Calculations run on this device. Scenario values are not sent anywhere — the engine is fully local.
ai calculator · free
Eval Confidence Interval Lab
Put binomial confidence intervals on leaderboard deltas before believing them.
Is 84.2% on 100 items actually better than 82%?
Engine 1.0.0 · calculation
0100
105000
0100
105000
- Point gap
- 2.0 pts
- Intervals overlap?
- Yes — the gap is not proven
- Method
- Wilson score intervalz=1.96
- Verdict
- Get more items before believing the delta
Method
- Wilson score intervals at z=1.96 for each model's accuracy, then a simple overlap test between intervals.
- The point: a 2-point lead on 100 items is usually inside the noise — size the eval before trusting the delta.