vanta Accuracy
Resolved-question track record by category — vanta vs the market's implied forecast. Lower Brier is better calibration.
Real-market backtest — Polymarket & Kalshi
5,420 resolved markets · T−7d snapshot · polymarket 5,420frozen run · computed 2026-08-10 · coverage 5% of resolved corpus
Brier (lower better)
0.1320
market 0.1326
Log score (lower better)
0.3966
market 0.3967
Directional accuracy
80%
market 80%
No-skill benchmark
0.1940
always-predict-26%
Leakage-free: for each market the pipeline saw only the venue price 7 days before close plus category base rates learned from other events; the market is scored on the identical snapshot. Expect vanta ≈ market — beating deep prediction markets is hard, and this scorecard reports whatever the data says.
Synthetic demo corpus below — exercises the UI, claims no edge by construction
Resolved questions
42
10 live · view archive
vanta accuracy
83%
market 79%
vanta Brier
0.128
market 0.128 · log 0.41 vs 0.40
Avg live edge
±8%
mean |vanta − market|
Directional accuracy by category
Calibration — observed vs predicted
| Category | Resolved | vanta acc. | Market acc. | vanta Brier | Market Brier |
|---|---|---|---|---|---|
| science | 6 | 100%▲vanta leads the market here | 67% | 0.145 | 0.165 |
| sports | 5 | 100%▲vanta leads the market here | 100% | 0.058 | 0.056 |
| technology | 10 | 80%▲vanta leads the market here | 80% | 0.128 | 0.115 |
| politics | 5 | 80%▲vanta leads the market here | 80% | 0.105 | 0.123 |
| finance | 12 | 75%▲vanta leads the market here | 75% | 0.153 | 0.144 |
| crypto | 4 | 75%▲vanta leads the market here | 75% | 0.137 | 0.151 |
Demo track record: seeded, deterministic data illustrating the intended edge. In production this table is written by the resolution pipeline as questions settle.