{# Evals > Compare: two runs picked (the latest two unless picked) and, once picked, only the cases that changed: flipped first, then a rule changed, added, removed; each opens on both runs' answers and every check before and after, a flip opened. A note first where the dataset differs between them. #} {% from "components/macros/feedback.html" import empty_state %} {% from "components/macros/form.html" import action, select %} {% from "components/macros/layout.html" import badge, card, goto %} {% from "components/macros/table.html" import data_table %} {% from "pages/overseer/ai/_evals_macros.html" import rule %} {% macro sides(row) %}
{{ side.label }} {{ badge(side.verdict.label, side.verdict.tone) }}
{{ side.answer or "-" }}
{% for r in compared.runs %}{{ goto(r.url, r.label) }} ({{ r.facts }}){% if loop.first %} to {% endif %}{% endfor %}
{% if compared.notice %}{{ compared.notice }}
{% endif %} {% call card(None, id="ai-evals-compare-card") %} {{ data_table([ {"key": "case", "label": "Case", "wrap": True}, {"key": "change", "label": "Change", "kind": "status"}, {"key": "before", "label": "Before", "kind": "status"}, {"key": "after", "label": "After", "kind": "status"}, {"key": "cost", "label": "Cost", "align": "right"}, {"key": "took", "label": "Took", "align": "right"}, ], compared.rows, empty="No differences.", detail=sides) }} {% endcall %}