Track record · measured from every signal we’ve fired
We grade ourselves in public
Since June 22, every signal is recorded the second it fires, then measured against what the market did 5 minutes, 4 hours, and one settlement later. All bands published, including the bad ones.
The AI Score’s live report card
Show the model two past alerts — one where the market moved ≥3¢ within an hour, one where it didn’t. This is how often it scored the mover higher, on recent signals it never saw in training.
retrained on a rolling window · currently 40,049 graded signals · last test Sep 3, 2026 · 50% would be guessing
This number updates itself on a schedule, whichever way it goes. If the model gets worse, you’ll see it here before we do anything about it.
What holds up
Sports markets that fire a signal move 2.3× more in the next hour than their own typical active hour — 13.0¢ vs 5.6¢.
n=7,318 signals vs 34,460 baseline market-hours
Odds a market moves ≥3¢ within an hour: 51% after a signal, 27% in a normal active hour.
same populations as above
The same movement lift in elections markets.
n=170 signals · smaller sample, watch this one
What doesn’t — and we say so
How often price keeps moving the signal’s way over the next 4 hours. That’s a coin flip: our signals locate action, they do not call direction.
n=4,343 matured directional signals
Per-contract result of buying the signal’s side and holding to settlement (rushes / whale prints). The market’s price already knew.
n=4,377 signals on settled markets, vs entry-implied odds
Six-figure trades settled below the odds they paid — following big money to settlement lost 7¢ per contract at meaningful odds. Big money is not smart money.
n=119 settled whale prints at 15–85¢
Methodology
Every emitted signal above our quality floor is snapshotted with the market’s price at that second (yes-equivalent cents). An analyzer measures the move at 5/15/30/60/240-minute horizons; settlement results come from Kalshi’s public API. “Baseline” is the same category’s average active hour (≥8 trades) with no signal requirement. Markets within 10¢ of settlement are excluded as mechanical noise.
Rule we hold ourselves to: signals are “associated with” movement — we will never say “predicts.” If a number on this page stops holding as n grows, the number changes, not the honesty.
SEE THEM LIVE ON THE BIG MONEY TRACKER