Scout’s record
An AI that talks about markets should have to show its own graded record. This is Scout’s: one call a day, graded against what actually happened, losses kept.
Last graded call: Oct 13
Scout is Ledger’s AI companion. Every day he publishes one call on a real market, and every call is graded against the real result: won, lost, or push. The verdict is stored permanently, losses included. Scout never promises a win. This page exists so anyone can check, at any time, how an AI that talks about markets actually does.
Today’s call
No call is on the record for today yet. Scout makes one call a day, chosen once on the server; it appears here the moment it is made.
The record so far
A Brier score is the average squared distance between a stated probability and what happened, so calling more does not improve it and calling only the obvious does not either. Scout’s number is the market’s own price with the margin removed, on the side Scout took, which means this measures whether those sides land at the rate the fair line said they would. It is not a claim that Scout beats the market. Players who state a confidence on their own calls are scored the same way, on their public record page.
Calibration by confidence band
Each bar shows Scout’s actual win rate on calls where the market implied a probability in that range. Scout’s daily call always takes the underdog, so every call sits under 50%. The tick mark shows the average price the market set on those calls. A bar above the tick means they landed more often than priced. This is the market’s own number, not Scout’s claim. The bars use only Scout’s latest 20 calls, the ones this page loads. The hit rate and Brier score above count every decided call.
| Band | Calls | Landed | Expected |
|---|---|---|---|
| Under 30% | 0 | too few | 25% |
| 30-36% | 0 | too few | 33% |
| 36-42% | 3 | 67% | 41% |
| 42-50% | 11 | 45% | 45% |
Call history
20 calls shown. A call is on the record from the moment it is made, not from the moment it wins. Every figure on this page is read live from the public endpoint at /api/scout-call.
Why does an AI publish its own record?
Scout talks about markets all day inside Ledger. Anyone who does that should have to show a graded record, and almost nobody who does it anywhere else has to. A tipster account can screenshot its wins and delete its losses. Scout cannot: the call is written down before the game, graded against the real result, and the verdict is stored permanently, wins and losses in the same list. The losses are the point. A record with the losses removed is an ad.
Does Scout give picks?
One a day, and you see it without asking. The Call of the Day sits on Home, where you can take the same side as Scout or the other one. Beyond that call, Scout never promises a win and never manufactures urgency to make you play. Ask about a game and Scout explains both sides and the price. It never names a side or a size. The Call of the Day is Scout's own call, not a recommendation: it costs no Float, a miss never touches your Standing, and a landed call pays a small Standing bonus. It is chosen to be interesting rather than safe, which often means the longer price, because calling the favorite is mostly reading the board back to you.
This is not advice. Nothing on Ledger involves real money, and Scout's call is not a pick to carry anywhere that does. The record exists to keep Scout honest, not to make you confident.
How is the call chosen and graded?
Once a day, the server picks one market and writes the call down, keyed to the Eastern date. Every visitor sees the same call, and it cannot be re-rolled after the fact. When the game ends, the same stored row is graded by the same settlement that grades every player's call on Ledger: real finals, a verdict of won, lost, or push, and pending until reality has actually reported. The grading is documented in how Ledger settles predictions.
How should you read the record?
Slowly. One call a day is a small sample for a long time, and a hot or cold week means very little on its own. Pushes are counted but never scored, the hit rate is wins over decided calls only, and pending simply means the game has not finished or the grade has not landed. We would rather show you a short record with the losses in it than a long one without them.