{% extends "base.html" %}
{% block title %}Run · FreeWeight{% endblock %}
{% block content %}
{% if error %}
<h2>Run {{ run_ref }}</h2>
<div class="error-state" role="alert">
  <p><strong>This run could not be shown.</strong></p>
  <p>{{ error }}</p>
  <p class="muted"><a href="/runs">Back to all runs</a></p>
</div>
{% else %}
{% set run = detail.run %}
<h2>Run <code>{{ run.id }}</code></h2>
<p>
  <span class="status status-{{ run.status }}" id="run-status" data-run-id="{{ run.id }}"
        data-last-sequence="{{ detail.last_event_sequence }}" data-terminal="{{ 'true' if run.is_terminal else 'false' }}">{{ run.status }}</span>
  · {{ run.suite_key }} <span class="muted">{{ run.suite_version }}</span>
  · <code>{{ run.model_canonical_id }}</code>
  {% if run.label %} · {{ run.label }}{% endif %}
  {% if run.suite_key.startswith('goal.') and run.status == 'completed' %}
  · <a href="/runs/{{ run.id }}/grade">grade its human criteria</a>
  {% endif %}
</p>
{% if run.error_code %}
<div class="error-state" role="alert">
  <p><strong>{{ run.error_code }}</strong></p>
  <p>{{ run.error_text }}</p>
</div>
{% endif %}

<p class="muted">
  Created {{ run.created_at | timestamp }} · started {{ run.started_at | timestamp }} ·
  completed {{ run.completed_at | timestamp }}<br>
  Fingerprint <code title="{{ run.reproducibility_fingerprint }}">{{ run.reproducibility_fingerprint }}</code>
</p>

{% if run.degradations %}
<div class="warning-state" role="status">
  <p><strong>This run was measured under conditions worth knowing about.</strong></p>
  <ul>
    {% for degradation in run.degradations %}
    <li><code>{{ degradation.kind }}</code> — {{ degradation.detail | tojson }}</li>
    {% endfor %}
  </ul>
</div>
{% endif %}

<h3>Provenance</h3>
<table>
  <caption class="muted">
    What would have to be true to repeat this run. Reproduce it with
    <code>freeweight run repeat {{ run.id[:10] }} --check</code>.
  </caption>
  <tbody>
    <tr>
      <th scope="row">Served context</th>
      <td>
        {% if run.served_context is none %}
        <span class="muted" title="Neither the caller, the provider nor the descriptor reported one.">—</span>
        {% else %}{{ run.served_context }} tokens
        <span class="muted">({{ run.served_context_source }})</span>{% endif %}
      </td>
    </tr>
    <tr>
      <th scope="row">Attributed device</th>
      <td>
        GPU {{ run.gpu_index if run.gpu_index is not none else "—" }}
        {% if run.multi_gpu_visible %}
        <span class="muted">· more than one GPU was visible; memory, KV and energy figures are
        unsupported unless the provider reports placement</span>
        {% endif %}
      </td>
    </tr>
    <tr>
      <th scope="row">Telemetry sampling overhead</th>
      <td class="numeric">
        {% if run.telemetry_overhead_percent is none %}
        <span class="muted" title="Not calibrated on this run.">—</span>
        {% else %}{{ "%.3f" | format(run.telemetry_overhead_percent) }} %{% endif %}
      </td>
    </tr>
    <tr>
      <th scope="row">Prompt pack</th>
      <td>
        {% if run.prompt_pack_id %}
        {{ run.prompt_pack_id }} {{ run.prompt_pack_version }}
        <code title="{{ run.prompt_pack_hash }}">{{ run.prompt_pack_hash[:19] }}…</code>
        <span class="muted">provenance only — the fingerprint takes each benchmark's own prompt
        subset hash</span>
        {% else %}<span class="muted">This suite renders no prompt records.</span>{% endif %}
      </td>
    </tr>
  </tbody>
</table>

<details>
  <summary>Fingerprint document</summary>
  <p class="muted">
    The inputs the fingerprint was taken over, stored in full. A hash that cannot be explained is
    no use during a regression hunt.
  </p>
  <pre><code>{{ run.fingerprint_document | tojson(indent=2) }}</code></pre>
</details>

{% if not run.is_terminal %}
<form method="post" action="/runs/{{ run.id }}/cancel">
  <button type="submit">Cancel run</button>
</form>
{% endif %}

<h3>Progress</h3>
<progress id="run-progress" value="0" max="1" aria-describedby="run-progress-text"></progress>
<p id="run-progress-text" class="muted">Waiting for events.</p>

<h3 id="tests">Tests</h3>
<table>
  <thead>
    <tr>
      <th scope="col">Test</th>
      <th scope="col">Status</th>
      <th scope="col">Cases</th>
      <th scope="col">Repetitions</th>
      <th scope="col">Skip reason</th>
      <th scope="col">Error</th>
    </tr>
  </thead>
  <tbody>
    {% for test in detail.tests %}
    <tr>
      <td><a href="/runs/{{ run.id }}/tests/{{ test.id }}">{{ test.test_name }}</a>
          <span class="muted">{{ test.test_key }}</span></td>
      <td><span class="status status-{{ test.status }}">{{ test.status }}</span></td>
      <td class="numeric">{{ test.completed_cases }} / {{ test.total_cases }}</td>
      <td class="numeric">{{ test.repetitions }}</td>
      <td>{{ test.skip_reason or "—" }}</td>
      <td>{{ test.error_code or "—" }}</td>
    </tr>
    {% else %}
    <tr><td colspan="6" class="muted">No tests enumerated yet.</td></tr>
    {% endfor %}
  </tbody>
</table>

<h3>Metrics</h3>
<table>
  <thead>
    <tr>
      <th scope="col">Metric</th>
      <th scope="col">Scope</th>
      <th scope="col">GPU</th>
      <th scope="col">Value</th>
      <th scope="col">Unit</th>
      <th scope="col">Aggregation</th>
      <th scope="col">Spread</th>
      <th scope="col">Samples</th>
      <th scope="col">Excluded</th>
      <th scope="col">Raw source</th>
    </tr>
  </thead>
  <tbody>
    {% for metric in detail.metrics %}
    <tr>
      <td>{{ metric.metric_key }}</td>
      <td>{{ "run" if metric.run_test_id is none else "test" }}</td>
      <td class="numeric">{{ metric.gpu_index if metric.gpu_index is not none else "—" }}</td>
      <td class="numeric">
        {% if metric.unavailable_reason %}
        <span class="muted" title="{{ metric.unavailable_reason }}">—</span>
        {% else %}
        {{ "%.4f" | format(metric.numeric_value) }}
        {% endif %}
      </td>
      <td>{{ metric.unit }}</td>
      <td>{{ metric.aggregation }}</td>
      <td class="numeric">
        {% if metric.stddev is none %}
        <span class="muted" title="One measurement has no spread.">—</span>
        {% else %}± {{ "%.4f" | format(metric.stddev) }}{% endif %}
      </td>
      <td class="numeric">{{ metric.sample_count if metric.sample_count is not none else "—" }}</td>
      <td class="numeric">{{ metric.excluded_count if metric.excluded_count is not none else "—" }}</td>
      {# UI standards §5: at most two clicks from a headline metric to the raw record behind it.
         A test-scope metric links straight at its samples; a run-scope one combines every test,
         so it links at the tests table, where each row links at its own samples. Either way the
         second click lands on stored rows. #}
      <td>
        {% if metric.run_test_id %}
        <a href="/runs/{{ detail.run.id }}/tests/{{ metric.run_test_id }}">samples</a>
        {% else %}
        <a href="#tests">every test</a>
        {% endif %}
      </td>
    </tr>
    {% else %}
    <tr><td colspan="10" class="muted">No aggregates yet — they are written once every sample is durable.</td></tr>
    {% endfor %}
  </tbody>
</table>

<h3>Telemetry</h3>
{% if charts %}
<p class="muted">
  {{ telemetry_samples }} observation(s) recorded while this run executed. Each series names its
  device; nothing is summed or averaged across GPUs. A gap in a line is a reading this machine
  could not take, never a zero.
</p>
{% for chart in charts %}
<figure class="chart">
  <figcaption id="chart-{{ chart.key }}-label">{{ chart.label }}</figcaption>
  <svg viewBox="0 0 100 100" preserveAspectRatio="none" role="img"
       aria-labelledby="chart-{{ chart.key }}-label" class="telemetry-chart">
    <polyline points="{{ chart.points }}" fill="none" stroke="currentColor" stroke-width="1"
              vector-effect="non-scaling-stroke" />
  </svg>
  <p class="muted">
    Axis 0 – {{ "%.2f" | format(chart.maximum) }} {{ chart.unit }} ·
    min {{ "%.2f" | format(chart.minimum) }} ·
    mean {{ "%.2f" | format(chart.mean) }} ·
    max {{ "%.2f" | format(chart.maximum) }} {{ chart.unit }} ·
    {{ chart.reported }} reported, {{ chart.missing }} not readable
  </p>
</figure>
{% endfor %}
{% else %}
<div class="empty-state">
  <p><strong>No telemetry was recorded for this run.</strong></p>
  <p>
    Either this machine reports none, or <code>telemetry.persist_during_runs</code> is off. It is
    real and empty, not missing.
  </p>
</div>
{% endif %}

<h3>Events</h3>
<ol id="run-events" class="event-log" aria-live="polite"></ol>
<noscript>
  <p class="muted">
    Live events need JavaScript. Everything above is rendered by the server and is complete as of
    this page load; reload to see the run's current state.
  </p>
</noscript>
<script src="/static/js/runs.js" defer></script>
{% endif %}
{% endblock %}