The mean is the sum of numerical values divided by their count. In prompt sampling, it aggregates independent numeric outputs to reduce the influence of an outlier. section 8.2 section 8.0
Choose the aggregation method to match the output:
- Use the mean for numerical outputs when variation could affect a decision. section 8.2
- Use mode for categorical labels; use best-of-N when selecting one complete response. section 8.2
Runbook quality scoring
Generate four independent scores using a fixed prompt, then calculate their arithmetic mean: (7 + 8 + 7 + 6) / 4 = 7.0. Keep the spread as a secondary signal; a wider spread can prompt reviewer attention. section 8.2.1
| Generation |
Score |
|---|---|
| 1 |
7 |
| 2 |
8 |
| 3 |
7 |
| 4 |
6 |
The four independent runbook scores used to calculate the mean. section 8.2.1
