TraceCore

Metrics Dashboard

Reproducibility rates, budget utilisation, and failure taxonomy across all stored runs.

Unique Tasks

Total Runs

{% if metrics_rows %} {% set ns = namespace(sum=0, count=0) %} {% for row in metrics_rows %} {% if row.reproducibility_rate is not none %} {% set ns.sum = ns.sum + row.reproducibility_rate %} {% set ns.count = ns.count + 1 %} {% endif %} {% endfor %}

Avg Repro Rate

{% if ns.count > 0 %}{{ "%.1f"|format(ns.sum / ns.count * 100) }}%{% else %}—{% endif %}

{% endif %}

{% if metrics_rows %}

Per-Task Breakdown

{% for row in metrics_rows %} {% endfor %}

Task	Agent	Runs	Repro Rate	Steps P50	TC P50	Avg Wall (s)	Failure Type	Termination Reason
`{{ row.task_ref or "—" }}`	{{ (row.agent or "—").split("/")[-1] }}	{{ row.run_count }}	{% set repro = row.reproducibility_rate %} {% if repro is not none %} {% set pct = repro * 100 %} {{ "%.1f"\|format(pct) }}% {% else %} — {% endif %}	{{ row.steps_p50 or "—" }}	{{ row.tool_calls_p50 or "—" }}	{{ row.avg_wall_clock_s or "—" }}	{% if row.failure_taxonomy %} {% for ft, cnt in row.failure_taxonomy.items() %} {{ ft }}: {{ cnt }} {% endfor %} {% else %} — {% endif %}	{% if row.termination_taxonomy %} {% for tr, cnt in row.termination_taxonomy.items() %} {{ tr }}: {{ cnt }} {% endfor %} {% else %} — {% endif %}

{% else %}

📊

No run data yet. Run some episodes first:

agent-bench run --agent agents/toy_agent.py --task filesystem_hidden_config@1 --seed 42

{% endif %}

Machine-readable data available at GET /api/metrics — supports ?task=<ref>, ?agent=<path>, and ?limit=N query params. CLI: agent-bench runs metrics --format json