{% extends "base.html" %} {% from "_macros.html" import rate %} {% set nav = 'history' %} {% block body %}

Benchmark history

{{ runs|length }} run(s) on this instance. Results never leave this machine — runs/ and data/ are gitignored.

Start a run

{% if busy %}

A run is in progress ({{ current }}). One at a time on purpose: two concurrent runs against the same endpoint would make stability.k measure queueing rather than the system.

{% endif %}

A real run against a hosted model costs money: cases × k API calls. The fixture system is free and offline. Running in parallel shortens the wall clock, not the bill — and it puts the provider's queueing inside latency_ms, so that one metric stops being comparable with a serial run. Every other metric is unaffected.

{% for g in gates %}{% endfor %} {% for r in runs %} {% for g in gates %}{% endfor %} {% else %} {% endfor %}
RunSystemDatasetk{{ g.replace('gate.','') }}Status
{{ r.run_id }} {{ r.sut_name }} {{ r.sut_version }} {{ r.dataset }} ({{ r.cases }}) {{ r.k }}{{ rate(r.rates[g]) }} {% if r.status in ('generating','scoring','starting') %} {{ r.status }} {% elif r.status == 'failed' %}failed {% elif r.scored %}scored {% else %}not scored{% endif %}
No runs yet. Start one above.
{% endblock %}