npm - @rpamis/comet - Versions diffs - 0.4.0-beta.3 → 0.4.0-beta.4 - Mend

@rpamis/comet 0.4.0-beta.3 → 0.4.0-beta.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (508) hide show

package/eval/local/report-style-demo.html ADDED Viewed

@@ -0,0 +1,867 @@
+<!doctype html>
+<html lang="zh-CN" data-lang="zh">
+  <head>
+    <meta charset="utf-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1" />
+    <title>Comet Eval 论文级报告样板</title>
+    <style>
+      :root {
+        color-scheme: light;
+        --paper: #f7f4ed;
+        --panel: #fffdf8;
+        --ink: #191817;
+        --muted: #66615a;
+        --line: #d7d0c3;
+        --grid: #e9e2d6;
+        --current: #1f5f8b;
+        --baseline: #9f3d2f;
+        --good: #28745c;
+        --bad: #a33a32;
+        --accent: #a87318;
+        --code: #eee7dc;
+        --paper-font:
+          'Times New Roman', SimSun, 'Songti SC', 'Noto Serif CJK SC', 'Source Han Serif SC', serif;
+        --mono: 'Times New Roman', SimSun, 'Songti SC', 'Noto Serif CJK SC', serif;
+      }
+      * {
+        box-sizing: border-box;
+      }
+      html {
+        background: var(--paper);
+      }
+      body {
+        margin: 0;
+        color: var(--ink);
+        background:
+          linear-gradient(90deg, rgba(25, 24, 23, 0.025) 1px, transparent 1px),
+          linear-gradient(180deg, rgba(25, 24, 23, 0.025) 1px, transparent 1px), var(--paper);
+        background-size: 28px 28px;
+        font: 16px/1.6 var(--paper-font);
+      }
+      .page {
+        width: min(1180px, calc(100vw - 40px));
+        margin: 28px auto 48px;
+        background: var(--panel);
+        border: 1px solid var(--line);
+        box-shadow: 0 18px 55px rgba(25, 24, 23, 0.08);
+      }
+      .toolbar {
+        display: flex;
+        justify-content: flex-end;
+        gap: 6px;
+        padding: 16px 20px 0;
+      }
+      .toolbar button {
+        height: 30px;
+        padding: 0 12px;
+        border: 1px solid var(--line);
+        border-radius: 4px;
+        background: #fff;
+        color: var(--muted);
+        font: 12px/1 var(--paper-font);
+        cursor: pointer;
+      }
+      html[data-lang='zh'] button[data-set-lang='zh'],
+      html[data-lang='en'] button[data-set-lang='en'] {
+        border-color: var(--ink);
+        color: var(--ink);
+      }
+      .hero {
+        display: grid;
+        grid-template-columns: 190px minmax(0, 1fr);
+        gap: 34px;
+        padding: 18px 42px 28px;
+        border-bottom: 1px solid var(--line);
+      }
+      .meta {
+        color: var(--muted);
+        font: 12px/1.45 var(--paper-font);
+        letter-spacing: 0.08em;
+        text-transform: uppercase;
+      }
+      .meta span {
+        display: block;
+        margin-bottom: 10px;
+      }
+      h1 {
+        max-width: 900px;
+        margin: 0 0 14px;
+        font: 700 clamp(32px, 4vw, 54px) / 1.12 var(--paper-font);
+        letter-spacing: 0;
+        text-wrap: balance;
+      }
+      .dek {
+        max-width: 880px;
+        margin: 0;
+        color: var(--muted);
+        font: 17px/1.65 var(--paper-font);
+        text-wrap: pretty;
+      }
+      code {
+        padding: 0.08rem 0.28rem;
+        border-radius: 4px;
+        background: var(--code);
+        font-family: var(--mono);
+        font-size: 0.92em;
+      }
+      .abstract {
+        display: grid;
+        grid-template-columns: minmax(0, 1.35fr) minmax(300px, 0.75fr);
+        gap: 30px;
+        padding: 26px 42px;
+        border-bottom: 1px solid var(--line);
+      }
+      .section-label {
+        margin: 0 0 10px;
+        color: var(--accent);
+        font: 700 11px/1.2 var(--paper-font);
+        letter-spacing: 0.1em;
+        text-transform: uppercase;
+      }
+      .abstract p {
+        margin: 0;
+        font-size: 18px;
+        line-height: 1.65;
+        text-wrap: pretty;
+      }
+      .findings {
+        margin: 0;
+        padding: 0 0 0 18px;
+        border-left: 2px solid var(--ink);
+        color: var(--muted);
+        font: 14px/1.55 var(--paper-font);
+        list-style: none;
+      }
+      .findings li + li {
+        margin-top: 10px;
+      }
+      .findings strong {
+        color: var(--ink);
+      }
+      .metrics {
+        display: grid;
+        grid-template-columns: repeat(4, minmax(0, 1fr));
+        border-bottom: 1px solid var(--line);
+      }
+      .metric {
+        min-height: 112px;
+        padding: 18px 22px;
+        border-right: 1px solid var(--line);
+      }
+      .metric:last-child {
+        border-right: 0;
+      }
+      .metric-label {
+        color: var(--muted);
+        font: 12px/1.25 var(--paper-font);
+        letter-spacing: 0.06em;
+        text-transform: uppercase;
+      }
+      .metric-value {
+        margin-top: 8px;
+        font: 700 32px/1 var(--paper-font);
+        font-variant-numeric: tabular-nums;
+      }
+      .metric-note {
+        margin-top: 8px;
+        color: var(--muted);
+        font: 13px/1.35 var(--paper-font);
+      }
+      .positive {
+        color: var(--good);
+      }
+      .negative {
+        color: var(--bad);
+      }
+      .content {
+        padding: 34px 42px 46px;
+      }
+      .figure {
+        background: #fff;
+        border: 1px solid var(--line);
+        border-radius: 4px;
+        overflow: hidden;
+      }
+      .figure + .figure,
+      .figure-grid,
+      .table-wrap,
+      .method {
+        margin-top: 28px;
+      }
+      .figure-head {
+        display: flex;
+        align-items: baseline;
+        justify-content: space-between;
+        gap: 18px;
+        padding: 16px 18px 12px;
+        border-bottom: 1px solid var(--line);
+      }
+      h2 {
+        margin: 0;
+        font: 700 15px/1.35 var(--paper-font);
+        letter-spacing: 0;
+      }
+      .kicker {
+        color: var(--muted);
+        font: 12px/1.25 var(--paper-font);
+        white-space: nowrap;
+      }
+      .figure-body {
+        padding: 16px 18px 18px;
+      }
+      .figure-body img {
+        display: block;
+        width: 100%;
+        height: auto;
+        border: 1px solid var(--grid);
+      }
+      .caption {
+        margin: 12px 0 0;
+        color: var(--muted);
+        font: 13px/1.55 var(--paper-font);
+        text-wrap: pretty;
+      }
+      .figure-grid {
+        display: grid;
+        grid-template-columns: minmax(0, 1fr) minmax(0, 1fr);
+        gap: 24px;
+      }
+      .table-wrap {
+        border: 1px solid var(--line);
+        border-radius: 4px;
+        overflow-x: auto;
+        background: #fff;
+      }
+      table {
+        width: 100%;
+        min-width: 920px;
+        border-collapse: collapse;
+        font: 13px/1.45 var(--paper-font);
+      }
+      caption {
+        caption-side: top;
+        padding: 14px 16px;
+        border-bottom: 1px solid var(--line);
+        color: var(--muted);
+        text-align: left;
+        font: 13px/1.5 var(--paper-font);
+      }
+      th,
+      td {
+        padding: 9px 11px;
+        border-bottom: 1px solid var(--grid);
+        text-align: left;
+        vertical-align: middle;
+      }
+      th {
+        background: #f5efe4;
+        font-weight: 700;
+        white-space: nowrap;
+      }
+      tbody tr:last-child td {
+        border-bottom: 0;
+      }
+      .num {
+        text-align: right;
+        font-family: var(--mono);
+        font-variant-numeric: tabular-nums;
+        white-space: nowrap;
+      }
+      .status {
+        display: inline-flex;
+        align-items: center;
+        justify-content: center;
+        min-width: 54px;
+        height: 22px;
+        border: 1px solid currentColor;
+        border-radius: 999px;
+        font: 700 11px/1 var(--paper-font);
+        letter-spacing: 0.04em;
+      }
+      .status.pass {
+        color: var(--good);
+        background: color-mix(in srgb, var(--good), transparent 92%);
+      }
+      .status.fail {
+        color: var(--bad);
+        background: color-mix(in srgb, var(--bad), transparent 92%);
+      }
+      .method {
+        display: grid;
+        grid-template-columns: 0.9fr 1.1fr;
+        gap: 28px;
+        padding-top: 28px;
+        border-top: 1px solid var(--line);
+      }
+      .method h2 {
+        margin: 0 0 12px;
+        font: 700 25px/1.2 var(--paper-font);
+      }
+      .method p,
+      .method li {
+        color: var(--muted);
+        font: 14px/1.6 var(--paper-font);
+      }
+      .method ul {
+        margin: 0;
+        padding-left: 18px;
+      }
+      [data-i18n='en'],
+      .chart-en {
+        display: none;
+      }
+      html[data-lang='en'] [data-i18n='zh'],
+      html[data-lang='en'] .chart-zh {
+        display: none;
+      }
+      html[data-lang='en'] [data-i18n='en'],
+      html[data-lang='en'] .chart-en {
+        display: revert;
+      }
+      html[data-lang='en'] img.chart-en {
+        display: block;
+      }
+      html[data-lang='zh'] .status[data-i18n='zh'],
+      html[data-lang='en'] .status[data-i18n='en'] {
+        display: inline-flex;
+      }
+      @media (max-width: 860px) {
+        .page {
+          width: 100%;
+          margin: 0;
+          border-left: 0;
+          border-right: 0;
+        }
+        .hero,
+        .abstract,
+        .figure-grid,
+        .method {
+          grid-template-columns: 1fr;
+        }
+        .hero,
+        .abstract,
+        .content {
+          padding-left: 22px;
+          padding-right: 22px;
+        }
+        .metrics {
+          grid-template-columns: repeat(2, minmax(0, 1fr));
+        }
+        .metric:nth-child(2) {
+          border-right: 0;
+        }
+        .metric:nth-child(-n + 2) {
+          border-bottom: 1px solid var(--line);
+        }
+      }
+    </style>
+  </head>
+  <body>
+    <main class="page">
+      <div class="toolbar" aria-label="Language">
+        <button type="button" data-set-lang="zh">中文</button>
+        <button type="button" data-set-lang="en">English</button>
+      </div>
+      <header class="hero">
+        <div class="meta">
+          <span data-i18n="zh">Comet Eval 报告</span>
+          <span data-i18n="en">Comet Evaluation Report</span>
+          <span data-i18n="zh">论文级样板</span>
+          <span data-i18n="en">Paper-style demo</span>
+          <span>2026-07-02</span>
+        </div>
+        <div>
+          <h1 data-i18n="zh">Workflow 回归分析：以论文图表方式呈现 eval 结果</h1>
+          <h1 data-i18n="en">Workflow regression analysis with paper-grade comparative figures</h1>
+          <p class="dek" data-i18n="zh">
+            示例数据来自现有 current-vs-0.3.9 串行 eval：<code>COMET_FULL</code> 对比
+            <code>COMET_FULL_039</code>。中文为准，英文仅作为同版式翻译层。
+          </p>
+          <p class="dek" data-i18n="en">
+            Demo data comes from the existing current-vs-0.3.9 serial eval run:
+            <code>COMET_FULL</code> compared with <code>COMET_FULL_039</code>.
+          </p>
+        </div>
+      </header>
+      <section class="abstract">
+        <div>
+          <p class="section-label" data-i18n="zh">摘要</p>
+          <p class="section-label" data-i18n="en">Abstract</p>
+          <p data-i18n="zh">
+            当前 Comet workflow 在任务通过率和 rubric 质量上高于 0.3.9 参考实现，但 token
+            与成本消耗更高。报告应优先呈现实验结论、效应大小、质量-成本权衡和失败归因，而不是只把日志表格包装成
+            HTML。
+          </p>
+          <p data-i18n="en">
+            The current Comet workflow improves task success and rubric quality over the 0.3.9
+            reference, while consuming more tokens and cost. The report should foreground findings,
+            effect sizes, quality-cost trade-offs, and failure attribution rather than wrapping log
+            tables in HTML.
+          </p>
+        </div>
+        <ul class="findings">
+          <li data-i18n="zh">
+            <strong>主结论。</strong>任务通过率从 33.3% 提升到 83.3%，覆盖六个任务。
+          </li>
+          <li data-i18n="en">
+            <strong>Primary finding.</strong> Task pass rate increases from 33.3% to 83.3% across
+            six tasks.
+          </li>
+          <li data-i18n="zh"><strong>权衡。</strong>总 token 增加 5.90M，主要集中在长流程任务。</li>
+          <li data-i18n="en">
+            <strong>Trade-off.</strong> Token use increases by 5.90M tokens, mostly concentrated in
+            long workflow tasks.
+          </li>
+          <li data-i18n="zh">
+            <strong>风险信号。</strong>效率和决策点遵循度仍低于 0.3.9 参考实现。
+          </li>
+          <li data-i18n="en">
+            <strong>Risk signal.</strong> Efficiency and decision-point compliance remain below the
+            0.3.9 reference.
+          </li>
+        </ul>
+      </section>
+      <section class="metrics" aria-label="Headline metrics">
+        <div class="metric">
+          <div class="metric-label" data-i18n="zh">任务通过率</div>
+          <div class="metric-label" data-i18n="en">Task pass rate</div>
+          <div class="metric-value positive">+50.0 pp</div>
+          <div class="metric-note" data-i18n="zh">当前 5/6，基线 2/6</div>
+          <div class="metric-note" data-i18n="en">5/6 current vs 2/6 baseline</div>
+        </div>
+        <div class="metric">
+          <div class="metric-label" data-i18n="zh">Rubric 平均分</div>
+          <div class="metric-label" data-i18n="en">Rubric average</div>
+          <div class="metric-value positive">+0.18</div>
+          <div class="metric-note" data-i18n="zh">当前 0.72，基线 0.54</div>
+          <div class="metric-note" data-i18n="en">0.72 current vs 0.54 baseline</div>
+        </div>
+        <div class="metric">
+          <div class="metric-label" data-i18n="zh">检查通过率</div>
+          <div class="metric-label" data-i18n="en">Check pass rate</div>
+          <div class="metric-value positive">+8.9 pp</div>
+          <div class="metric-note" data-i18n="zh">当前 76/79，基线 69/79</div>
+          <div class="metric-note" data-i18n="en">76/79 current vs 69/79 baseline</div>
+        </div>
+        <div class="metric">
+          <div class="metric-label" data-i18n="zh">总成本</div>
+          <div class="metric-label" data-i18n="en">Total cost</div>
+          <div class="metric-value negative">+$3.63</div>
+          <div class="metric-note" data-i18n="zh">当前 $7.5947，基线 $3.9688</div>
+          <div class="metric-note" data-i18n="en">$7.5947 current vs $3.9688 baseline</div>
+        </div>
+      </section>
+      <section class="content">
+        <article class="figure">
+          <div class="figure-head">
+            <h2 data-i18n="zh">图 1. Rubric 维度相对 0.3.9 的变化</h2>
+            <h2 data-i18n="en">Figure 1. Rubric dimension deltas relative to 0.3.9</h2>
+            <span class="kicker" data-i18n="zh">均值差，越靠右越好</span>
+            <span class="kicker" data-i18n="en">Mean score difference, higher is better</span>
+          </div>
+          <div class="figure-body">
+            <img
+              class="chart-zh"
+              src="report-style-demo-assets/rubric_delta.zh.svg"
+              alt="Rubric 维度变化图"
+            />
+            <img
+              class="chart-en"
+              src="report-style-demo-assets/rubric_delta.en.svg"
+              alt="Rubric dimension delta chart"
+            />
+            <p class="caption" data-i18n="zh">
+              Python 生成的 lollipop/forest plot 比手写 SVG 更接近论文图表：零线是 0.3.9
+              参考实现，蓝色代表当前 workflow 提升，红色代表退步。
+            </p>
+            <p class="caption" data-i18n="en">
+              The Python-generated lollipop/forest plot is closer to paper figures than hand-coded
+              SVG. The zero line is the 0.3.9 reference; blue favors the current workflow and red
+              marks regressions.
+            </p>
+          </div>
+        </article>
+        <div class="figure-grid">
+          <article class="figure">
+            <div class="figure-head">
+              <h2 data-i18n="zh">图 2. 质量-成本前沿</h2>
+              <h2 data-i18n="en">Figure 2. Quality-cost frontier</h2>
+              <span class="kicker" data-i18n="zh">Rubric vs token</span>
+              <span class="kicker" data-i18n="en">Rubric vs token budget</span>
+            </div>
+            <div class="figure-body">
+              <img
+                class="chart-zh"
+                src="report-style-demo-assets/quality_cost.zh.svg"
+                alt="质量成本前沿图"
+              />
+              <img
+                class="chart-en"
+                src="report-style-demo-assets/quality_cost.en.svg"
+                alt="Quality-cost frontier chart"
+              />
+              <p class="caption" data-i18n="zh">
+                这个图把质量和消耗分开呈现，避免把 token 成本混进质量分数里。
+              </p>
+              <p class="caption" data-i18n="en">
+                This chart separates quality and spend so token cost does not get mixed into the
+                quality score.
+              </p>
+            </div>
+          </article>
+          <article class="figure">
+            <div class="figure-head">
+              <h2 data-i18n="zh">图 3. 任务通过矩阵</h2>
+              <h2 data-i18n="en">Figure 3. Task outcome matrix</h2>
+              <span class="kicker" data-i18n="zh">六个 benchmark 任务</span>
+              <span class="kicker" data-i18n="en">Six benchmark tasks</span>
+            </div>
+            <div class="figure-body">
+              <img
+                class="chart-zh"
+                src="report-style-demo-assets/task_outcomes.zh.svg"
+                alt="任务通过矩阵"
+              />
+              <img
+                class="chart-en"
+                src="report-style-demo-assets/task_outcomes.en.svg"
+                alt="Task outcome matrix"
+              />
+              <p class="caption" data-i18n="zh">
+                任务矩阵保留 per-task 粒度，适合在正文给结论、在附录追踪具体失败。
+              </p>
+              <p class="caption" data-i18n="en">
+                The task matrix preserves per-task granularity, supporting conclusions in the main
+                text and failure tracing in the appendix.
+              </p>
+            </div>
+          </article>
+        </div>
+        <div class="table-wrap">
+          <table>
+            <caption data-i18n="zh">
+              表 1. Per-task 对比摘要。Tokens 和成本为每个 task/treatment run 的观测总量。
+            </caption>
+            <caption data-i18n="en">
+              Table 1. Per-task comparison summary. Tokens and costs are total observed usage for
+              each task/treatment run.
+            </caption>
+            <thead>
+              <tr>
+                <th data-i18n="zh">任务</th>
+                <th data-i18n="en">Task</th>
+                <th>Treatment</th>
+                <th data-i18n="zh">状态</th>
+                <th data-i18n="en">Status</th>
+                <th class="num">Checks</th>
+                <th class="num">Rubric</th>
+                <th class="num">Tokens</th>
+                <th class="num">Cost</th>
+                <th data-i18n="zh">失败归因</th>
+                <th data-i18n="en">Failure attribution</th>
+              </tr>
+            </thead>
+            <tbody>
+              <tr>
+                <td data-i18n="zh">API 缓存 TTL</td>
+                <td data-i18n="en">API cache TTL</td>
+                <td><code>COMET_FULL</code></td>
+                <td>
+                  <span class="status pass" data-i18n="zh">通过</span
+                  ><span class="status pass" data-i18n="en">PASS</span>
+                </td>
+                <td class="num">13/13</td>
+                <td class="num">0.79</td>
+                <td class="num">4,280,454</td>
+                <td class="num">$2.6856</td>
+                <td data-i18n="zh">正常</td>
+                <td data-i18n="en">OK</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">API 缓存 TTL</td>
+                <td data-i18n="en">API cache TTL</td>
+                <td><code>COMET_FULL_039</code></td>
+                <td>
+                  <span class="status fail" data-i18n="zh">失败</span
+                  ><span class="status fail" data-i18n="en">FAIL</span>
+                </td>
+                <td class="num">10/13</td>
+                <td class="num">0.33</td>
+                <td class="num">202,121</td>
+                <td class="num">$0.1795</td>
+                <td data-i18n="zh">缺少 OpenSpec 产物和状态机证据</td>
+                <td data-i18n="en">OpenSpec artifacts and state machine evidence missing</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">修复 median</td>
+                <td data-i18n="en">Fix median</td>
+                <td><code>COMET_FULL</code></td>
+                <td>
+                  <span class="status pass" data-i18n="zh">通过</span
+                  ><span class="status pass" data-i18n="en">PASS</span>
+                </td>
+                <td class="num">13/13</td>
+                <td class="num">0.70</td>
+                <td class="num">2,216,220</td>
+                <td class="num">$1.4787</td>
+                <td data-i18n="zh">正常</td>
+                <td data-i18n="en">OK</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">修复 median</td>
+                <td data-i18n="en">Fix median</td>
+                <td><code>COMET_FULL_039</code></td>
+                <td>
+                  <span class="status fail" data-i18n="zh">失败</span
+                  ><span class="status fail" data-i18n="en">FAIL</span>
+                </td>
+                <td class="num">10/13</td>
+                <td class="num">0.33</td>
+                <td class="num">148,330</td>
+                <td class="num">$0.1217</td>
+                <td data-i18n="zh">缺少 OpenSpec 产物和状态机证据</td>
+                <td data-i18n="en">OpenSpec artifacts and state machine evidence missing</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">完整流程</td>
+                <td data-i18n="en">Full workflow</td>
+                <td><code>COMET_FULL</code></td>
+                <td>
+                  <span class="status pass" data-i18n="zh">通过</span
+                  ><span class="status pass" data-i18n="en">PASS</span>
+                </td>
+                <td class="num">13/13</td>
+                <td class="num">0.84</td>
+                <td class="num">161,476</td>
+                <td class="num">$0.0942</td>
+                <td data-i18n="zh">正常</td>
+                <td data-i18n="en">OK</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">完整流程</td>
+                <td data-i18n="en">Full workflow</td>
+                <td><code>COMET_FULL_039</code></td>
+                <td>
+                  <span class="status pass" data-i18n="zh">通过</span
+                  ><span class="status pass" data-i18n="en">PASS</span>
+                </td>
+                <td class="num">13/13</td>
+                <td class="num">0.88</td>
+                <td class="num">224,163</td>
+                <td class="num">$0.1309</td>
+                <td data-i18n="zh">正常</td>
+                <td data-i18n="en">OK</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">性能去重</td>
+                <td data-i18n="en">Perf dedupe</td>
+                <td><code>COMET_FULL</code></td>
+                <td>
+                  <span class="status pass" data-i18n="zh">通过</span
+                  ><span class="status pass" data-i18n="en">PASS</span>
+                </td>
+                <td class="num">13/13</td>
+                <td class="num">0.79</td>
+                <td class="num">3,042,862</td>
+                <td class="num">$2.0644</td>
+                <td data-i18n="zh">正常</td>
+                <td data-i18n="en">OK</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">性能去重</td>
+                <td data-i18n="en">Perf dedupe</td>
+                <td><code>COMET_FULL_039</code></td>
+                <td>
+                  <span class="status fail" data-i18n="zh">失败</span
+                  ><span class="status fail" data-i18n="en">FAIL</span>
+                </td>
+                <td class="num">10/13</td>
+                <td class="num">0.34</td>
+                <td class="num">49,682</td>
+                <td class="num">$0.0280</td>
+                <td data-i18n="zh">缺少 OpenSpec 产物和状态机证据</td>
+                <td data-i18n="en">OpenSpec artifacts and state machine evidence missing</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">重构 counter</td>
+                <td data-i18n="en">Refactor counter</td>
+                <td><code>COMET_FULL</code></td>
+                <td>
+                  <span class="status pass" data-i18n="zh">通过</span
+                  ><span class="status pass" data-i18n="en">PASS</span>
+                </td>
+                <td class="num">14/14</td>
+                <td class="num">0.84</td>
+                <td class="num">1,536,724</td>
+                <td class="num">$1.0945</td>
+                <td data-i18n="zh">正常</td>
+                <td data-i18n="en">OK</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">重构 counter</td>
+                <td data-i18n="en">Refactor counter</td>
+                <td><code>COMET_FULL_039</code></td>
+                <td>
+                  <span class="status pass" data-i18n="zh">通过</span
+                  ><span class="status pass" data-i18n="en">PASS</span>
+                </td>
+                <td class="num">14/14</td>
+                <td class="num">0.80</td>
+                <td class="num">4,304,050</td>
+                <td class="num">$3.0287</td>
+                <td data-i18n="zh">正常</td>
+                <td data-i18n="en">OK</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">健壮配置</td>
+                <td data-i18n="en">Robust config</td>
+                <td><code>COMET_FULL</code></td>
+                <td>
+                  <span class="status fail" data-i18n="zh">失败</span
+                  ><span class="status fail" data-i18n="en">FAIL</span>
+                </td>
+                <td class="num">10/13</td>
+                <td class="num">0.33</td>
+                <td class="num">208,867</td>
+                <td class="num">$0.1774</td>
+                <td data-i18n="zh">缺少 OpenSpec 产物和状态机证据</td>
+                <td data-i18n="en">OpenSpec artifacts and state machine evidence missing</td>
+              </tr>
+              <tr>
+                <td data-i18n="zh">健壮配置</td>
+                <td data-i18n="en">Robust config</td>
+                <td><code>COMET_FULL_039</code></td>
+                <td>
+                  <span class="status fail" data-i18n="zh">失败</span
+                  ><span class="status fail" data-i18n="en">FAIL</span>
+                </td>
+                <td class="num">12/13</td>
+                <td class="num">0.56</td>
+                <td class="num">621,935</td>
+                <td class="num">$0.4801</td>
+                <td data-i18n="zh">workflow 阶段证据不完整</td>
+                <td data-i18n="en">Workflow phase evidence incomplete</td>
+              </tr>
+            </tbody>
+          </table>
+        </div>
+        <section class="method">
+          <div>
+            <p class="section-label" data-i18n="zh">方法</p>
+            <p class="section-label" data-i18n="en">Methods</p>
+            <h2 data-i18n="zh">图表由 Python 生成，HTML 只负责报告排版</h2>
+            <h2 data-i18n="en">Charts are generated by Python; HTML owns report layout only</h2>
+            <p data-i18n="zh">
+              这个样板把数据可视化从 HTML 中拆出来：Python 脚本生成可复现 SVG，HTML
+              引用图表资产并提供中英文切换。后续真实 report renderer 可以继续复用这个边界。
+            </p>
+            <p data-i18n="en">
+              This sample separates visualization from report layout: the Python script generates
+              reproducible SVG assets, while HTML references those figures and provides bilingual
+              switching.
+            </p>
+          </div>
+          <ul>
+            <li data-i18n="zh">中文内容是主版本；英文只作为同结构翻译。</li>
+            <li data-i18n="en">
+              Chinese is the source version; English is the same-structure translation.
+            </li>
+            <li data-i18n="zh">图表文件位于 <code>eval/local/report-style-demo-assets/</code>。</li>
+            <li data-i18n="en">
+              Chart files live under <code>eval/local/report-style-demo-assets/</code>.
+            </li>
+            <li data-i18n="zh">当前没有多次 run 的方差，因此不画置信区间。</li>
+            <li data-i18n="en">
+              The current demo has no multi-run variance, so confidence intervals are not drawn.
+            </li>
+          </ul>
+        </section>
+      </section>
+    </main>
+    <script>
+      const applyLanguage = (lang) => {
+        const next = lang === 'en' ? 'en' : 'zh';
+        document.documentElement.dataset.lang = next;
+        document.documentElement.lang = next === 'zh' ? 'zh-CN' : 'en';
+        localStorage.setItem('comet-report-demo-lang', next);
+      };
+      document.querySelectorAll('[data-set-lang]').forEach((button) => {
+        button.addEventListener('click', () => applyLanguage(button.dataset.setLang));
+      });
+      applyLanguage(localStorage.getItem('comet-report-demo-lang') || 'zh');
+    </script>
+  </body>
+</html>