Track Record
Every pick the model has made this season, scored against the final. The predictions are recomputed from the frozen algorithm exactly as they stood pregame — the same numbers the slate showed — so there is nothing to curate and nowhere to hide. † marks pitcher-adjusted calls.
Season 2026, through 2026-07-09 · back to the picks
$100 at real market prices
The season back-test, on honest terms: simulated bankrolls, real prices. Every paper position is bought at the actual traded Kalshi price at first pitch and settled by the real result — no model prices, no assumptions. The strategies trade only where the model genuinely disagrees with the market, sized quarter-Kelly with a 10% cap, constants chosen before any market data was seen. Coverage: 905 of 1,387 games (65%).
In aggregate the market's prices are sharper than the model — which is exactly why the strategies bet only the biggest disagreements instead of every game. Exchange fees are not modeled; thin early markets can move between the priced snapshot and first pitch. Hover the chart for any day’s balances.
The Stake
real moneyday 1On July 8, 2026 we put $100 of our own cash in a Kalshi account and handed it to the strategy: it buys the published strong signals automatically, and everything below is read from the exchange — we can’t retouch it.
Strong signals only, limit orders at the ask, quarter-Kelly with hard caps — the full rules, every fill, and the paper benchmark live on The Stake.
By month, by confidence tier
| Month | All picks | Lean | Solid | Strong |
|---|---|---|---|---|
| July | 60.0%63–42 | 46%13–15 | 69%25–11 | 61%25–16 |
| June | 57.0%225–170 | 49%53–56 | 63%55–32 | 59%117–82 |
| May | 52.7%221–198 | 50%58–59 | 52%54–50 | 55%109–89 |
| April | 51.5%202–190 | 48%114–123 | 58%72–53 | 53%16–14 |
| March | 60.5%46–30 | 61%23–15 | 55%16–13 | 78%7–2 |
Tiers: lean = 50–56% stated confidence, solid = 56–62%, strong = 62%+.