Project: eval-harness · Branch: feat/html-report · Mode: default · Model: gpt-5-codex

Summary

Generated a self-contained HTML report per eval run: summary table, per-metric distributions, a diff view against the previous run, and drill-down into failing cases with judge rationales.

Conversation

Turn 1 — User

Give me a report I can share, comparing this run with the last one.

Turn 1 — Assistant

Single HTML file with inline CSS/JS, metrics summary, histogram per metric, regressions highlighted against the baseline run, and expandable failing cases.

Tools used in this turn:

- Write: harness/report/html.py
- Write: harness/report/template.html
- Bash: python -m harness report runs/latest --baseline runs/prev — report.html

Connections

- eval-harness
- Reporting
- RegressionTracking