Public track record

Every call, logged and scored.

We timestamp every AI prediction with the price at the time and the full reasoning, then score it against the resolved market. No cherry-picking, no retroactive edits, no hidden losses. Download the CSV and audit the whole book line by line.

Predictions resolved

27 pending · 47 total

Win rate

25.0%

on calls @ any confidence

ROI (flat stake)

-41.3%

P/L: -8.25 units

Brier score

0.1239

lower is better · 0.25 = coin flip

Cumulative PnL

1-unit flat stake per prediction

Total

-8.25u

Calibration curve

Perfect predictions sit on the diagonal · dot size = sample count

Brier

0.124

Dots above the diagonal = AI overconfident; below = underconfident. We rebalance the ensemble weights weekly based on this curve.

Volume by confidence bucket

Live

Biggest wins

5 logged

The Quiet Engine
Best Picture
+3.28u
18.3% @ 4.28
Republicans
Winning Party
+1.24u
45.7% @ 2.24
Labour
Most Seats
+1.03u
55.0% @ 2.03
Labor
Winning Party
+0.65u
56.0% @ 1.65
Hold
Rate Move
+0.55u
46.7% @ 1.55

Recent losses

Every miss on the record

Reform UK
Most Seats
-1.00u
8.0% @ 5.57
Lib Dem
Most Seats
-1.00u
3.7% @ 49.81
Conservative
Most Seats
-1.00u
30.0% @ 3.54
Cut 25bp
Rate Move
-1.00u
43.0% @ 2.81
Hike
Rate Move
-1.00u
3.7% @ 50.00

We publish every loss. Transparent track records have to include the bad calls.

Transparency, not tips. Proppa Bet is market intelligence, not a betting service. Every prediction is versioned with the model and prompt strategy used, and the full audit trail is one CSV download away. We don't retro-edit history.