{% extends "base.html" %}
{% block title %}Evidence — FreeWeight{% endblock %}
{% block content %}
<h2>Capability evidence</h2>
<p class="muted">
  What LoadCoach routes on: one record per model, runtime profile, machine and capability, each
  with ADR-0017's confidence beside it. A capability with no evidence is absent, never scored
  zero; a goal below its calibration gate emits nothing at all. The file form is
  <a href="/api/v1/evidence/export">/api/v1/evidence/export</a>.
</p>

{% if error %}
<div class="error-state" role="alert"><p>{{ error }}</p></div>
{% else %}

<form method="get" action="/evidence" class="filters">
  <div class="field">
    <label for="f-capability">Capability</label>
    <input id="f-capability" name="capability" value="{{ filters.capability }}" placeholder="tool_use">
  </div>
  <div class="field">
    <label for="f-model">Model</label>
    <input id="f-model" name="model" value="{{ filters.model }}" placeholder="ollama/qwen3.5:9b">
  </div>
  <div class="field">
    <label for="f-stale">Show</label>
    <select id="f-stale" name="stale">
      <option value="" {% if not filters.stale %}selected{% endif %}>everything</option>
      <option value="only" {% if filters.stale == 'only' %}selected{% endif %}>stale only</option>
    </select>
  </div>
  <button type="submit">Filter</button>
</form>

<dl class="card-grid">
  <div class="card">
    <dt>Records</dt>
    <dd>{{ rows | length }}</dd>
    <p class="card-note">Across {{ capability_count }} capabilit{{ 'y' if capability_count == 1 else 'ies' }}.</p>
  </div>
  <div class="card">
    <dt>Stale</dt>
    <dd>{{ stale_count }}</dd>
    <p class="card-note">Freshness below the threshold, or the environment drifted. Re-run to refresh.</p>
  </div>
</dl>

{% if groups %}
<h3>Subjects, grouped by base</h3>
<p class="muted">
  A subject is a base, or a base with one LoRA adapter applied. Subjects are shown side by side and
  never merged: evidence measured under an adapter describes that subject and nothing else — not
  the bare base, not a sibling adapter. A subject nobody has measured shows <code>—</code>, never
  the base's numbers.
</p>
<div class="table-scroll">
  <table data-table="subjects" data-complete="true">
    <caption>
      Every base with the subjects measured on it, and where each subject's evidence came from.
    </caption>
    <thead>
      <tr>
        <th scope="col">Base</th>
        <th scope="col">Subject</th>
        <th scope="col" class="numeric">Records</th>
        <th scope="col">Capabilities</th>
        <th scope="col">Evidence source</th>
      </tr>
    </thead>
    <tbody>
      {% for group in groups %}
      {% for subject in group.subjects %}
      <tr>
        {% if loop.first %}
        <td class="mono" rowspan="{{ group.subjects | length + (1 if group.hidden_subject_count else 0) }}">
          {{ group.base_canonical_id }}
          {% if group.adapter_count %}
          <span class="unit">{{ group.adapter_count }} adapter subject{{ '' if group.adapter_count == 1 else 's' }}</span>
          {% endif %}
        </td>
        {% endif %}
        <td class="mono">
          {% if subject.adapter_name %}
          <strong>{{ subject.adapter_name }}</strong>
          {% else %}
          <span class="unit">bare base</span>
          {% endif %}
          {% if subject.name_only %}
          <span class="status status-interrupted" title="The base was named, not proven by digest">name only</span>
          {% endif %}
        </td>
        <td class="numeric">{{ subject.record_count or '—' }}</td>
        <td>
          {% if subject.capabilities %}
          {% for capability in subject.capabilities %}<code>{{ capability }}</code>{% if not loop.last %} {% endif %}{% endfor %}
          {% else %}
          —
          {% endif %}
        </td>
        <td>
          {% if subject.source_run_count %}
          measured here, {{ subject.source_run_count }} run{{ '' if subject.source_run_count == 1 else 's' }}
          {% else %}
          <span class="muted">not measured — an adapter subject inherits nothing from its base</span>
          {% endif %}
        </td>
      </tr>
      {% endfor %}
      {% if group.hidden_subject_count %}
      <tr>
        <td class="muted" colspan="4">
          and {{ group.hidden_subject_count }} more subject{{ '' if group.hidden_subject_count == 1 else 's' }}
          on this base, not shown — filter for one by name to see it
        </td>
      </tr>
      {% endif %}
      {% endfor %}
    </tbody>
  </table>
</div>
{% endif %}

{% if rows %}
<div class="table-scroll">
  <table data-table="evidence" data-sortable="true" data-complete="true">
    <caption>
      Every stored record. Expand a row to see the six confidence factors and the metrics that
      contributed, with their weights and sample counts — the answer to "why is this score what it
      is".
    </caption>
    <thead>
      <tr>
        <th scope="col">Capability</th>
        <th scope="col">Subject</th>
        <th scope="col" class="numeric">Score</th>
        <th scope="col" class="numeric">Confidence</th>
        <th scope="col" class="numeric">Samples</th>
        <th scope="col">Measured</th>
        <th scope="col">Status</th>
        <th scope="col">Why</th>
      </tr>
    </thead>
    <tbody>
      {% for row in rows %}
      {% set record = row.record %}
      {% set staleness = row.staleness %}
      <tr>
        <td>
          <code>{{ record.capability_id }}</code>
          {% if record.is_goal_sourced %}
          <span class="status status-preparing" title="A user-authored goal produced or contributed to this record">goal</span>
          {% endif %}
        </td>
        <td class="mono">{{ record.subject_canonical_id }}
          {% if record.adapter_name %}
          <span class="status status-preparing" title="Measured with this LoRA adapter applied; it describes this subject and nothing else">{{ record.adapter_name }}</span>
          {% endif %}
          {% if record.identity_confidence == 'name_only' %}
          <span class="status status-interrupted" title="No digest: the weights may have changed under this name">name only</span>
          {% endif %}
        </td>
        <td class="numeric">{{ '%.3f' % record.score }}</td>
        <td class="numeric">{{ '%.2f' % record.confidence }}</td>
        <td class="numeric">{{ record.sample_count }}{% if record.excluded_count %} <span class="unit">(+{{ record.excluded_count }} excluded)</span>{% endif %}</td>
        <td>{{ record.measured_at.strftime('%Y-%m-%d') }} <span class="unit">{{ '%.0f' % staleness.age_days }} d ago</span></td>
        <td>
          {% if staleness.stale %}
          <span class="status status-interrupted">stale</span>
          {% else %}
          <span class="status status-completed">fresh</span>
          {% endif %}
        </td>
        <td>
          <details>
            <summary>explain</summary>
            {% if staleness.stale %}
            <p class="muted">{{ staleness.reasons | join(' ') }} <a href="/runs">Start a run</a> to refresh it.</p>
            {% endif %}
            <dl>
              <dt>Confidence factors</dt>
              <dd>
                sample {{ '%.2f' % record.factors.get('sample_factor', 0) }} ·
                consistency {{ '%.2f' % record.factors.get('consistency_factor', 0) }} ·
                freshness {{ '%.2f' % record.factors.get('freshness_factor', 0) }} (now {{ '%.2f' % staleness.freshness }}) ·
                environment {{ '%.2f' % record.factors.get('environment_factor', 0) }} ·
                identity {{ '%.2f' % record.factors.get('identity_factor', 0) }} ·
                judge validity {{ '%.2f' % record.judge_validity_factor }}
              </dd>
              <dt>Contributing metrics</dt>
              <dd>
                <ul>
                  {% for metric in record.contributing_metrics %}
                  <li><code>{{ metric.metric_key }}</code> weight {{ '%.2f' % metric.weight }}, n = {{ metric.sample_count }}, contributed {{ '%.3f' % metric.contribution }}</li>
                  {% endfor %}
                </ul>
              </dd>
              {% if record.calibration %}
              <dt>Judge agreement</dt>
              <dd>
                κ<sub>w</sub> {{ '%.2f' % record.calibration.kappa_w }} over {{ record.calibration.n_holdout }} held-out samples graded by {{ record.calibration.graded_by }};
                validity factor {{ '%.2f' % record.judge_validity_factor }}.
                {% if record.score_method_mix %}Score method mix:
                {% for rung, share in record.score_method_mix.items() %}{{ rung }} {{ '%.0f' % (share * 100) }}%{% if not loop.last %}, {% endif %}{% endfor %}.{% endif %}
              </dd>
              {% endif %}
              <dt>Provenance</dt>
              <dd>
                policy {{ record.policy_version }} · vocabulary {{ record.vocabulary_version }} ·
                profile <code>{{ record.runtime_profile_hash }}</code> ·
                machine <code>{{ record.machine_fingerprint[:12] }}…</code> ·
                runs {% for run_id in record.source_run_ids %}<a href="/runs/{{ run_id }}">{{ run_id[:8] }}</a>{% if not loop.last %}, {% endif %}{% endfor %}
              </dd>
            </dl>
          </details>
        </td>
      </tr>
      {% endfor %}
    </tbody>
  </table>
</div>
{% else %}
<div class="empty-state">
  <h3>No capability evidence yet</h3>
  <p>Evidence is computed when a run completes. Start one:</p>
  <pre>freeweight run start --model &lt;model&gt; --suite native.tool_use</pre>
  <p>A goal with judged criteria also needs to be calibrated before it emits evidence.</p>
</div>
{% endif %}
{% endif %}
{% endblock %}