How accurate is the Oracle?
As of September 4, 2026, the BTC Event Oracle's 1-week directional Brier score is 0.253 versus a random-walk baseline of 0.250, over N=1307 resolved forecasts.
Short-horizon Bitcoin is close to a random walk, so the honest, expected result is to roughly tie a naive “no change” benchmark — and we publish exactly that, including our misses. These figures update daily; the live dashboard shows up-to-the-hour numbers and the full calibration charts.
The record, by horizon
| Horizon | N | Brier | vs RW | CRPS skill | Coverage |
|---|---|---|---|---|---|
| 1 Week | 1307 | 0.253 | -0.01 | +0.17 | 71% / 60% |
| 1 Month | 702 | 0.255 | -0.02 | +0.15 | 74% / 60% |
| 1 Year | insufficient data — fills in as calls mature | ||||
Brier scores the up/down call (lower is better; 0.25 = a coin-flip). “vs RW” and “CRPS skill” are skill scores against the random walk — near zero is the honest, expected result at short horizons. Coverage is how often the real price landed inside the stated range vs how often it should have.
Dig deeper
- · How the forecast is made & graded
- · Weekly recap — 2026-W36 (42% of calls correct)
- · Weekly recap — 2026-W35 (30% of calls correct)
- · Weekly recap — 2026-W34 (17% of calls correct)
- · Weekly recap — 2026-W33 (75% of calls correct)
- · Weekly recap — 2026-W32 (28% of calls correct)
- · Weekly recap — 2026-W31 (60% of calls correct)