niceeval 0.6.1 → 0.6.2

package/src/report/react/chart-math.ts

@@ -1,5 +1,4 @@
-// 通用图表数值工具:round-number 刻度生成与标签防撞布局,零依赖纯函数。
-// 从 src/view/app/lib/chart.ts 迁来(view 侧从这里 re-export),报告组件与 view 共用同一份。
+// 通用图表数值工具:round-number 刻度生成与点标签布局,零依赖纯函数。
 
 function niceNumber(range: number, round: boolean): number {
   if (range <= 0) return 1;
@@ -35,34 +34,114 @@ export function niceTicks(min: number, max: number, count = 5): number[] {
   return ticks;
 }
 
-export interface LabelInput {
+export interface PointLabelInput {
   cx: number;
   cy: number;
+  /** 标签文本的估算宽度(px)。 */
   width: number;
-  /** true = 文字锚在点左侧(text-anchor: end),false = 锚在右侧(text-anchor: start)。 */
-  anchorLeft: boolean;
+}
+
+/** 标签允许占用的画布范围(含坐标轴边距,标签可以借用空白边距)。 */
+export interface LabelBounds {
+  x0: number;
+  y0: number;
+  x1: number;
+  y1: number;
+}
+
+export interface PlacedPointLabel {
+  /** <text> 的锚点坐标(y 为文字基线)。 */
+  x: number;
+  y: number;
+  anchor: "start" | "end" | "middle";
+  /** 标签不在点左右紧邻位:渲染层补 leader line 连回原点。 */
+  leader: boolean;
+}
+
+/** 标签框高度(基线上方 10px、下方 2px,对应 11px 字号)。 */
+const LABEL_ASCENT = 10;
+const LABEL_DESCENT = 2;
+/** 数据点视为边长 2×DOT_R 的方形障碍(点半径 4.5 + 命中余量)。 */
+const DOT_R = 6;
+/** 紧邻位与点的水平间距;候选环按它逐环外扩。 */
+const NEAR = 10;
+const RINGS = [NEAR, NEAR + 12, NEAR + 26];
+
+interface Box {
+  x0: number;
+  y0: number;
+  x1: number;
+  y1: number;
+}
+
+function labelBoxAt(x: number, y: number, anchor: PlacedPointLabel["anchor"], width: number): Box {
+  const x0 = anchor === "start" ? x : anchor === "end" ? x - width : x - width / 2;
+  return { x0, y0: y - LABEL_ASCENT, x1: x0 + width, y1: y + LABEL_DESCENT };
+}
+
+function overlapArea(a: Box, b: Box): number {
+  const w = Math.min(a.x1, b.x1) - Math.max(a.x0, b.x0);
+  const h = Math.min(a.y1, b.y1) - Math.max(a.y0, b.y0);
+  return w > 0 && h > 0 ? w * h : 0;
+}
+
+function outOfBoundsArea(box: Box, bounds: LabelBounds): number {
+  const total = (box.x1 - box.x0) * (box.y1 - box.y0);
+  const w = Math.min(box.x1, bounds.x1) - Math.max(box.x0, bounds.x0);
+  const h = Math.min(box.y1, bounds.y1) - Math.max(box.y0, bounds.y0);
+  const inside = w > 0 && h > 0 ? w * h : 0;
+  return total - inside;
+}
+
+/** 一个点在距离 d 的候选环:左右紧邻、四个斜角、正上正下。顺序即同代价时的偏好序。 */
+function ringCandidates(p: PointLabelInput, d: number): Omit<PlacedPointLabel, "leader">[] {
+  const diag = d * 0.75;
+  return [
+    { x: p.cx + d, y: p.cy + 4, anchor: "start" },
+    { x: p.cx - d, y: p.cy + 4, anchor: "end" },
+    { x: p.cx + diag, y: p.cy - diag - 2, anchor: "start" },
+    { x: p.cx - diag, y: p.cy - diag - 2, anchor: "end" },
+    { x: p.cx, y: p.cy - d - LABEL_DESCENT - 2, anchor: "middle" },
+    { x: p.cx + diag, y: p.cy + diag + LABEL_ASCENT, anchor: "start" },
+    { x: p.cx - diag, y: p.cy + diag + LABEL_ASCENT, anchor: "end" },
+    { x: p.cx, y: p.cy + d + LABEL_ASCENT + 2, anchor: "middle" },
+  ];
 }
 
 /**
- * 贪心地把互相重叠的直接标签往下推开(按 y 排序、逐个查重叠再下移),点本身位置不变。
- * 散点常见同分不同价的聚簇,这一步避免标签叠成一团糊字。
+ * 散点直接标签的候选位择优布局(docs/feature/reports/library.md「MetricScatter」):
+ * 每个标签在点四周由近及远的候选环上打分——与已放置标签的重叠、与任何数据点的重叠、
+ * 越出画布的面积、离点距离逐项累加,取代价最小的候选。存在无冲突候选时标签必不遮点、
+ * 不叠标签、不越界;全候选冲突时取重叠最小者,绝不丢标签。重合点簇因此向不同方向散开,
+ * 而不是单向堆叠压到下方的数据点上。
  */
-export function layoutLabelOffsets(items: LabelInput[], lineHeight = 14, padX = 4): number[] {
-  const offsets = new Array(items.length).fill(0);
-  const order = [...items.keys()].sort((a, b) => items[a].cy - items[b].cy || items[a].cx - items[b].cx);
-  const placed: { x0: number; x1: number; y: number }[] = [];
+export function placePointLabels(points: PointLabelInput[], bounds: LabelBounds): PlacedPointLabel[] {
+  const dots: Box[] = points.map((p) => ({ x0: p.cx - DOT_R, y0: p.cy - DOT_R, x1: p.cx + DOT_R, y1: p.cy + DOT_R }));
+  const placedBoxes: Box[] = [];
+  const result: PlacedPointLabel[] = new Array(points.length);
+  // 由上到下、由左到右逐个放置,布局确定可复现
+  const order = [...points.keys()].sort((a, b) => points[a].cy - points[b].cy || points[a].cx - points[b].cx);
+
   for (const i of order) {
-    const it = items[i];
-    const x0 = it.anchorLeft ? it.cx - it.width : it.cx;
-    const x1 = it.anchorLeft ? it.cx : it.cx + it.width;
-    let y = it.cy;
-    let guard = 0;
-    while (guard < 40 && placed.some((p) => x0 < p.x1 + padX && x1 > p.x0 - padX && Math.abs(y - p.y) < lineHeight)) {
-      y += lineHeight;
-      guard++;
+    const p = points[i];
+    let best: { candidate: Omit<PlacedPointLabel, "leader">; box: Box; cost: number; leader: boolean } | null = null;
+    for (let ring = 0; ring < RINGS.length; ring++) {
+      const candidates = ringCandidates(p, RINGS[ring]);
+      for (let c = 0; c < candidates.length; c++) {
+        const candidate = candidates[c];
+        const box = labelBoxAt(candidate.x, candidate.y, candidate.anchor, p.width);
+        let cost = 0;
+        for (const placed of placedBoxes) cost += overlapArea(box, placed) * 4;
+        for (const dot of dots) cost += overlapArea(box, dot) * 3;
+        cost += outOfBoundsArea(box, bounds) * 6;
+        cost += ring * 8 + c * 0.5; // 近环、偏好序靠前者优先,只作同冲突程度下的排序
+        if (best === null || cost < best.cost) {
+          best = { candidate, box, cost, leader: !(ring === 0 && c <= 1) };
+        }
+      }
     }
-    placed.push({ x0, x1, y });
-    offsets[i] = y - it.cy;
+    placedBoxes.push(best!.box);
+    result[i] = { ...best!.candidate, leader: best!.leader };
   }
-  return offsets;
+  return result;
 }

package/src/report/react/enhance.js

@@ -1,5 +1,5 @@
 // niceeval 报告的渐进增强 runtime:纯 vanilla JS、零依赖、IIFE、幂等。
-// 只作用于 .nre DOM 与 data-nre-* 属性;三个行为——表格排序、行过滤、SVG 点 tooltip。
+// 只作用于 .nre DOM 与 data-nre-* 属性;四个行为——实验组切换、表格排序、行过滤、SVG 点 tooltip。
 // 静态 HTML 无 JS 时内容完整可读是硬约束:排序有数据侧预排、tooltip 退化为原生
 // <title>、过滤输入框静默无功能。全部经 document 级事件委托绑定,重复注入本文件
 // 只在首次生效(window.__nreEnhanced 守卫),DOM 被搬动(如 view 把 <template> 内容
@@ -14,6 +14,38 @@
     return target && target.closest ? target.closest(selector) : null;
   }
 
+  // ───────────────────────── 实验组:[data-nre-experiment-group-select] ─────────────────────────
+  // 数据侧已经把每组的摘要、散点和列表分别算好；这里仅在同一报告节点内切换 <details>，
+  // 不重新请求、不重算，也不触碰 Runs / Traces / Attempt 证据室。
+
+  function selectExperimentGroup(control) {
+    var root = control.closest("[data-nre-experiment-groups]");
+    if (!root) return;
+    var selected = control.getAttribute("data-nre-experiment-group-select");
+    var controls = root.querySelectorAll("[data-nre-experiment-group-select]");
+    var panels = root.querySelectorAll("[data-nre-experiment-group-panel]");
+    for (var i = 0; i < controls.length; i++) {
+      var active = controls[i].getAttribute("data-nre-experiment-group-select") === selected;
+      controls[i].setAttribute("aria-selected", active ? "true" : "false");
+      controls[i].tabIndex = active ? 0 : -1;
+    }
+    for (var j = 0; j < panels.length; j++) {
+      panels[j].open = panels[j].getAttribute("data-nre-experiment-group-panel") === selected;
+    }
+  }
+
+  document.addEventListener("click", function (e) {
+    var control = closest(e.target, "[data-nre-experiment-group-select]");
+    if (control) selectExperimentGroup(control);
+  });
+
+  document.addEventListener("keydown", function (e) {
+    var control = closest(e.target, "[data-nre-experiment-group-select]");
+    if (!control || (e.key !== "Enter" && e.key !== " ")) return;
+    e.preventDefault();
+    selectExperimentGroup(control);
+  });
+
   // ───────────────────────── 排序:th[data-nre-sort] ─────────────────────────
   // 点击按该列排序 tbody 行(td/th 的 data-sort-value,数值优先、退回字符串;
   // 空值恒沉底),再点反向;方向指示由 th 上的 nre-sort-asc / nre-sort-desc 类驱动。

package/src/report/react/fixtures.ts

@@ -21,7 +21,7 @@ import type { AttemptLocator } from "../../results/locator.ts";
 
 const locator = (s: string): AttemptLocator => s as AttemptLocator;
 
-export const passRateColumn: MetricColumn = { key: "pass-rate", label: "pass rate", unit: "%", better: "higher" };
+export const passRateColumn: MetricColumn = { key: "task-pass-rate", label: "pass rate", unit: "%", better: "higher" };
 export const codeLinesColumn: MetricColumn = { key: "code-lines", label: "code lines", unit: "lines", better: "lower" };
 export const costColumn: MetricColumn = { key: "cost", label: "cost", unit: "$", better: "lower" };
 
@@ -73,7 +73,7 @@ export const groupSummaryData: GroupSummaryData = {
   lastRunAt: "2026-07-01T11:30:00Z",
 };
 
-export const tableData: TableData<"pass-rate" | "code-lines"> = {
+export const tableData: TableData<"task-pass-rate" | "code-lines"> = {
   dimension: "agent",
   // 行顺序故意不按 passRate 排:组件必须按传入顺序渲染,不重排
   columns: [passRateColumn, codeLinesColumn],
@@ -81,7 +81,7 @@ export const tableData: TableData<"pass-rate" | "code-lines"> = {
     {
       key: "codex",
       cells: {
-        "pass-rate": { value: 0.5, display: "50%", samples: 6, total: 6, refs: [] },
+        "task-pass-rate": { value: 0.5, display: "50%", samples: 6, total: 6, refs: [] },
         // 全 null:一个有效样本都没有 → 缺数据文案,绝不画 0
         "code-lines": { value: null, display: "—", samples: 0, total: 6, refs: [] },
       },
@@ -89,7 +89,7 @@ export const tableData: TableData<"pass-rate" | "code-lines"> = {
     {
       key: "bub",
       cells: {
-        "pass-rate": {
+        "task-pass-rate": {
           value: 0.87,
           display: "87%",
           samples: 6,
@@ -104,13 +104,13 @@ export const tableData: TableData<"pass-rate" | "code-lines"> = {
 };
 
 /** rows: "experiment" 的榜单形态:行携带 agent/model 元信息与 eval 级折叠计票。 */
-export const tableDataWithMeta: TableData<"pass-rate"> = {
+export const tableDataWithMeta: TableData<"task-pass-rate"> = {
   dimension: "experiment",
   columns: [passRateColumn],
   rows: [
     {
       key: "compare/bub",
-      cells: { "pass-rate": { value: 0.5, display: "50%", samples: 2, total: 2, refs: [] } },
+      cells: { "task-pass-rate": { value: 0.5, display: "50%", samples: 2, total: 2, refs: [] } },
       meta: {
         agent: "bub",
         model: "gpt-5.4",
@@ -119,7 +119,7 @@ export const tableDataWithMeta: TableData<"pass-rate"> = {
     },
     {
       key: "compare/codex",
-      cells: { "pass-rate": { value: 1, display: "100%", samples: 2, total: 2, refs: [] } },
+      cells: { "task-pass-rate": { value: 1, display: "100%", samples: 2, total: 2, refs: [] } },
       meta: {
         agent: "codex",
         verdicts: { passed: 2, failed: 0, errored: 0, skipped: 0 },
@@ -261,7 +261,7 @@ export const lineData: LineData = {
   ],
 };
 
-export const deltaData: DeltaData<"pass-rate" | "cost"> = {
+export const deltaData: DeltaData<"task-pass-rate" | "cost"> = {
   columns: [passRateColumn, costColumn],
   rows: [
     {
@@ -270,7 +270,7 @@ export const deltaData: DeltaData<"pass-rate" | "cost"> = {
       b: { experimentId: "compare/bub--agents-md" },
       cells: {
         // 通过率 +12pp:better higher → 好(绿)
-        "pass-rate": {
+        "task-pass-rate": {
           a: { value: 0.5, display: "50%", samples: 6, total: 6, refs: [] },
           b: { value: 0.62, display: "62%", samples: 6, total: 6, refs: [] },
           delta: 0.12,
@@ -290,7 +290,7 @@ export const deltaData: DeltaData<"pass-rate" | "cost"> = {
       a: { experimentId: "compare/codex" },
       b: { experimentId: "compare/codex--agents-md" },
       cells: {
-        "pass-rate": {
+        "task-pass-rate": {
           a: { value: 0.4, display: "40%", samples: 6, total: 6, refs: [] },
           b: { value: 0.4, display: "40%", samples: 6, total: 6, refs: [] },
           delta: 0,
@@ -322,8 +322,9 @@ const failedAttempt: AttemptListItem = {
       name: "roots-correct",
       severity: "gate",
       score: 0,
-      passed: false,
-      detail: "expected x=2, got x=3",
+      outcome: "failed" as const,
+      expected: "x=2",
+      received: "x=3",
       evidence: "judge: sign flipped when substituting into the quadratic formula",
     },
   ],
@@ -338,7 +339,17 @@ const erroredAttempt: AttemptListItem = {
   attempt: 0,
   agent: "codex",
   verdict: "errored",
-  error: "TypeError: cannot read properties of undefined (reading 'foo')",
+  // 结构化 error:列表只渲染 message 一层摘要;cause/stack 与 diagnostics 是下钻详情,随数据携带
+  error: {
+    code: "unexpected-error",
+    message: "TypeError: cannot read properties of undefined (reading 'foo')",
+    phase: "eval.run",
+    stack: "TypeError: cannot read properties of undefined (reading 'foo')\n    at run (adapter.ts:42:7)",
+    cause: { name: "TypeError", message: "cannot read properties of undefined (reading 'foo')" },
+  },
+  diagnostics: [
+    { code: "sandbox-teardown-failed", level: "warning", message: "sandbox teardown timed out", phase: "sandbox.teardown" },
+  ],
   assertions: [],
   durationMs: 4_500,
   locator: locator("@1c1c1c1"),
@@ -358,7 +369,6 @@ export const evalListItems: EvalListItem[] = [
     evalId: "algebra/quadratic",
     experimentId: "compare/bub",
     verdict: "failed",
-    reason: "roots-correct: expected x=2, got x=3",
     score: { value: 0, display: "0%", samples: 1, total: 1, refs: [failedAttempt.locator] },
     duration: { value: 32_000, display: "32.0s", samples: 1, total: 1, refs: [failedAttempt.locator] },
     cost: { value: 0.12, display: "$0.12", samples: 1, total: 1, refs: [failedAttempt.locator] },
@@ -368,7 +378,6 @@ export const evalListItems: EvalListItem[] = [
     evalId: "geometry/angles",
     experimentId: "compare/codex",
     verdict: "errored",
-    reason: erroredAttempt.error,
     score: { value: 0, display: "0%", samples: 1, total: 1, refs: [erroredAttempt.locator] },
     duration: { value: 4_500, display: "4.5s", samples: 1, total: 1, refs: [erroredAttempt.locator] },
     cost: { value: null, display: "—", samples: 0, total: 1, refs: [] },
@@ -394,7 +403,6 @@ export const experimentListItems: ExperimentListItem[] = [
       {
         evalId: "algebra/quadratic",
         verdict: "failed",
-        reason: "roots-correct: expected x=2, got x=3",
         duration: { value: 32_000, display: "32.0s", samples: 1, total: 1, refs: [failedAttempt.locator] },
         cost: { value: 0.12, display: "$0.12", samples: 1, total: 1, refs: [failedAttempt.locator] },
         attempts: [failedAttempt, failedRetryAttempt],
@@ -423,7 +431,6 @@ export const experimentListItems: ExperimentListItem[] = [
       {
         evalId: "geometry/angles",
         verdict: "errored",
-        reason: erroredAttempt.error,
         duration: { value: 4_500, display: "4.5s", samples: 1, total: 1, refs: [erroredAttempt.locator] },
         cost: { value: null, display: "—", samples: 0, total: 1, refs: [] },
         attempts: [erroredAttempt],

package/src/report/react/render.test.tsx

@@ -1,11 +1,9 @@
+// cases: docs/engineering/unit-tests/reports/cases.md
 // 「不 hydrate 也完整」的验收测试:每个组件过 renderToStaticMarkup,
 // 断言纯静态 HTML 里就有全部关键内容——数字、覆盖率角标、缺数据文案、
 // 散点的 SVG 与系列名、truncated 行、attemptHref 链接。
-// 另外锁两条契约:源码零 hooks;维度键跨组件同色。
+// 另外锁一条契约:维度键跨组件同色。
 
-import { readFileSync, readdirSync } from "node:fs";
-import { dirname, join } from "node:path";
-import { fileURLToPath } from "node:url";
 import { renderToStaticMarkup } from "react-dom/server";
 import { describe, expect, it } from "vitest";
 import type { AttemptLocator } from "../../results/locator.ts";
@@ -72,12 +70,6 @@ describe("RunOverview", () => {
 describe("MetricTable", () => {
   const html = renderToStaticMarkup(<MetricTable data={tableData} attemptHref={attemptHref} />);
 
-  it("按传入顺序渲染行,不重排(排序在数据侧)", () => {
-    // fixture 里 codex(50%)在 bub(87%)前面,输出必须保持
-    expect(html.indexOf(">codex<")).toBeGreaterThan(-1);
-    expect(html.indexOf(">codex<")).toBeLessThan(html.indexOf(">bub<"));
-  });
-
   it("列头带 label、unit 与 better 方向", () => {
     expect(html).toContain("pass rate");
     expect(html).toContain("(%)");
@@ -157,21 +149,12 @@ describe("MetricMatrix", () => {
     expect(html).toContain("eval × agent");
   });
 
-  it("稀疏格子:没有样本的格子空着(恰好一个)", () => {
-    expect(html.match(/nre-td-empty/g)).toHaveLength(1);
-  });
-
   it("格子数字与 refs 下钻链接", () => {
     expect(html).toContain("100%");
     expect(html).toContain("0%");
     expect(html).toContain('href="/attempts/@1b3b3b3"');
     expect(html).toContain('href="/attempts/@1b7b7b7"');
   });
-
-  it("列头(维度键)带稳定散列配色 class", () => {
-    expect(html).toContain(colorClassForKey("bub"));
-    expect(html).toContain(colorClassForKey("codex"));
-  });
 });
 
 describe("Scoreboard", () => {
@@ -243,11 +226,6 @@ describe("MetricScatter", () => {
     expect(html).toContain(">codex-mid</text>");
   });
 
-  it("null 点不画,底部注脚如实报数", () => {
-    expect(html).not.toContain('data-key="compare/codex-broken"');
-    expect(html).toContain("1 point missing data");
-  });
-
   it("hover 退化为 <title>:display 与 samples/total", () => {
     expect(html).toContain("<title>");
     expect(html).toContain("50%(6/6)");
@@ -304,11 +282,6 @@ describe("DeltaTable", () => {
     expect(html).toContain("nre-delta-flat");
   });
 
-  it("任一侧缺数据:Δ 显示为缺,不硬算", () => {
-    expect(html).toContain("nre-delta-missing");
-    expect(html).toContain("no data");
-  });
-
   it("每行标出 A → B 的 experimentId", () => {
     expect(html).toContain("compare/bub → compare/bub--agents-md");
   });
@@ -322,14 +295,11 @@ describe("DeltaTable", () => {
 describe("AttemptList", () => {
   const html = renderToStaticMarkup(<AttemptList items={attemptListItems} />);
 
-  it("逐条断言:name、score、detail、evidence", () => {
+  it("只显示一条主失败摘要，不内联完整 evidence", () => {
     expect(html).toContain("roots-correct");
-    expect(html).toContain("expected x=2, got x=3");
-    expect(html).toContain("judge: sign flipped when substituting into the quadratic formula");
-  });
-
-  it("errored 的 error 摘要", () => {
-    expect(html).toContain("TypeError: cannot read properties of undefined");
+    expect(html).toContain("expected x=2");
+    expect(html).toContain("received x=3");
+    expect(html).not.toContain("judge: sign flipped when substituting into the quadratic formula");
   });
 
   it("total > items.length 时如实报「还有 n 条没列」", () => {
@@ -339,28 +309,14 @@ describe("AttemptList", () => {
     expect(html).not.toContain("more not shown");
   });
 
-  it("每条 attempt 带 locator + 默认证据室深链,不追加字母缩写", () => {
-    expect(html).toContain('href="#/attempt/@1a4a4a4"');
-    expect(html).toContain('href="#/attempt/@1c1c1c1"');
-    expect(html).not.toMatch(/\[[EXD⏱,]+\]/);
-  });
-
-  it("长文本(evidence)收进 <details>,零 JS 可展开", () => {
-    expect(html).toContain("<details");
-    expect(html).toContain("<summary>");
+  it("AttemptList 自身不展开断言详情", () => {
+    expect(html).not.toContain("<details");
   });
 });
 
 describe("EvalList", () => {
   const html = renderToStaticMarkup(<EvalList items={evalListItems} />);
 
-  it("每项一个 experimentId + evalId,判定与分数在场", () => {
-    expect(html).toContain("algebra/quadratic");
-    expect(html).toContain("geometry/angles");
-    expect(html).toContain("compare/bub");
-    expect(html).toContain("compare/codex");
-  });
-
   it("展开到这道题的 Attempt(与 AttemptList 同一套 AttemptRow 渲染)", () => {
     expect(html).toContain('href="#/attempt/@1a4a4a4"');
     expect(html).toContain("roots-correct");
@@ -377,7 +333,7 @@ describe("ExperimentList", () => {
 
   it("web 面是带过滤框与八个固定列头的 experiment 比较表", () => {
     expect(html).toContain('data-nre-experiment-filter=""');
-    for (const header of ["Experiment", "Model", "Agent", "Avg duration", "Pass rate", "Tokens", "Est. cost", "Result"]) {
+    for (const header of ["Experiment", "Model", "Agent", "Avg duration", "End-to-end pass rate", "Tokens", "Est. cost", "Result"]) {
       expect(html).toContain(`>${header}</button>`);
     }
     expect(html).toContain('data-nre-experiment-sort="4" class="nre-sort-desc"');
@@ -438,15 +394,4 @@ describe("跨组件契约", () => {
     ].join("");
     expect(all).not.toContain("<script");
   });
-
-  it("组件源码零 hooks(本实验的「不 hydrate 也完整」用最笨的方式保证)", () => {
-    const dir = dirname(fileURLToPath(import.meta.url));
-    const sources = readdirSync(dir)
-      .filter((f) => (f.endsWith(".tsx") || f.endsWith(".ts")) && !f.includes(".test."))
-      .map((f) => readFileSync(join(dir, f), "utf8"));
-    expect(sources.length).toBeGreaterThanOrEqual(12);
-    for (const src of sources) {
-      expect(src).not.toMatch(/\buse[A-Z][A-Za-z]*\s*\(/);
-    }
-  });
 });

package/src/report/react/styles.css

@@ -61,6 +61,7 @@
 .nre-attempt-list .nre-attempt-experiment,
 .nre-attempt-list .nre-attempt-duration,
 .nre-attempt-list .nre-attempt-cost { color:var(--muted); font-size:.9em; font-variant-numeric:tabular-nums; }
+.nre-attempt-list .nre-attempt-result { display:-webkit-box; overflow:hidden; -webkit-box-orient:vertical; -webkit-line-clamp:2; margin:.4rem 0 0; color:color-mix(in oklch,var(--muted),var(--text) 12%); overflow-wrap:anywhere; }
 .nre-attempt-list .nre-attempt-error { margin:.4rem 0 0; padding:6px 10px; border:1px solid color-mix(in oklch,var(--bad),var(--line) 50%); border-radius:7px; background:color-mix(in oklch,var(--bad),var(--panel) 92%); color:var(--bad); font-family:ui-monospace,SFMono-Regular,Menlo,monospace; font-size:.85em; white-space:pre-wrap; word-break:break-word; }
 .nre-attempt-list .nre-assertions { list-style:none; margin:.4rem 0 0; padding:0; display:grid; gap:6px; }
 .nre-attempt-list .nre-assertion { padding-left:.25rem; }
@@ -75,6 +76,77 @@
 .nre-attempt-list .nre-truncated { color:var(--warn); font-size:.9em; }
 .nre-attempt-list .nre-attempt-list-empty { border:1px dashed var(--line-strong); border-radius:8px; padding:18px; color:var(--muted); text-align:center; background:var(--panel); }
 
+/* ---- ExperimentComparison:完整组索引 + 当前组独立 panel ---- */
+.nre-experiment-comparison .nre-experiment-group-tabs {
+  display:grid;
+  grid-template-columns:repeat(auto-fit,minmax(320px,1fr));
+  gap:12px;
+  margin-bottom:20px;
+}
+.nre-experiment-comparison .nre-experiment-group-tab {
+  min-width:0;
+  border:1px solid var(--line);
+  border-radius:9px;
+  padding:14px 16px 8px;
+  background:var(--panel);
+  cursor:pointer;
+  transition:border-color .12s,background .12s;
+}
+.nre-experiment-comparison .nre-experiment-group-tab:hover {
+  border-color:var(--line-strong);
+  background:color-mix(in oklch,var(--panel-2),var(--panel) 35%);
+}
+.nre-experiment-comparison .nre-experiment-group-tab:focus-visible {
+  outline:2px solid var(--nre-c0);
+  outline-offset:2px;
+}
+.nre-experiment-comparison .nre-experiment-group-tab[aria-selected="true"] {
+  border-color:var(--nre-c0);
+  box-shadow:inset 3px 0 0 var(--nre-c0);
+}
+.nre-experiment-comparison .nre-experiment-group-name {
+  display:block;
+  margin-bottom:10px;
+  color:var(--text);
+  font-size:15px;
+  overflow-wrap:anywhere;
+}
+.nre-experiment-comparison .nre-experiment-group-tab .nre-group-kpis {
+  column-gap:16px;
+}
+.nre-experiment-comparison .nre-experiment-group-tab .nre-group-kpi {
+  min-width:58px;
+}
+.nre-experiment-comparison .nre-experiment-group-panel {
+  border:1px solid var(--line);
+  border-radius:9px;
+  padding:16px;
+  background:var(--panel);
+}
+.nre-experiment-comparison .nre-experiment-group-panel > summary {
+  cursor:pointer;
+  font-size:16px;
+  font-weight:700;
+  overflow-wrap:anywhere;
+}
+.nre-experiment-comparison .nre-experiment-group-panel[open] > summary {
+  margin-bottom:16px;
+}
+.nre-experiment-comparison .nre-experiment-group-panel > .nre {
+  margin:0 0 18px;
+}
+.nre-experiment-comparison .nre-experiment-group-panel > .nre:last-child {
+  margin-bottom:0;
+}
+.nre-experiment-comparison .nre-experiment-groups-empty {
+  border:1px dashed var(--line-strong);
+  border-radius:8px;
+  padding:18px;
+  color:var(--muted);
+  text-align:center;
+  background:var(--panel);
+}
+
 /* ---- EvalList:实体列表的第二级,一条 <details> 一个 experimentId × evalId ---- */
 .nre-eval-list .nre-evals { list-style:none; margin:0; padding:0; }
 .nre-eval-list .nre-eval { border-bottom:1px solid var(--line); }
@@ -88,7 +160,6 @@
 .nre-eval-list .nre-eval-attempts-count,
 .nre-eval-list .nre-eval-duration,
 .nre-eval-list .nre-eval-cost { color:var(--muted); font-size:.9em; font-variant-numeric:tabular-nums; }
-.nre-eval-list .nre-eval-reason { color:color-mix(in oklch,var(--muted),var(--text) 6%); }
 .nre-eval-list .nre-attempts { list-style:none; margin:0 0 12px; padding:0 0 0 20px; border-left:2px solid color-mix(in oklch,var(--nre-c0),transparent 55%); }
 .nre-eval-list .nre-eval-list-empty { border:1px dashed var(--line-strong); border-radius:8px; padding:18px; color:var(--muted); text-align:center; background:var(--panel); }
 
@@ -132,7 +203,7 @@
 .nre-experiment-list .nre-experiment-attempt-row { padding:5px 0 5px 20px; color:var(--muted); }
 .nre-experiment-list .nre-eval-id { font-weight:560; overflow-wrap:anywhere; }
 .nre-experiment-list .nre-eval-attempt-count { color:var(--soft); font-size:12px; }
-.nre-experiment-list .nre-eval-summary { color:var(--muted); font-size:.92em; overflow-wrap:anywhere; }
+.nre-experiment-list .nre-eval-rollup { color:var(--muted); font-size:.92em; overflow-wrap:anywhere; }
 .nre-experiment-list .nre-attempt-branch { color:var(--line-strong); font-variant-numeric:tabular-nums; }
 .nre-experiment-list .nre-eval-attempt-badges { display:inline-flex; flex-wrap:wrap; gap:.6em; }
 .nre-experiment-list .nre-eval-attempt-metrics,