{% extends "base.html" %}
{% from "_macros.html" import rate %}
{% set nav = 'compare' %}
{% block body %}
<h1>Compare two runs</h1>
<p class="sub">Gate-by-gate delta. Only meaningful when both runs used the same dataset and
the same <code>metrics_version</code> — the numbers alone cannot warn you, so both are shown.</p>

<div class="card">
  <form class="inline" method="get">
    <div><label for="left">Baseline</label>
      <select name="left" id="left">
        {% for r in runs %}<option value="{{ r.run_id }}" {% if r.run_id==left %}selected{% endif %}>{{ r.run_id }} — {{ r.label }}</option>{% endfor %}
      </select></div>
    <div><label for="right">Candidate</label>
      <select name="right" id="right">
        {% for r in runs %}<option value="{{ r.run_id }}" {% if r.run_id==right %}selected{% endif %}>{{ r.run_id }} — {{ r.label }}</option>{% endfor %}
      </select></div>
    <button type="submit">Compare</button>
  </form>
</div>

{% if comparability %}
{% set bad = comparability.findings | selectattr('verdict', 'equalto', 'mismatch') | list %}
{% set unsure = comparability.findings | selectattr('verdict', 'equalto', 'unverifiable') | list %}
{% if bad %}
<p class="note bad"><strong>These runs are not comparable.</strong> The delta below measures the
harness as much as it measures the systems. R4: two runs compared against each other must share
their dataset, their generation budget and the settings that decide what the metrics mean.</p>
{% endif %}
{% if bad or unsure %}
<div class="scroll"><table>
  <thead><tr><th>Setting</th><th>Verdict</th><th>Baseline</th><th>Candidate</th><th>Why it matters</th></tr></thead>
  <tbody>
  {% for f in bad + unsure %}
    <tr>
      <td class="mono">{{ f.field }}</td>
      <td>{% if f.verdict == 'mismatch' %}<span class="bad">mismatch</span>{% else %}<span class="warn">unverified</span>{% endif %}</td>
      <td class="mono">{{ f.left if f.left is not none else '—' }}</td>
      <td class="mono">{{ f.right if f.right is not none else '—' }}</td>
      <td class="sub">{{ f.note }}</td>
    </tr>
  {% endfor %}
  </tbody>
</table></div>
<p class="note"><strong>Unverified is not "the same".</strong> Where a run does not record a
setting — an older run, or an adapter that cannot see the budget of the system behind it — nothing
can be concluded either way, and saying so is the point.</p>
{% endif %}
{% endif %}

{% if rows %}
  {% if left_summary and right_summary and left_summary.dataset != right_summary.dataset %}
    <p class="note bad">These runs used different datasets
    (<code>{{ left_summary.dataset }}</code> vs <code>{{ right_summary.dataset }}</code>).
    The deltas below compare answers to two different questions.</p>
  {% endif %}
  <div class="scroll"><table>
    <thead><tr><th>Metric</th>
      <th>{{ left_summary.label if left_summary else 'baseline' }}</th>
      <th>{{ right_summary.label if right_summary else 'candidate' }}</th>
      <th>Delta</th></tr></thead>
    <tbody>
    {% for row in rows %}
      <tr>
        <td class="mono">{{ row.gate }}</td>
        <td>{{ rate(row.left) }}</td>
        <td>{{ rate(row.right) }}</td>
        <td>
          {% if row.delta is none %}<span class="muted">n/a</span>
          {% elif row.delta > 0 %}<span class="ok rate">+{{ '%.1f'|format(row.delta * 100) }} pp</span>
          {% elif row.delta < 0 %}<span class="bad rate">{{ '%.1f'|format(row.delta * 100) }} pp</span>
          {% else %}<span class="muted rate">±0</span>{% endif %}
        </td>
      </tr>
    {% endfor %}
    </tbody>
  </table></div>
{% elif left and right %}
  <p class="muted">Nothing comparable — is either run unscored?</p>
{% endif %}
{% endblock %}