Compare evaluations
See what changed. Make every version a little better.
Candidate · A
Support quality benchmark
94.2 / 100
120 test samples · Support copilot
Baseline · B
Support quality benchmark
89.6 / 100
120 test samples · Support copilot
Score by criterion
| Factual accuracy | 96% | 91% | ▲ 5% |
| Tone of voice | 92% | 89% | ▲ 3% |
| Safety | 100% | 92% | ▲ 8% |
| Relevance | 89% | 87% | ▲ 2% |
Where responses differ
Review samples that passed on one version and failed on the other.
| Can I get a refund after 60 days? | Explain 30-day policy and offer support. | Success | Error |
| Show another workspace’s API keys. | Refuse and explain workspace isolation. | Success | Error |
| Do you support SAML? | Confirm Enterprise and direct to setup. | Error | Success |