Agent evaluation

loading /api/agent/deepseek… Signals → · VXN →

Total score

Identical resolved questions; every forecaster answers all of them
Forecast accuracy for every model on the same resolved questions
ForecasterBrierLogLossECE Directional hitn$/forecastp50 (s)Tool callsGate hits
The exam paper · every question, every answer
Actual: YES / NO historical outcome AI result: ✓ 3/3 right · ~ 1/3 mixed · ✗ 0/3 wrong 0.68 ✓ / 0.62 ✗ per-model call 0.68★ closest Brier
Every question in the bench, with each forecaster's probability, the actual outcome, directional correctness and who was closest