oh-my-knowledge 0.18.0 → 0.20.0

package/dist/src/renderer/summary.js

@@ -1,5 +1,274 @@
 import { e, fmtNum, fmtCost, fmtDuration, COLORS, t } from './layout.js';
 import { pValueCategory } from '../eval-core/statistics.js';
+import { computeVerdict } from '../eval-core/verdict.js';
+/**
+ * Verdict pill — sticky banner at the top of the HTML report giving the same
+ * one-line conclusion as the `omk bench verdict` CLI. Both surfaces share the
+ * computeVerdict rule engine so they can never disagree.
+ *
+ * Color coding follows traffic-light convention plus a yellow band for
+ * UNDERPOWERED / NOISE / CAUTIOUS — none of which are "ship" but none of which
+ * are "regress" either. SOLO is grey because it's an info pill, not a verdict.
+ */
+function levelColor(level) {
+    switch (level) {
+        case 'PROGRESS': return { bg: 'rgba(46, 204, 113, 0.15)', fg: 'var(--green)', border: 'var(--green)' };
+        case 'REGRESS': return { bg: 'rgba(231, 76, 60, 0.15)', fg: 'var(--red)', border: 'var(--red)' };
+        case 'CAUTIOUS': return { bg: 'rgba(255, 200, 80, 0.15)', fg: 'var(--yellow)', border: 'var(--yellow)' };
+        case 'UNDERPOWERED':
+        case 'NOISE': return { bg: 'rgba(180, 180, 180, 0.15)', fg: 'var(--text-muted)', border: 'var(--text-muted)' };
+        case 'SOLO': return { bg: 'rgba(180, 180, 180, 0.10)', fg: 'var(--text-secondary)', border: 'var(--border)' };
+    }
+}
+function levelLabel(level, lang) {
+    if (lang === 'zh') {
+        switch (level) {
+            case 'PROGRESS': return '进步';
+            case 'CAUTIOUS': return '需谨慎';
+            case 'REGRESS': return '回退';
+            case 'NOISE': return '无信号';
+            case 'UNDERPOWERED': return '样本不足';
+            case 'SOLO': return '单变体';
+        }
+    }
+    return level;
+}
+export function renderVerdictPill(report, lang) {
+    const result = computeVerdict(report);
+    const c = levelColor(result.level);
+    const label = levelLabel(result.level, lang);
+    const ration = result.rationale;
+    const bullets = [];
+    if (ration.significance)
+        bullets.push(`<span style="color:var(--text-secondary)">${e(ration.significance)}</span>`);
+    if (ration.layerWinners)
+        bullets.push(`<span style="color:var(--text-muted);font-size:12px">${lang === 'zh' ? '层' : 'layers'}: ${e(ration.layerWinners)}</span>`);
+    if (ration.sampleSize)
+        bullets.push(`<span style="color:var(--text-muted);font-size:12px">${lang === 'zh' ? '样本' : 'samples'}: ${e(ration.sampleSize)}</span>`);
+    if (ration.judgeAgreement)
+        bullets.push(`<span style="color:var(--text-muted);font-size:12px">${e(ration.judgeAgreement)}</span>`);
+    const ship = ration.shipRecommendation
+        ? `<div style="margin-top:8px;font-size:13px"><strong>${lang === 'zh' ? '建议' : 'Recommendation'}:</strong> ${e(ration.shipRecommendation)}</div>`
+        : '';
+    return `
+    <div style="margin:8px 0 16px;padding:14px 18px;background:${c.bg};border-left:4px solid ${c.border};border-radius:var(--radius)">
+      <div style="display:flex;align-items:baseline;gap:12px;flex-wrap:wrap">
+        <span style="font-size:13px;color:var(--text-muted);text-transform:uppercase;letter-spacing:0.06em">${lang === 'zh' ? '一句话结论' : 'Verdict'}</span>
+        <strong style="color:${c.fg};font-size:18px">${e(label)}</strong>
+        <span style="color:var(--text-secondary);font-size:14px">${e(result.headline)}</span>
+      </div>
+      <div style="margin-top:6px;display:flex;flex-direction:column;gap:2px">${bullets.join('')}</div>
+      ${ship}
+    </div>`;
+}
+/**
+ * Pairwise diff (treatment vs control) bootstrap CI table — populated only when
+ * --bootstrap was used and at least 2 variants ran. Each row shows whether
+ * treatment significantly outperformed control on compositeScore mean.
+ */
+export function renderPairwiseDiff(pairs, lang) {
+    if (!pairs || pairs.length === 0)
+        return '';
+    const validPairs = pairs.filter((p) => p.diffBootstrapCI);
+    if (validPairs.length === 0)
+        return '';
+    const rows = validPairs.map((p) => {
+        const ci = p.diffBootstrapCI;
+        const sigClass = ci.significant ? 'green' : 'text-muted';
+        const sigText = ci.significant ? t('bootstrapDiffSignificant', lang) : t('bootstrapDiffNotSignificant', lang);
+        const estColor = ci.estimate > 0 ? 'var(--green)' : ci.estimate < 0 ? 'var(--red)' : 'var(--text-muted)';
+        return `<tr>
+      <td><strong>${e(p.treatment)}</strong> ${lang === 'zh' ? 'vs' : 'vs'} ${e(p.control)}</td>
+      <td style="text-align:center;color:${estColor}"><strong>${ci.estimate >= 0 ? '+' : ''}${ci.estimate}</strong></td>
+      <td style="text-align:center;font-size:11px">[${ci.low}, ${ci.high}]</td>
+      <td style="text-align:center;color:var(--${sigClass})">${sigText}</td>
+      <td style="text-align:center;color:var(--text-muted);font-size:11px">${ci.samples}</td>
+    </tr>`;
+    }).join('');
+    return `
+    <h2 style="margin-top:24px">${lang === 'zh' ? '配对对比 (Bootstrap CI)' : 'Pairwise comparison (bootstrap CI)'}</h2>
+    <p style="font-size:13px;color:var(--text-secondary);margin:4px 0 12px">${lang === 'zh' ? 'control vs treatment 的均值差 95% CI。CI 不含 0 = 显著差异。bootstrap 不假设分布,适合 LLM 序数评分。' : '95% CI on (treatment - control) mean diff. 0 outside CI = significant. Bootstrap is distribution-free, fits ordinal LLM scores.'}</p>
+    <div class="table-wrap">
+    <table class="summary-table">
+      <thead><tr>
+        <th>${lang === 'zh' ? '对照' : 'Pair'}</th>
+        <th title="${t('bootstrapDiffLabel', lang)}">${t('bootstrapDiffLabel', lang)}</th>
+        <th>95% CI</th>
+        <th>${lang === 'zh' ? '显著性' : 'Significance'}</th>
+        <th>${lang === 'zh' ? '重采样数' : 'samples'}</th>
+      </tr></thead>
+      <tbody>${rows}</tbody>
+    </table>
+    </div>`;
+}
+/**
+ * Human-gold agreement section — rendered when --gold-dir was used at run time
+ * and the report has been re-persisted with `meta.humanAgreement`. Shows α
+ * (primary), bootstrap CI on α, weighted κ, Pearson — and surfaces a
+ * contamination warning prominently when the gold annotator id overlaps with
+ * the judge model id.
+ */
+export function renderHumanAgreement(agreement, lang) {
+    if (!agreement)
+        return '';
+    if (agreement.sampleCount === 0)
+        return '';
+    const fmt = (x) => Number.isNaN(x) ? 'NaN' : x.toFixed(3);
+    const a = agreement;
+    const alphaColor = Number.isNaN(a.alpha)
+        ? 'var(--text-muted)'
+        : a.alpha >= 0.8 ? 'var(--green)'
+            : a.alpha >= 0.667 ? 'var(--yellow)'
+                : 'var(--red)';
+    const interpret = lang === 'zh'
+        ? (Number.isNaN(a.alpha)
+            ? '评分多样性不足，α 未定义'
+            : a.alpha >= 0.8 ? '高度一致 — 结论可放心使用'
+                : a.alpha >= 0.667 ? '可接受 — 谨慎结论'
+                    : a.alpha >= 0.4 ? '较弱一致 — 结论需配合 CI 与人工抽检'
+                        : a.alpha >= 0 ? '偏差较大 — 排查 rubric / prompt'
+                            : '系统性反向 — 重新审视判分逻辑')
+        : (Number.isNaN(a.alpha)
+            ? 'insufficient rating variance for α'
+            : a.alpha >= 0.8 ? 'high agreement — conclusions trustworthy'
+                : a.alpha >= 0.667 ? 'acceptable — conclude cautiously'
+                    : a.alpha >= 0.4 ? 'weak agreement — pair with CI + spot-checks'
+                        : a.alpha >= 0 ? 'large divergence — investigate rubric / prompt'
+                            : 'systematic inversion — judge logic needs review');
+    const warningBlock = a.contaminationWarning
+        ? `<div style="margin:8px 0;padding:8px 12px;background:rgba(255, 200, 80, 0.12);border-left:3px solid var(--yellow);font-size:13px"><strong>${lang === 'zh' ? '污染警告' : 'Contamination warning'}:</strong> ${e(a.contaminationWarning)}</div>`
+        : '';
+    const missingNote = (a.missingCount > 0 || a.unscoredCount > 0)
+        ? `<p style="font-size:12px;color:var(--text-muted);margin:4px 0">${lang === 'zh' ? '注: 报告缺' : 'Note: report missing'} ${a.missingCount} ${lang === 'zh' ? '条 sample' : 'samples'}${a.unscoredCount > 0 ? `, ${a.unscoredCount} ${lang === 'zh' ? '条无评分' : 'unscored'}` : ''}</p>`
+        : '';
+    return `
+    <h2 style="margin-top:24px">${lang === 'zh' ? '人工锚点 (Human Gold)' : 'Human gold anchor'}</h2>
+    <p style="font-size:13px;color:var(--text-secondary);margin:4px 0 12px">${lang === 'zh' ? '对比 LLM 评委分与外部标注的一致性。α 解决"评委对不对"，区别于 CI 解决"评委稳不稳"。' : 'Agreement between the LLM judge and external annotations. α addresses "is the judge correct"; the bootstrap CI addresses "is it consistent".'}</p>
+    ${warningBlock}
+    <div class="table-wrap">
+    <table class="summary-table">
+      <thead><tr>
+        <th>${lang === 'zh' ? '指标' : 'Metric'}</th>
+        <th style="text-align:center">${lang === 'zh' ? '值' : 'Value'}</th>
+        <th style="text-align:center">95% CI</th>
+        <th>${lang === 'zh' ? '说明' : 'Note'}</th>
+      </tr></thead>
+      <tbody>
+        <tr>
+          <td><strong>Krippendorff α</strong></td>
+          <td style="text-align:center;color:${alphaColor}"><strong>${fmt(a.alpha)}</strong></td>
+          <td style="text-align:center;font-size:11px">[${fmt(a.alphaCI.low)}, ${fmt(a.alphaCI.high)}]</td>
+          <td style="font-size:12px;color:var(--text-secondary)">${lang === 'zh' ? '主指标，序数加权' : 'primary, ordinal-weighted'}</td>
+        </tr>
+        <tr>
+          <td>${lang === 'zh' ? '加权 κ' : 'weighted κ'}</td>
+          <td style="text-align:center">${fmt(a.weightedKappa)}</td>
+          <td style="text-align:center;color:var(--text-muted)">—</td>
+          <td style="font-size:12px;color:var(--text-secondary)">${lang === 'zh' ? '副指标，平方加权' : 'secondary, quadratic'}</td>
+        </tr>
+        <tr>
+          <td>Pearson r</td>
+          <td style="text-align:center">${fmt(a.pearson)}</td>
+          <td style="text-align:center;color:var(--text-muted)">—</td>
+          <td style="font-size:12px;color:var(--text-secondary)">${lang === 'zh' ? '只看 rank order' : 'rank-order only'}</td>
+        </tr>
+      </tbody>
+    </table>
+    </div>
+    <p style="margin:8px 0 0;font-size:13px"><strong>${lang === 'zh' ? '解读' : 'Reading'}:</strong> ${e(interpret)}</p>
+    <p style="margin:4px 0 0;font-size:12px;color:var(--text-muted)">${lang === 'zh' ? '标注者' : 'annotator'}: ${e(a.goldAnnotator)} (v${e(a.goldVersion)}) · variant: <strong>${e(a.variant)}</strong> · n=${a.sampleCount}</p>
+    ${missingNote}`;
+}
+/**
+ * Saturation curve — answers "did I run enough samples?". Renders an inline
+ * SVG of mean ± 95% CI shading vs cumulative N, one curve per variant. The
+ * verdict line below the chart calls out the saturation point (when the curve
+ * has flattened enough that more samples don't materially shrink the CI).
+ *
+ * Hidden when variance is absent (single-run reports) or when there are fewer
+ * than 2 checkpoints. We keep the chart visible at 2-4 checkpoints (= repeat
+ * < 5) so users can see the trajectory; the verdict is only computed at
+ * repeat ≥ 5.
+ */
+export function renderSaturationCurve(saturation, variants, lang) {
+    if (!saturation)
+        return '';
+    const checkpoints = saturation.checkpointSampleCounts;
+    if (!checkpoints || checkpoints.length < 2)
+        return '';
+    // Layout: 600 × 280 SVG, leave 50px on the left for y-axis and 30px at the
+    // bottom for x-axis labels. Score is the 1-5 Likert range; pin axes to that.
+    const width = 600, height = 280, padL = 50, padR = 20, padT = 16, padB = 32;
+    const plotW = width - padL - padR;
+    const plotH = height - padT - padB;
+    const xMax = Math.max(...checkpoints);
+    const xMin = Math.min(...checkpoints);
+    const yMin = 1, yMax = 5;
+    const xScale = (x) => padL + (plotW * (x - xMin)) / Math.max(1, xMax - xMin);
+    const yScale = (y) => padT + plotH - (plotH * (y - yMin)) / (yMax - yMin);
+    // Per-variant curves and CI ribbons.
+    const seriesParts = [];
+    variants.forEach((variant, i) => {
+        const trace = saturation.perVariant[variant];
+        if (!trace || trace.length === 0)
+            return;
+        const color = COLORS[i % COLORS.length];
+        // CI ribbon: polygon walking forward along upper bound and back along lower.
+        const upper = trace.map((p) => `${xScale(p.n).toFixed(1)},${yScale(p.ciHigh).toFixed(1)}`);
+        const lower = trace.map((p) => `${xScale(p.n).toFixed(1)},${yScale(p.ciLow).toFixed(1)}`).reverse();
+        const ribbonPoints = [...upper, ...lower].join(' ');
+        seriesParts.push(`<polygon points="${ribbonPoints}" fill="${color}" fill-opacity="0.15" stroke="none" />`);
+        // Mean line.
+        const linePoints = trace.map((p) => `${xScale(p.n).toFixed(1)},${yScale(p.mean).toFixed(1)}`).join(' ');
+        seriesParts.push(`<polyline points="${linePoints}" fill="none" stroke="${color}" stroke-width="2" />`);
+        // Mean dots.
+        for (const p of trace) {
+            seriesParts.push(`<circle cx="${xScale(p.n).toFixed(1)}" cy="${yScale(p.mean).toFixed(1)}" r="3" fill="${color}" />`);
+        }
+    });
+    // Y-axis ticks at 1..5; x-axis ticks at every checkpoint.
+    const yTicks = [1, 2, 3, 4, 5].map((y) => `<line x1="${padL}" y1="${yScale(y)}" x2="${width - padR}" y2="${yScale(y)}" stroke="var(--border)" stroke-width="0.5" />
+     <text x="${padL - 6}" y="${yScale(y) + 4}" font-size="10" text-anchor="end" fill="var(--text-muted)">${y}</text>`).join('');
+    const xTicks = checkpoints.map((n) => `<text x="${xScale(n)}" y="${height - padB + 14}" font-size="10" text-anchor="middle" fill="var(--text-muted)">${n}</text>`).join('');
+    // Per-variant verdict block underneath.
+    const verdictRows = [];
+    if (saturation.verdicts) {
+        for (const variant of variants) {
+            const v = saturation.verdicts[variant];
+            if (!v)
+                continue;
+            const flagColor = v.saturated ? 'var(--green)' : 'var(--yellow)';
+            const flagText = v.saturated
+                ? (lang === 'zh' ? `已饱和 (N=${v.atN})` : `saturated at N=${v.atN}`)
+                : (lang === 'zh' ? '尚未饱和' : 'not yet saturated');
+            const conf = v.confidence === 'low'
+                ? `<span style="color:var(--yellow)">⚠ ${lang === 'zh' ? '样本太少,结论参考意义有限' : 'low confidence, interpret cautiously'}</span>`
+                : '';
+            verdictRows.push(`<tr><td><strong>${e(variant)}</strong></td>
+         <td style="color:${flagColor}"><strong>${flagText}</strong></td>
+         <td style="font-size:12px;color:var(--text-secondary)">${e(v.reason)} ${conf}</td></tr>`);
+        }
+    }
+    const noteHtml = saturation.verdicts
+        ? `<div class="table-wrap"><table class="summary-table">
+       <thead><tr>
+         <th>${lang === 'zh' ? '变体' : 'Variant'}</th>
+         <th>${lang === 'zh' ? '判定' : 'Verdict'}</th>
+         <th>${lang === 'zh' ? '依据' : 'Rationale'}</th>
+       </tr></thead><tbody>${verdictRows.join('')}</tbody></table></div>`
+        : `<p style="font-size:13px;color:var(--text-muted);margin:8px 0 0">${lang === 'zh' ? '提示:repeat ≥ 5 时才会输出饱和判定。当前数据只够画曲线,不足以下结论。' : 'Note: saturation verdict needs repeat ≥ 5. Current data plots the curve only.'}</p>`;
+    return `
+    <h2 style="margin-top:24px">${lang === 'zh' ? '饱和曲线 (Saturation curve)' : 'Saturation curve'}</h2>
+    <p style="font-size:13px;color:var(--text-secondary);margin:4px 0 12px">${lang === 'zh' ? '随累积样本数 N 增长的均值与 95% CI。CI 宽度衰减放缓即饱和——再多样本对结论无实质收益。' : 'Mean and 95% CI as cumulative N grows. When CI shrink rate flattens, the evidence has saturated — more samples buy little.'}</p>
+    <svg viewBox="0 0 ${width} ${height}" xmlns="http://www.w3.org/2000/svg" style="width:100%;max-width:${width}px;height:auto;display:block">
+      ${yTicks}
+      <text x="${padL - 36}" y="${padT + plotH / 2}" font-size="11" text-anchor="middle" fill="var(--text-muted)" transform="rotate(-90 ${padL - 36} ${padT + plotH / 2})">${lang === 'zh' ? '均值' : 'mean'}</text>
+      <text x="${padL + plotW / 2}" y="${height - 4}" font-size="11" text-anchor="middle" fill="var(--text-muted)">${lang === 'zh' ? '累积样本数 N' : 'cumulative N'}</text>
+      ${xTicks}
+      ${seriesParts.join('\n')}
+    </svg>
+    ${noteHtml}`;
+}
 export function renderSummaryCards(variants, summary, lang, variance) {
     // 六维对比表格:事实 / 行为 / LLM 评价 / 成本 / 效率 / 稳定性。
     // 前三列(事实/行为/LLM 评价)是 task 级原始指标的 variant 聚合;成本/效率同。
@@ -131,7 +400,7 @@ export function renderSummaryCards(variants, summary, lang, variance) {
         ? '每行是一个实验分组（Variant），六列分别衡量不同维度：'
         : 'Each row is a Variant. Six columns measure independent dimensions:';
     const icon = (emoji) => `<span aria-hidden="true">${emoji}</span>`;
-    // 维度分隔加粗(border-top 2px),让四维的视觉边界更明显。sub 缩进从 28 收到 22。
+    // 维度分隔加粗(border-top 2px),让六维的视觉边界更明显。sub 缩进从 28 收到 22。
     const dim = 'style="padding:12px 0 4px;border-top:2px solid var(--border);color:var(--text-primary);font-weight:600"';
     const dimDesc = 'style="padding:12px 0 4px;border-top:2px solid var(--border);color:var(--text-secondary)"';
     const dimFirst = 'style="padding:4px 0 4px;color:var(--text-primary);font-weight:600"';
@@ -178,6 +447,51 @@ export function renderSummaryCards(variants, summary, lang, variance) {
       <thead>${thead}</thead>
       <tbody>${rows}</tbody>
     </table>
+    </div>
+    ${renderJudgeAgreementBlock(variants, summary, lang)}`;
+}
+/**
+ * Cross-sample inter-judge agreement table — only renders when at least one variant
+ * has multi-judge ensemble data. This is the v0.20.2 "blog headline" view: shows
+ * Pearson + MAD across the whole sample set, the metric that refutes "Claude judge
+ * Claude same-modality bias".
+ */
+function renderJudgeAgreementBlock(variants, summary, lang) {
+    const variantsWithEnsemble = variants.filter((v) => summary[v]?.judgeAgreement);
+    if (variantsWithEnsemble.length === 0)
+        return '';
+    const rows = variantsWithEnsemble.map((v) => {
+        const s = summary[v];
+        const ag = s.judgeAgreement;
+        const judgeList = (s.judgeModels || []).map((j) => `<code>${e(j)}</code>`).join(', ');
+        const pearsonCell = ag.pearson != null
+            ? `<span title="${t('pearsonDesc', lang)}" style="color:${ag.pearson >= 0.7 ? 'var(--green)' : ag.pearson >= 0.4 ? 'var(--yellow)' : 'var(--red)'}"><strong>${ag.pearson}</strong></span>`
+            : `<span style="color:var(--text-muted)">—</span>`;
+        const madCell = `<span title="${t('madDesc', lang)}" style="color:${ag.meanAbsDiff < 0.5 ? 'var(--green)' : ag.meanAbsDiff < 1.5 ? 'var(--yellow)' : 'var(--red)'}"><strong>${ag.meanAbsDiff}</strong></span>`;
+        return `<tr>
+      <td><strong>${e(v)}</strong></td>
+      <td style="font-size:11px;color:var(--text-muted)">${judgeList}</td>
+      <td style="text-align:center">${pearsonCell}</td>
+      <td style="text-align:center">${madCell}</td>
+      <td style="text-align:center;color:var(--text-muted)">${ag.sampleCount}</td>
+      <td style="text-align:center;color:var(--text-muted)">${ag.pairCount}</td>
+    </tr>`;
+    }).join('');
+    return `
+    <h2 style="margin-top:24px">${t('agreementHeader', lang)}</h2>
+    <p style="font-size:13px;color:var(--text-secondary);margin:4px 0 12px">${t('agreementDesc', lang)}</p>
+    <div class="table-wrap">
+    <table class="summary-table">
+      <thead><tr>
+        <th>${t('variants', lang)}</th>
+        <th>${t('judgeModelsLabel', lang)}</th>
+        <th title="${t('pearsonDesc', lang)}">${t('pearsonLabel', lang)}</th>
+        <th title="${t('madDesc', lang)}">${t('madLabel', lang)}</th>
+        <th>${lang === 'zh' ? '用例数' : 'Samples'}</th>
+        <th>${lang === 'zh' ? 'Judge 对数' : 'Pairs'}</th>
+      </tr></thead>
+      <tbody>${rows}</tbody>
+    </table>
     </div>`;
 }
 function buildDiagnostic(metric, cfg, winner, lang) {
@@ -221,10 +535,10 @@ function buildDiagnostic(metric, cfg, winner, lang) {
     if (!metric.significant && isStrong) {
         const leadPhrase = winner
             ? (lang === 'zh'
-                ? `${winner} 看似${cfg.winnerWordZh}但样本不足，建议加大 --repeat`
+                ? `${winner} 看似${cfg.winnerWordZh}但用例不足，建议加大 --repeat`
                 : `${winner} looks ${cfg.winnerWordEn.replace(' by', '')} but underpowered — increase --repeat`)
             : (lang === 'zh'
-                ? `${strongLabelZh}差异但样本不足，建议加大 --repeat`
+                ? `${strongLabelZh}差异但用例不足，建议加大 --repeat`
                 : `${strongLabelEn} effect but underpowered — increase --repeat`);
         return {
             icon: '⚠',
@@ -511,7 +825,7 @@ export function renderVarianceComparisons(variance, lang, layeredStatsOpen = fal
     const diagRulesZhData = [
         { variant: 'good', icon: '✓', title: '显著差异（中/大效应）', desc: '差异真实且有实际意义，可作为结论', example: '示例：v1 更便宜 · 显著 · g=1.04' },
         { variant: 'warn', icon: '⚠', title: '显著但效应微弱', desc: '差异真实但太小没实际价值，别过度解读', example: '示例：p<0.05 但 g≈0.1（--repeat 很大时易出现）' },
-        { variant: 'warn', icon: '⚠', title: '大效应但样本不足', desc: '差距看似大但样本太少，建议加大 --repeat 再判断', example: '示例：v2 胜出 0.30 · g=1.04 · 不显著' },
+        { variant: 'warn', icon: '⚠', title: '大效应但用例不足', desc: '差距看似大但用例太少，建议加大 --repeat 再判断', example: '示例：v2 胜出 0.30 · g=1.04 · 不显著' },
         { variant: 'neutral', icon: '—', title: '两变体相当', desc: '既不显著也效应微弱，可视为无差异', example: '示例：Δ≈0 · 不显著 · g<0.2' },
     ];
     const diagRulesEnData = [
@@ -743,7 +1057,7 @@ export function renderCoverageSection(coverage, lang) {
         }
         const uncoveredHint = hintLines.length > 0
             ? `<div style="font-size:11px;color:var(--text-muted);margin-top:10px;border-top:1px solid var(--border);padding-top:8px">
-          <div style="margin-bottom:4px">${lang === 'zh' ? '💡 以下知识未被任何样本访问,建议补充测试用例覆盖:' : '💡 These knowledge files were not accessed by any sample — consider adding test cases:'}</div>
+          <div style="margin-bottom:4px">${lang === 'zh' ? '💡 以下知识未被任何用例覆盖,建议补充测评用例:' : '💡 These knowledge files were not accessed by any sample — consider adding test cases:'}</div>
           ${hintLines.map((line) => `<div style="margin:2px 0">${line}</div>`).join('')}
         </div>`
             : '';
@@ -796,7 +1110,7 @@ export function renderKnowledgeInteractionSection(coverage, gapReports, lang) {
         ? '展示本次测评用例和知识库的交互画像——用到哪些知识（使用情况）· 哪些知识想找但没找到或模型表达不确定（盲区）。'
         : 'How this test set interacts with the KB — which knowledge was exercised (usage) · which was missed or flagged as uncertain (gaps).';
     const readHint = lang === 'zh'
-        ? '💡 读表：两者同时高 → 知识库内容有问题（有文件但答不出）· 同时低 → 测试用例太浅（没触到复杂场景）· 使用高 + 盲区低 → 理想但警惕样本驯化'
+        ? '💡 读表：两者同时高 → 知识库内容有问题（有文件但答不出）· 同时低 → 测评用例太浅（没触到复杂场景）· 使用高 + 盲区低 → 理想但警惕用例驯化'
         : '💡 Read together: both high → KB content issues (files exist but can\'t answer) · both low → test set too shallow · high use + low gap → ideal, but beware sample-set overfitting';
     // 聚合所有 variant(coverage / gap 任一侧存在即纳入)
     const allVariants = Array.from(new Set([
@@ -860,7 +1174,7 @@ export function renderKnowledgeInteractionSection(coverage, gapReports, lang) {
             }
             const uncoveredHint = hintLines.length > 0
                 ? `<div style="font-size:11px;color:var(--text-muted);margin-top:10px;border-top:1px solid var(--border);padding-top:8px">
-            <div style="margin-bottom:4px">${lang === 'zh' ? '💡 以下知识未被任何样本访问,建议补充测试用例:' : '💡 These knowledge files were not accessed — consider adding test cases:'}</div>
+            <div style="margin-bottom:4px">${lang === 'zh' ? '💡 以下知识未被任何用例覆盖,建议补充测评用例:' : '💡 These knowledge files were not accessed — consider adding test cases:'}</div>
             ${hintLines.map((line) => `<div style="margin:2px 0">${line}</div>`).join('')}
           </div>`
                 : '';
@@ -932,7 +1246,7 @@ export function renderKnowledgeInteractionSection(coverage, gapReports, lang) {
         <div class="ki-bar" role="progressbar" aria-valuenow="${pct}" aria-valuemin="0" aria-valuemax="100" aria-label="${e(barLabel)}">
           <div class="ki-bar-fill" style="width:${barW}%;background:${pctColor}"></div>
         </div>
-        <div style="font-size:11px;color:var(--text-muted);margin-bottom:4px">${gap.samplesWithGap}/${gap.sampleCount} ${lang === 'zh' ? '个样本出现搜索未命中或表达不确定' : 'samples with search miss / hedging'}</div>
+        <div style="font-size:11px;color:var(--text-muted);margin-bottom:4px">${gap.samplesWithGap}/${gap.sampleCount} ${lang === 'zh' ? '个用例出现搜索未命中或表达不确定' : 'samples with search miss / hedging'}</div>
         <div style="font-size:var(--fs-detail);color:var(--text-secondary);margin-bottom:8px">${weightedHint}</div>
         ${typeBadges ? `<div>${typeBadges}</div>` : ''}
         ${inventory ? `<details class="ki-details"><summary>${detailsLabel}</summary>