Forecast track record
How well past probability estimates matched what happened. Only forecasts published before their target date count; research backtests are kept on separate pages.
These results score synthetic fixture forecasts against synthetic outcomes. They show how the page works, not how any model has performed.
- Settled forecasts
- 8
- Unique targets
- 8 pair-dates, 2 dates
- Period
- 31 Aug 2026 – 30 Sep 2026
- Method version
- v1 · [version ID]
- Exclusions
- None. USD/EUR is not counted separately from EUR/USD.
8 settled forecasts are far too few to tell sources apart. Sources are listed in a fixed order and are not ranked. Uncertainty intervals appear from 30 settled forecasts.
Sources compared
| Source | Settled forecasts | Mean Brier (0–2) | Single results, lowest–highest | 95% interval |
|---|---|---|---|---|
| AI average uncalibrated | 8 | 0.741 | 0.494–0.975 | Not shown below 30 |
| OpenAI | 8 | 0.748 | 0.450–1.005 | Not shown below 30 |
| Claude | 8 | 0.741 | 0.441–0.950 | Not shown below 30 |
| Statistical reference Synthetic fixture | 8 | 0.725 | 0.588–0.869 | Not shown below 30 |
| Prediction markets | 0 | Not scored | — | No matching markets |
OpenAI and Claude see the same evidence, so their results are correlated and are not independent tests.
Reliability
When estimates say 30%, does the outcome happen about 30% of the time? Each point groups outcome probabilities into a bin and compares them with how often the outcome occurred, with its count and an uncertainty range.
Shown from 100 outcome probabilities per source. Currently 24 (8 forecasts × 3 outcomes).
Not enough settled outcomes yet
Detailed results
AI average, default threshold per pair family. Each row links to the original forecast page.
| Target date | Pair | Threshold | Lead time | Down / Within / Up | Outcome | Brier |
|---|---|---|---|---|---|---|
| 31 Aug 2026 | BTC/USD | ±10% | 31 days | 30% / 39% / 31% | Up | 0.719 |
| 31 Aug 2026 | ETH/USD | ±10% | 31 days | 33% / 31% / 36% | Down | 0.672 |
| 31 Aug 2026 | SOL/USD | ±10% | 31 days | 35% / 26% / 39% | Within | 0.832 |
| 31 Aug 2026 | EUR/USD | ±1% | 31 days | 30% / 27% / 43% | Up | 0.494 |
| 30 Sep 2026 | BTC/USD | ±10% | 61 days | 38% / 28% / 34% | Within | 0.776 |
| 30 Sep 2026 | ETH/USD | ±10% | 61 days | 39% / 23% / 38% | Within | 0.900 |
| 30 Sep 2026 | SOL/USD | ±10% | 61 days | 40% / 18% / 42% | Down | 0.563 |
| 30 Sep 2026 | EUR/USD | ±1% | 61 days | 36% / 20% / 44% | Within | 0.975 |
Mean of the 8 AI average scores: 0.741.
- One version per target is scored, chosen in advance by the documented rule; later updates of the same target are correlated with it and are not counted as extra evidence.
- Thresholds of the same forecast are correlated, so each threshold is reported separately; this table uses the default threshold only.
- USD/EUR settles on the same observation as EUR/USD and is not counted twice.
- 35 live forecasts are still pending; the earliest target is 31 Oct 2026.