{% set c = comparison %}
<section>
  <h2>{{ a }} vs {{ b }} <span class="muted small">raw metrics, no composite score</span></h2>
  <table class="data compare">
    <thead><tr><th>metric</th><th class="num">{{ a }}</th><th class="num">{{ b }}</th><th class="num">Δ (B − A)</th></tr></thead>
    <tbody>
    {% for d in c.deltas %}
      <tr>
        <td>{{ d.label }}{% if d.lower_is_better %} <span class="muted small">(lower is better)</span>{% endif %}</td>
        {% if d.metric == "success_rate" %}
          <td class="num">{{ d.a|pct }}</td><td class="num">{{ d.b|pct }}</td>
          <td class="num {% if d.delta %}{{ 'better' if d.delta > 0 else 'worse' }}{% endif %}">{% if d.delta is not none %}{{ (d.delta * 100)|delta(0) }} pts{% else %}—{% endif %}</td>
        {% elif d.metric in ("reported_cost_usd", "estimated_cost_usd") %}
          <td class="num">{{ d.a|cost }}</td><td class="num">{{ d.b|cost }}</td>
          <td class="num {% if d.delta %}{{ 'better' if (d.delta < 0) == d.lower_is_better else 'worse' }}{% endif %}">{{ d.delta|delta(4) }}</td>
        {% elif d.metric == "wall_time_seconds" %}
          <td class="num">{{ d.a|seconds }}</td><td class="num">{{ d.b|seconds }}</td>
          <td class="num {% if d.delta %}{{ 'better' if (d.delta < 0) == d.lower_is_better else 'worse' }}{% endif %}">{{ d.delta|delta(1) }} s</td>
        {% else %}
          <td class="num">{{ d.a|num }}</td><td class="num">{{ d.b|num }}</td>
          <td class="num {% if d.delta %}{{ 'better' if (d.delta < 0) == d.lower_is_better else 'worse' }}{% endif %}">{{ d.delta|delta }}</td>
        {% endif %}
      </tr>
    {% endfor %}
      <tr><td>valid runs</td><td class="num">{{ c.a_aggregate.n_valid }} / {{ c.a_aggregate.n_total }}</td><td class="num">{{ c.b_aggregate.n_valid }} / {{ c.b_aggregate.n_total }}</td><td></td></tr>
    </tbody>
  </table>
</section>

<section>
  <h2>Statistical evidence <span class="muted small">paired by task, cluster bootstrap over tasks</span></h2>
  {% set p = c.paired %}
  {% include "partials/paired.html" %}
</section>

<section>
  <h2>Per-task agreement</h2>
  <p class="chips">
    <span class="chip chip-pass">both passed: {{ c.summary.both_passed }}</span>
    <span class="chip chip-fail">both failed: {{ c.summary.both_failed }}</span>
    <span class="chip chip-a">{{ a }} passed / {{ b }} failed: {{ c.summary.a_only }}</span>
    <span class="chip chip-b">{{ b }} passed / {{ a }} failed: {{ c.summary.b_only }}</span>
    {% if c.summary.mixed %}<span class="chip">mixed (repetitions disagree): {{ c.summary.mixed }}</span>{% endif %}
    {% if c.summary.unverified %}<span class="chip">unverified: {{ c.summary.unverified }}</span>{% endif %}
  </p>
  <table class="data">
    <thead><tr><th>task</th><th class="center">{{ a }}</th><th class="center">{{ b }}</th><th>category</th></tr></thead>
    <tbody>
    {% for t in c.tasks %}
      {% set ca = matrix[t.task_key][a] %}{% set cb = matrix[t.task_key][b] %}
      <tr>
        <td>{{ t.task_key }}</td>
        <td class="center cell cell-{{ ca.state }}">{% if ca.primary_run %}<a href="/runs/{{ ca.primary_run.run_id }}">{{ ca.state|state_symbol }} {{ t.a_score|num }}</a>{% else %}—{% endif %}</td>
        <td class="center cell cell-{{ cb.state }}">{% if cb.primary_run %}<a href="/runs/{{ cb.primary_run.run_id }}">{{ cb.state|state_symbol }} {{ t.b_score|num }}</a>{% else %}—{% endif %}</td>
        <td><span class="chip chip-{{ t.category }}">{{ t.category|replace("_", " ") }}</span></td>
      </tr>
    {% endfor %}
    </tbody>
  </table>
</section>