Event Edge

Scorecard

Model forecasts begin when experiment v1 goes live

Live since July 2026: events from 1 July 2026 with forecasts made before the filing.

stats · all · n=544
Brier · exceeds typical
0.2448
Brier · up
0.2511
Log loss · up
0.6954
Pinball q10
0.0113
Pinball q50
0.0248
Pinball q90
0.0122
stats · post · n=238
Brier · exceeds typical
0.2411
Brier · up
0.2509
Log loss · up
0.6951
Pinball q10
0.0117
Pinball q50
0.0268
Pinball q90
0.0137
stats · pre · n=306
Brier · exceeds typical
0.2477
Brier · up
0.2512
Log loss · up
0.6956
Pinball q10
0.0110
Pinball q50
0.0233
Pinball q90
0.0109
Live window scores by role and session
rolesessionnBrier · exceeds typicalBrier · upLog loss · upPinball q10Pinball q50Pinball q90
statsall5440.24480.25110.69540.01130.02480.0122
statspost2380.24110.25090.69510.01170.02680.0137
statspre3060.24770.25120.69560.0110.02330.0109

Reliability of P(up) · stats

0.000.000.250.250.500.500.750.751.001.00forecast probabilityobserved frequency

Each point is a bin of forecast probabilities; its size is the number of events in the bin and its height the share that actually went up. Points on the dashed line mean the stated probabilities were right on average.

Backtest (model training period): all scored events, including those that overlap the models' training data and forecasts made after the fact.

stats · all · n=4120
Brier · exceeds typical
0.2375
Brier · up
0.2502
Log loss · up
0.6935
Pinball q10
0.0120
Pinball q50
0.0256
Pinball q90
0.0122
stats · post · n=1882
Brier · exceeds typical
0.2385
Brier · up
0.2504
Log loss · up
0.6940
Pinball q10
0.0126
Pinball q50
0.0277
Pinball q90
0.0136
stats · pre · n=2238
Brier · exceeds typical
0.2367
Brier · up
0.2500
Log loss · up
0.6931
Pinball q10
0.0115
Pinball q50
0.0238
Pinball q90
0.0111
Backtest scores by role and session
rolesessionnBrier · exceeds typicalBrier · upLog loss · upPinball q10Pinball q50Pinball q90
statsall41200.23750.25020.69350.0120.02560.0122
statspost18820.23850.25040.6940.01260.02770.0136
statspre22380.23670.250.69310.01150.02380.0111

Reliability of P(up) · stats

0.000.000.250.250.500.500.750.751.001.00forecast probabilityobserved frequency

Each point is a bin of forecast probabilities; its size is the number of events in the bin and its height the share that actually went up. Points on the dashed line mean the stated probabilities were right on average.