{% extends "base.html" %}
{% block title %}{{ goal.pack.name }} results — FreeWeight{% endblock %}
{% block content %}
<h2>{{ goal.pack.name }}</h2>
{% if error %}
<div class="error-state" role="alert"><p>{{ error }}</p></div>
{% else %}
<p class="muted">{{ goal.pack.intent }}</p>
<p>
  <span class="unit mono">{{ goal.pack.capability_id }}</span> ·
  <span class="unit mono">{{ goal.goal_hash }}</span>
  {% if goal.pack.unforked %}
  <span class="status status-interrupted"
        title="Unedited starter content. A voice measured on someone else's prompts is not your voice.">unforked</span>
  {% endif %}
</p>

{% if goal.pack.unforked %}
<div class="warning-state">
  <strong>This goal is running unedited starter content.</strong> Its results are badged
  <code>unforked</code> here, in the dashboard and in every export. Edit its criteria or its tasks
  and the badge comes off.
</div>
{% endif %}

<h3>Score method mix</h3>
<p class="muted">
  Beside the score, always. A 0.82 that is 80&nbsp;% rules is a different fact from a 0.82 a jury
  produced, and this is what tells the two apart.
</p>
<dl class="card-grid">
  {% for rung in ['rule', 'reference', 'human', 'judge'] %}
  <div class="card">
    <dt>{{ rung }}</dt>
    <dd>{{ '%.0f' % (mix[rung] * 100) }}%</dd>
    <p class="card-note">
      {% if rung == 'rule' %}Deterministic, free, and never disagrees with you.
      {% elif rung == 'reference' %}Deterministic against your own ground truth.
      {% elif rung == 'human' %}You graded it. Validity 1.0 by definition.
      {% else %}A jury graded it. This is the part calibration has to measure.{% endif %}
    </p>
  </div>
  {% endfor %}
</dl>

<h3>Calibration</h3>
{% if outcome %}
<dl class="card-grid">
  <div class="card">
    <dt>State</dt>
    <dd style="font-size: 17px;">{{ outcome.verdict.state.value | replace('_', ' ') }}</dd>
    <p class="card-note">
      {% if outcome.verdict.state.value == 'calibrated' %}Evidence is emitted.
      {% else %}No evidence is emitted — not discounted evidence, none.{% endif %}
    </p>
  </div>
  <div class="card">
    <dt>Weighted &kappa;<sub>w</sub></dt>
    <dd>{{ '%.2f' % outcome.verdict.weighted_kappa_w if outcome.verdict.weighted_kappa_w is not none else '—' }}</dd>
    <p class="card-note">over n<sub>holdout</sub> = {{ outcome.verdict.n_holdout }}</p>
  </div>
  <div class="card">
    <dt>Judge validity factor</dt>
    <dd>{{ '%.2f' % outcome.verdict.judge_validity_factor }}</dd>
    <p class="card-note">The sixth confidence factor (ADR-0032 §2).</p>
  </div>
</dl>
<p><a href="/goals/{{ goal.slug }}/report">Read the full agreement report</a></p>
{% else %}
<div class="empty-state">
  <p>This goal has never been calibrated. It will still run, and every sample will be
    inspectable — it just cannot emit capability evidence yet.
    <a href="/goals/{{ goal.slug }}/grade">Grade its samples</a> to change that.</p>
</div>
{% endif %}

<h3>Runs</h3>
{% if rows %}
<div class="table-scroll">
  <table data-table="goal-results" data-sortable="true" data-complete="true">
    <caption>
      Every stored metric of this goal, newest run first. A different <code>goal_hash</code> is a
      different suite version, so two rubrics never share a row.
    </caption>
    <thead>
      <tr>
        <th scope="col">Metric</th>
        <th scope="col" class="numeric">Value</th>
        <th scope="col" class="numeric">n</th>
        <th scope="col">Suite version</th>
        <th scope="col">Model</th>
        <th scope="col">Raw source</th>
      </tr>
    </thead>
    <tbody>
      {% for row in rows %}
      <tr>
        <td>{{ row.metric_key }}</td>
        <td class="numeric">
          {% if row.numeric_value is none %}
          <span class="muted" title="Unsupported: {{ row.unavailable_reason }}">—</span>
          {% else %}{{ '%.4g' % row.numeric_value }}{% endif %}
          <span class="unit">{{ row.unit }}</span>
        </td>
        <td class="numeric">{{ row.sample_count }}</td>
        <td class="mono">{{ row.suite_version }}</td>
        <td class="mono">{{ row.model_canonical_id }}</td>
        <td>
          <a href="/runs/{{ row.run_id }}{% if row.run_test_id %}/tests/{{ row.run_test_id }}{% endif %}">
            {{ row.run_id[:8] }}
          </a>
        </td>
      </tr>
      {% endfor %}
    </tbody>
  </table>
</div>
{% else %}
<div class="empty-state">
  <p>No completed run of this goal yet.</p>
  <pre>freeweight run start --model &lt;model&gt; --suite goal.{{ goal.slug }}</pre>
</div>
{% endif %}

<h3>Exports</h3>
<ul>
  <li><a href="/api/v1/goals/{{ goal.slug }}/export">benchmark.goal_pack</a> — the rubric itself,
    hash-pinned, for a consumer deciding comparability.</li>
  <li><a href="/api/v1/goals/{{ goal.slug }}/calibration/report/export">benchmark.calibration_report</a>
    — the agreement figures with their <code>n_holdout</code>, in the schema a consumer reads.</li>
  <li><a href="/api/v1/results/export?scope=suite&amp;selector=goal.{{ goal.slug }}&amp;format=json">results</a>
    — every run of this goal.</li>
</ul>
{% endif %}
{% endblock %}