Free Engineering Workbench calculator
AI output safety rates
Track unsafe outputs found in reviewed samples, with an upper bound when you find none.
AI evaluationHow it's worked out
- Rate = found ÷ reviewed. Upper bound = Wilson 95%, or 3 ÷ n when none found
- Random sample of production outputs
Note“Up to” is the 95% upper bound. With zero found in n samples, the rule of three gives 3 ÷ n, so 2,000 clean samples still allow up to 0.15%.
Source: Hanley and Lippman-Hand (1983) rule of three; OWASP Top 10 for LLM Applications
Need help acting on this number?
VVnT SeQuor's AI Testing & Evaluations practice helps teams turn this metric into a plan.