{% macro performance_table(rows) %}
<div class="scroll"><table class="performance-table"><thead><tr><th>指标 / 对象</th><th>差值</th><th>第一次</th><th>第二次</th><th>资格 / 原因</th></tr></thead><tbody>
{% for row in rows %}<tr data-metric="{{ row.metric_id }}" data-eligible="{{ 'true' if row.eligible else 'false' }}"><td>{{ row.metric_id }} · {{ row.statistic }}<br>{{ row.category or '—' }} · {{ row.case_id or '题型汇总' }}</td><td>{{ '%+.3f'|format(row.difference) if row.difference is not none else '—' }} {{ row.unit or '' }}</td><td>{{ '%.3f'|format(row.left) if row.left is not none else '—' }}</td><td>{{ '%.3f'|format(row.right) if row.right is not none else '—' }}</td><td>{{ '已获准' if row.eligible else '未获准' }}{% if row.reasons %}<br>{{ row.reasons|join('; ') }}{% endif %}<details><summary>来源与限制</summary><pre>{{ row.source }}
第一次：{{ row.left_limits|join('; ') }}
第二次：{{ row.right_limits|join('; ') }}</pre></details>{% if row.resource_windows is defined %}<details class="resource-window"><summary>观测窗口核对</summary><p>差值仅适用于所列采样窗口。覆盖率是样本时间跨度，不代表连续观测；系统读数不归因于单个模型。</p>{% for window in row.resource_windows %}<p>第{{ loop.index }}次：{% if window.coverage is defined %}覆盖 {{ window.coverage|percent }} · 样本 {{ window.samples }} · 排除 {{ window.excluded }} · 请求内相对位置 {{ window.offsets[0]|percent }}～{{ window.offsets[1]|percent }}{% else %}窗口证据不足{% endif %}</p>{% endfor %}</details>{% endif %}</td></tr>{% endfor %}
</tbody></table></div>
{% endmacro %}
{% if index.comparison %}<section class="comparison"{% if index.comparison_source is defined %} data-bound-pair="true"{% endif %}><h2>两次运行的比较</h2><p>模式：{{ index.comparison.mode }} · 因素：{{ index.comparison.factor or '无' }}。差值方向：第二次减第一次。</p><p>受控结论资格 · 质量：{{ '允许' if index.comparison.eligibility.quality else '不可比' }} · 性能：{{ '仅逐项允许' if index.comparison.eligibility.performance else '未获准' }}</p>
{% if index.comparison_source is defined %}<p><a href="{{ index.comparison_source.url }}">已复核的比较证据</a> · 筛选隐藏任一运行时，本比较一同隐藏。</p>{% endif %}
{% if index.comparison.calibration_status is defined %}
<p>总开销校准：{% for status in index.comparison.calibration_status %}第{{ loop.index }}次 {{ '已获准' if status.status == 'qualified' else '未提供' if status.status == 'not_supplied' else '受限' }}{% if not loop.last %} · {% endif %}{% endfor %}。增量 ABBA 是独立诊断。</p>
<details><summary>校准状态与适用范围</summary><pre>{{ index.comparison.calibration_status|pretty }}</pre></details>
{% endif %}
{% if index.comparison.observed_differences is defined %}
{% set observed = index.comparison.observed_differences %}
<section class="observed-differences"><h3>本次观测差值</h3>
<p>仅描述所列样本，不能归因于模型或设备。总体质量未完整评分或口径不匹配时不计算差值。</p>
<p>完成率差值：{{ observed.completion_rate.difference|percent }}</p>
<div class="scroll"><table><thead><tr><th>质量指标 / 题型</th><th>第一次</th><th>第二次</th><th>差值</th></tr></thead><tbody>
{% for row in observed.quality %}<tr><td>{{ row.metric_id }} · {{ row.category or '全部' }}</td><td>{{ row.left|display }}</td><td>{{ row.right|display }}</td><td>{{ row.difference|display }}</td></tr>{% endfor %}
</tbody></table></div>
<details><summary>样本范围与口径限制</summary><pre>{{ observed.sample_scope|pretty }}
{{ observed.reasons|pretty }}</pre></details>
<details><summary>条件差异</summary><pre>{{ observed.condition_differences|pretty }}</pre></details>
<details><summary>已有逐题评分观测</summary><pre>{{ observed.per_case|pretty }}</pre></details>
<h4>性能观测</h4><p>同口径观测值；差值仍须通过逐指标资格。</p>{{ performance_table(observed.performance) }}</section>
{% endif %}
{% if index.comparison.performance_analysis is defined %}
{% set analysis = index.comparison.performance_analysis %}
{% set permitted = analysis.differences|selectattr('eligible')|list %}
{% set refused = analysis.differences|rejectattr('eligible')|list %}
<section class="performance-deltas"><h3>逐指标性能差值</h3><p>{{ permitted|length }} 行获准。只描述这两次运行，不自动判断赢家；获准行数不代表独立题数。— 表示没有获准差值。</p>
<p class="id muted">第一次：{{ index.runs[0].source.run_id }}<br>第二次：{{ index.runs[1].source.run_id }}</p>
{% if analysis.blockers %}<details open><summary>共同阻断原因</summary><pre>{{ analysis.blockers|join('
') }}</pre></details>{% endif %}
{% if permitted %}{{ performance_table(permitted) }}{% else %}<p>暂无获准的性能差值。</p>{% endif %}
<details class="performance-refused"><summary>未获准指标与原因（{{ refused|length }} 行）</summary>{{ performance_table(refused) }}</details>
<details><summary>比较范围与限制</summary><pre>{{ analysis.limitations|join('
') }}</pre></details></section>
<details><summary>查看冻结条件</summary><pre>{{ index.comparison.conditions|pretty }}</pre></details>
{% else %}<details><summary>查看条件与阻断原因</summary><pre>{{ index.comparison|pretty }}</pre></details>{% endif %}
</section>{% endif %}