The moneyline, the spread and the total are different claims with different records. A combined figure would describe none of them.
- Every record against a baseline. We compare each one to the best trivial rule for that market, not the most flattering. Chosen badly, a baseline turns a systematic lean into an apparent edge — which is exactly what happened when we first measured the MLB run line against the wrong rule.
- Each market, its own denominator. A moneyline call needs a pre-game price, a spread call needs a line and a total call needs a number. The same slate does not produce all three, so the game counts differ and each is shown with its own.
- Where a market is missing. Our football model projects a margin and not a score, so it makes no call on the total. That slot says so rather than disappearing.
- What's included. Games where enough sportsbooks posted a line for us to publish against — which leaves thinly-priced games under-represented in the college sports and barely matters in the NFL. For MLB: July 2026 onward only — earlier projections were rewritten in a single pass on 6 July and are not what the site showed at the time, and any projection amended once a game had started is excluded. Each sport's page states its own window.
- No stakes, no returns. This is a record of what the model said and what happened. It is not a betting result and there are no money figures on it.