Every forecast is scored once its time comes, including the misses. A 90% range should hold about 9 times out of 10, so roughly one in ten is expected to miss.
v2 (HAR-RV-J) has been forecasting live since 28 Oct 2026, 13:00 UTC (13 days). Live results need about two weeks before they mean much; the out-of-sample backtest is the longer test.
Prices: Binance BTC/USDT. The fallback to CoinGecko hasn't been used since live forecasting began.
Last 7 days: 89.3% held (±4.5, 2,016 forecasts, all live). 216 misses in the period shown (216 live). Within ±4.5 of 90% is consistent with the target. Forecasts overlap, so a single day is close to one independent observation at 24h; 7-day windows are more meaningful.
| In | Scored | Hit rate | Interval score |
|---|---|---|---|
| 1 hour | 3,781 | 90.4% | 2.821 |
| 4 hours | 3,745 | 87.9% | 6.603 |
| 24 hours | 3,505 | 79.4% | 17.205 |
| In | Scored | Hit rate | Interval score |
|---|---|---|---|
| 1 hour | 3,813 | 89.6% | 2.977 |
| 4 hours | 3,777 | 88.5% | 6.502 |
| 24 hours | 3,537 | 75.2% | 18.058 |
| In | Scored | Hit rate | Interval score |
|---|---|---|---|
| 1 hour | 225 | 88.4% | 2.813 |
| 4 hours | 189 | 98.9% | 4.703 |
| 24 hours | 0 | – | – |
One email when the new model has two weeks of live forecasts: how often the ranges held, the misses, and what changed. Nothing else.
Sign-up is handled by MailerLite, which asks you to confirm and lets you unsubscribe at any time. Your address is used only for this report and is never stored on this site.