Your LLM red-team report should be reproducible, not a screenshot

Most "we red-teamed our model" posts I see are a screenshot of a bad output and a shrug. That is not...