First public release.
- Load local JSONL or Hugging Face datasets
- Evaluate offline predictions, Python callables, and HTTP guardrail endpoints
- Report TP, FP, TN, FN, false-positive rate, pair accuracy, and per-family slices
- Export complete Promptfoo tests and an HTTP provider
- Generate privacy-minimized JSON, Markdown, and JSONL reports
- Validate with 38 tests, 89% coverage, type checks, security scans, and Python 3.10/3.12/3.14 CI