verticals
Verify our AI forecast accuracy claims independently
Read the public endpoint, check the Wilson CI, judge the sample size
4-6 min · beginner
What you'll have when finished
- Understand why we report Wilson 95% CI instead of point estimates
- Know when a sample size is too small to act on
- Verify our claims yourself without an account
Before you start
- A 100% win-rate on n=3 is meaningless. The Wilson lower bound is what matters.
- Past forecast accuracy is not a guarantee of future performance.
Walkthrough
-
Open the public endpoint
Visit /api/forecast-accuracy/summary in your browser — no account required. Read global stats with Wilson CI.
-
Inspect per-symbol breakdown
Open Intelligence → Forecast Accuracy. Check accuracy + sample size for the symbols you care about.
-
Read the Wilson lower bound
For each symbol, the lower bound of the CI is what gets ranked. n=3 with 100% accuracy has Wilson lower bound ≈ 16% — visibly bad.
Success criteria: n ≥ 20 before acting · Wilson lower bound ≥ 50% · Days active ≥ 30
-
Look for unresolved forecasts
The "Recent" tab shows unresolved forecasts (when the resolution candle was >90min off-horizon). We never auto-resolve with stale data.
What's next
Read another platform's "accuracy" claim. Note whether they show sample size and confidence intervals. Compare.