Reports

A live example is published from the deterministic sample (npm run sample-report).

HTML: behavtest report <run> [--against <base>] --out report.html writes one self-contained file: no network access, no external assets, opens from file://, light and dark themes, filter and search, and per-case drill-down with inputs, outputs, scores, judge reasoning and traces. Pipeline outputs are untrusted text and are only ever inserted as text, never HTML.
Markdown: behavtest run --md summary.md and behavtest compare --md compare.md write GitHub-flavoured summaries, ready for a CI job summary (cat summary.md >> "$GITHUB_STEP_SUMMARY") or a PR comment.
JSON: --json on run and compare, for scripts and dashboards.
Console: behavtest runs lists saved runs; behavtest show <run> [case] prints a run, or one case's input, outputs, scores and trace.