{% extends "layout.html" %} {% block title %}Trajectory · run {{ run_id }} · cua-speedrun{% endblock %} {% block content %}

Trajectory · run {{ run_id }}

Every observation the agent saw and every action it took, in order. Back to run.

{% for t in tasks %}

{{ t.task_id }} · seed {{ t.seed }}

{% if t.instruction %}

{{ t.instruction }}

{% endif %} {% if not t.has_frames %}

No screenshots were captured for this task (older run, or the environment did not return frames).

{% endif %} {% for l in t.agent_logs %}
{{ "Agent stderr" if l.name == "agent.stderr" else "Agent stdout" }} {{ l.name }} · full captured stream
Open to load.
{% endfor %}
    {% for s in t.steps %} {% if s.kind == "observe" %}
  1. observe{% if s.dur %} · {{ "%.2f" | format(s.dur) }}s{% endif %}
    {% if s.frame %} screenshot {% else %}no frame{% endif %}
  2. {% elif s.kind == "step" %}
  3. ↳ act{% if s.dur %} · {{ "%.2f" | format(s.dur) }}s{% endif %} {{ s.actions | tojson }}
  4. {% elif s.kind == "done" %}
  5. agent marked done
  6. {% elif s.kind == "verdict" %}
  7. {{ "pass" if s.passed else "fail" }} {{ s.detail }}
  8. {% endif %} {% endfor %}
{% endfor %} {% endblock %}