niceeval 0.6.0 → 0.6.1

package/src/report/react/enhance.js

@@ -73,6 +73,45 @@
     }
   });
 
+  // ExperimentList 的 web 面以原生 details 表达「八列主行 + 展开明细」，不是 table/tbody。
+  // 单独按 summary 的列排序、按整条 details 文本过滤；无 JS 时保持数据侧成功率降序。
+  document.addEventListener("click", function (e) {
+    var control = closest(e.target, "[data-nre-experiment-sort]");
+    if (!control) return;
+    var board = control.closest(".nre-experiment-table");
+    if (!board) return;
+    var index = Number(control.getAttribute("data-nre-experiment-sort"));
+    var dir = control.classList.contains("nre-sort-asc") ? "desc" : "asc";
+    var controls = board.querySelectorAll("[data-nre-experiment-sort]");
+    for (var i = 0; i < controls.length; i++) controls[i].classList.remove("nre-sort-asc", "nre-sort-desc");
+    control.classList.add(dir === "asc" ? "nre-sort-asc" : "nre-sort-desc");
+    var entries = Array.prototype.slice.call(board.querySelectorAll(":scope > .nre-experiment-entry"));
+    entries.sort(function (a, b) {
+      var ac = a.querySelector(".nre-experiment-summary").children[index];
+      var bc = b.querySelector(".nre-experiment-summary").children[index];
+      var av = ac ? ac.getAttribute("data-sort-value") || ac.textContent.trim() : "";
+      var bv = bc ? bc.getAttribute("data-sort-value") || bc.textContent.trim() : "";
+      if (av === "" && bv === "") return 0;
+      if (av === "") return 1;
+      if (bv === "") return -1;
+      var an = Number(av), bn = Number(bv);
+      var out = !isNaN(an) && !isNaN(bn) ? an - bn : String(av).localeCompare(String(bv));
+      return dir === "asc" ? out : -out;
+    });
+    for (var j = 0; j < entries.length; j++) board.appendChild(entries[j]);
+  });
+
+  document.addEventListener("input", function (e) {
+    var input = closest(e.target, "input[data-nre-experiment-filter]");
+    if (!input) return;
+    var scope = input.parentElement;
+    var entries = scope ? scope.querySelectorAll(".nre-experiment-entry") : [];
+    var query = input.value.trim().toLowerCase();
+    for (var i = 0; i < entries.length; i++) {
+      entries[i].classList.toggle("nre-row-hidden", query !== "" && entries[i].textContent.toLowerCase().indexOf(query) === -1);
+    }
+  });
+
   // ───────────────────────── tooltip:.nre-scatter-point / .nre-line-point ─────────────────────────
   // 首次 hover 时把点内 <title> 的内容搬进 data-nre-title(避免与原生 tooltip 重影),
   // 渲染样式化 tooltip div(定位在点上方,挂在所属 figure 里)。无 JS 时 <title> 原样生效。

package/src/report/react/fixtures.ts

@@ -330,7 +330,6 @@ const failedAttempt: AttemptListItem = {
   durationMs: 32_000,
   costUSD: 0.12,
   locator: locator("@1a4a4a4"),
-  capabilities: { eval: true, execution: true, timing: true, diff: false },
 };
 
 const erroredAttempt: AttemptListItem = {
@@ -343,7 +342,13 @@ const erroredAttempt: AttemptListItem = {
   assertions: [],
   durationMs: 4_500,
   locator: locator("@1c1c1c1"),
-  capabilities: { eval: false, execution: true, timing: false, diff: false },
+};
+
+const failedRetryAttempt: AttemptListItem = {
+  ...failedAttempt,
+  attempt: 4,
+  durationMs: 35_000,
+  locator: locator("@1b5b5b5"),
 };
 
 export const attemptListItems: AttemptListItem[] = [failedAttempt, erroredAttempt];
@@ -376,13 +381,14 @@ export const experimentListItems: ExperimentListItem[] = [
     experimentId: "compare/bub",
     agent: "bub",
     model: "gpt-5.4",
+    flags: { memory: true },
     verdicts: { passed: 1, failed: 1, errored: 0, skipped: 0 },
     passRate: { value: 0.5, display: "50%", samples: 2, total: 2, refs: [] },
     cost: { value: 0.12, display: "$0.12", samples: 1, total: 2, refs: [failedAttempt.locator] },
     duration: { value: 32_000, display: "32.0s", samples: 2, total: 2, refs: [] },
     tokens: { value: null, display: "—", samples: 0, total: 2, refs: [] },
     evals: 2,
-    attempts: 2,
+    attempts: 3,
     lastRunAt: "2026-07-01T10:00:00Z",
     evalRows: [
       {
@@ -391,7 +397,7 @@ export const experimentListItems: ExperimentListItem[] = [
         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],
+        attempts: [failedAttempt, failedRetryAttempt],
       },
       {
         evalId: "algebra/simple",

package/src/report/react/format.ts

@@ -1,7 +1,7 @@
 // web 面的小工具:展示格式化统一住在计算侧的 ../format.ts(两个渲染面同一份),
 // 这里 re-export 并补 class 名拼接。MetricCell 一律自带 display,组件不重算。
 
-export { MISSING_TEXT, capabilityBadge, formatDurationMs, formatPercent, formatUSD, verdictMark } from "../format.ts";
+export { MISSING_TEXT, formatDurationMs, formatPercent, formatUSD, verdictMark } from "../format.ts";
 
 /** 拼 class 名:过滤空值,末尾接使用者透传的 className。 */
 export function cx(...parts: (string | undefined | false)[]): string {

package/src/report/react/index.tsx

@@ -38,7 +38,6 @@ export type {
 
 // 数据契约类型(家在 ../types.ts,「算」与「画」两侧共用同一份)
 export type {
-  AttemptEvidenceCapabilities,
   AttemptListItem,
   AttemptLocator,
   DeltaData,

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

@@ -271,7 +271,7 @@ describe("MetricScatter", () => {
     expect(empty).toContain("4 points missing data");
   });
 
-  it("恰好 1 个可画点:比较至少要两个实验,不画孤点图", () => {
+  it("恰好 1 个可画点:照常画出单点", () => {
     const single = renderToStaticMarkup(
       <MetricScatter
         data={{
@@ -281,8 +281,8 @@ describe("MetricScatter", () => {
         }}
       />,
     );
-    expect(single).not.toContain("<svg");
-    expect(single).toContain("At least 2 experiments needed to compare");
+    expect(single).toContain("<svg");
+    expect(single).toContain("nre-scatter-point");
   });
 });
 
@@ -339,10 +339,10 @@ describe("AttemptList", () => {
     expect(html).not.toContain("more not shown");
   });
 
-  it("每条 attempt 带 locator + 默认证据室深链 + 证据能力标记", () => {
+  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).toContain("[E,X,⏱]"); // failedAttempt: eval + execution + timing
+    expect(html).not.toMatch(/\[[EXD⏱,]+\]/);
   });
 
   it("长文本(evidence)收进 <details>,零 JS 可展开", () => {
@@ -373,18 +373,34 @@ describe("EvalList", () => {
 });
 
 describe("ExperimentList", () => {
-  const html = renderToStaticMarkup(<ExperimentList items={experimentListItems} />);
+  const html = renderToStaticMarkup(<ExperimentList items={experimentListItems} filter />);
 
-  it("主行:身份、agent/model、官方两级聚合指标", () => {
+  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"]) {
+      expect(html).toContain(`>${header}</button>`);
+    }
+    expect(html).toContain('data-nre-experiment-sort="4" class="nre-sort-desc"');
+  });
+
+  it("主行:身份、agent/model、官方两级聚合指标与判定构成", () => {
     expect(html).toContain("compare/bub");
     expect(html).toContain("compare/codex");
     expect(html).toContain("gpt-5.4");
     expect(html).toContain("50%"); // passRate.display
+    expect(html).toContain("1 passed / 1 failed");
+    expect(html).toContain("memory=true");
   });
 
-  it("展开到 Eval:判定符 + 该题 Attempt 的 locator 徽标内联", () => {
+  it("展开到 Eval:同一道题的全部 Attempt 逐条显示", () => {
     expect(html).toContain("algebra/quadratic");
     expect(html).toContain('href="#/attempt/@1a4a4a4"');
+    expect(html).toContain('href="#/attempt/@1b5b5b5"');
+    expect(html.match(/algebra\/quadratic/g)).toHaveLength(1);
+    expect(html).toContain("nre-experiment-eval-header");
+    expect(html.match(/nre-experiment-attempt-row/g)).toHaveLength(3);
+    expect(html).toContain("├─");
+    expect(html).toContain("└─");
   });
 
   it("零 JS 靠原生 <details>,无 <script>", () => {

package/src/report/react/styles.css

@@ -42,7 +42,7 @@
   margin: 1rem 0;
 }
 
-/* ---- 判定符与证据能力徽标:AttemptList / EvalList / ExperimentList 共用的最小单元 ---- */
+/* ---- locator 与判定符:AttemptList / EvalList / ExperimentList 共用的最小单元 ---- */
 .nre .nre-locator { display:inline-flex; align-items:center; gap:4px; color:var(--nre-c0); font-variant-numeric:tabular-nums; text-decoration:none; white-space:nowrap; }
 .nre .nre-locator:hover { text-decoration:underline; }
 .nre .nre-locator-mark { font-weight:700; }
@@ -50,7 +50,6 @@
 .nre .nre-locator.nre-verdict-failed .nre-locator-mark,
 .nre .nre-locator.nre-verdict-errored .nre-locator-mark { color:var(--bad); }
 .nre .nre-locator.nre-verdict-skipped .nre-locator-mark { color:var(--soft); }
-.nre .nre-locator-caps { color:var(--muted); font-size:.85em; }
 
 /* ---- AttemptList:实体列表的叶子层,一条卡片一个 Attempt ---- */
 .nre-attempt-list .nre-attempts { list-style:none; margin:0; padding:0; }
@@ -93,25 +92,53 @@
 .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); }
 
-/* ---- ExperimentList:实体列表的第一级,一条 <details> 一个 experiment ---- */
-.nre-experiment-list .nre-experiments { list-style:none; margin:0; padding:0; border:1px solid var(--line); border-radius:9px; overflow:hidden; background:var(--panel); }
+/* ---- ExperimentList:web 面固定八列比较表,一条 <details> 一个 experiment ---- */
+.nre-experiment-table { border:1px solid var(--line); border-radius:9px; overflow:hidden; background:var(--panel); }
+.nre-experiment-head,
+.nre-experiment-summary { display:grid; grid-template-columns:minmax(220px,1.25fr) minmax(120px,.8fr) minmax(90px,.6fr) minmax(100px,.65fr) minmax(90px,.55fr) minmax(110px,.7fr) minmax(90px,.6fr) minmax(160px,.9fr); column-gap:16px; align-items:center; }
+.nre-experiment-head { padding:12px 20px; color:var(--muted); font-size:12px; font-weight:560; text-transform:uppercase; letter-spacing:.05em; border-bottom:1px solid var(--line); }
+.nre-experiment-head button { appearance:none; border:0; padding:0; background:transparent; color:inherit; font:inherit; text-transform:inherit; letter-spacing:inherit; text-align:left; cursor:pointer; }
+.nre-experiment-head button::after { content:"↕"; margin-left:6px; color:var(--soft); font-size:11px; }
+.nre-experiment-head button.nre-sort-asc::after { content:"↑"; color:var(--text); }
+.nre-experiment-head button.nre-sort-desc::after { content:"↓"; color:var(--text); }
+.nre-experiment-table-wrap > .nre-filter { height:34px; min-width:260px; margin-bottom:16px; border:1px solid var(--line); border-radius:7px; padding:0 11px; background:var(--panel); color:var(--text); font:inherit; }
 .nre-experiment-list .nre-experiment-entry { border-bottom:1px solid var(--line); }
 .nre-experiment-list .nre-experiment-entry:last-child { border-bottom:0; }
-.nre-experiment-list .nre-experiment-summary { display:flex; flex-wrap:wrap; gap:.9em; align-items:baseline; padding:13px 20px; cursor:pointer; list-style:none; }
+.nre-experiment-list .nre-experiment-summary { padding:13px 20px; cursor:pointer; list-style:none; transition:background .12s; }
 .nre-experiment-list .nre-experiment-summary::-webkit-details-marker { display:none; }
 .nre-experiment-list .nre-experiment-summary:hover { background:color-mix(in oklch,var(--panel-2),var(--panel) 25%); }
-.nre-experiment-list .nre-experiment-id { font-weight:650; overflow-wrap:anywhere; }
-.nre-experiment-list .nre-experiment-agent,
-.nre-experiment-list .nre-experiment-model { color:var(--muted); font-size:.9em; }
-.nre-experiment-list .nre-experiment-cell { font-variant-numeric:tabular-nums; }
-.nre-experiment-list .nre-experiment-meta-sub { flex-basis:100%; color:var(--soft); font-size:12px; }
-.nre-experiment-list .nre-experiment-evals { list-style:none; margin:0; padding:8px 20px 16px; display:grid; gap:6px; background:color-mix(in oklch,var(--panel-2),var(--panel) 24%); }
-.nre-experiment-list .nre-experiment-eval-row { display:flex; flex-wrap:wrap; gap:.6em; align-items:baseline; font-size:13px; }
+.nre-experiment-list .nre-experiment-summary > span { min-width:0; }
+.nre-experiment-list .nre-experiment-name { position:relative; padding-left:30px; }
+.nre-experiment-list .nre-experiment-name::before { content:""; position:absolute; left:5px; top:6px; width:7px; height:7px; border-right:1.5px solid var(--muted); border-bottom:1.5px solid var(--muted); transform:rotate(-45deg); transition:transform .15s; }
+.nre-experiment-list .nre-experiment-entry[open] .nre-experiment-name::before { transform:rotate(45deg); }
+.nre-experiment-list .nre-experiment-id { display:block; overflow-wrap:anywhere; font-weight:650; }
+.nre-experiment-list .nre-experiment-name small { display:block; margin-top:3px; color:var(--muted); font-size:12px; font-weight:400; }
+.nre-experiment-list .nre-num { font-variant-numeric:tabular-nums; font-feature-settings:"tnum"; white-space:nowrap; }
+.nre-experiment-list .nre-good { color:var(--good); }
+.nre-experiment-list .nre-warn { color:var(--warn); }
+.nre-experiment-list .nre-bad { color:var(--bad); }
+.nre-experiment-list .nre-experiment-pill { display:inline-flex; min-height:22px; align-items:center; border:1px solid var(--line); border-radius:999px; padding:0 8px; color:var(--muted); font-size:12px; white-space:nowrap; }
+.nre-experiment-list .nre-experiment-detail { padding:18px 30px 22px; background:color-mix(in oklch,var(--panel-2),var(--panel) 24%); border-top:1px solid var(--line); }
+.nre-experiment-list .nre-experiment-flags { display:flex; flex-wrap:wrap; gap:8px; margin:0 0 14px; color:var(--soft); font-size:12px; }
+.nre-experiment-list .nre-experiment-flags b { border:1px solid color-mix(in oklch,var(--line),transparent 16%); border-radius:999px; padding:3px 8px; color:var(--text); font-weight:560; }
+.nre-experiment-list .nre-experiment-evals { list-style:none; margin:0; padding:0; display:grid; gap:6px; }
+.nre-experiment-list .nre-experiment-eval,
+.nre-experiment-list .nre-experiment-attempts { list-style:none; margin:0; padding:0; }
+.nre-experiment-list .nre-experiment-eval { padding:4px 0 9px; border-bottom:1px solid color-mix(in oklch,var(--line),transparent 28%); }
+.nre-experiment-list .nre-experiment-eval:last-child { border-bottom:0; }
+.nre-experiment-list .nre-experiment-eval-header,
+.nre-experiment-list .nre-experiment-attempt-row { display:grid; grid-template-columns:20px minmax(220px,1fr) minmax(150px,.7fr) minmax(260px,1fr); column-gap:12px; align-items:baseline; font-size:13px; }
+.nre-experiment-list .nre-experiment-eval-header { padding:7px 0 5px; }
+.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-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-avg,
+.nre-experiment-list .nre-eval-attempt-metrics,
 .nre-experiment-list .nre-eval-reason { color:var(--muted); font-size:.92em; }
 .nre-experiment-list .nre-experiment-list-empty { border:1px dashed var(--line-strong); border-radius:8px; padding:18px; color:var(--muted); text-align:center; background:var(--panel); }
+@media(max-width:1000px){.nre-experiment-table{overflow-x:auto}.nre-experiment-head,.nre-experiment-summary{min-width:1040px}.nre-experiment-detail{min-width:1040px}}
 
 /* ---- verdict 判定色(AttemptList/EvalList/ExperimentList 的 verdict 文字与判定符共用) ---- */
 .nre .nre-verdict-passed { color:var(--good); }

package/src/report/report.test.ts

@@ -136,7 +136,7 @@ function selection(snapshots: Snapshot[], warnings: SelectionWarning[]): Selecti
 }
 
 describe("ExperimentList.data", () => {
-  it("身份、flags、Eval 判定构成、官方两级聚合汇总指标,展开到 Eval 携带 locator 与证据能力", async () => {
+  it("身份、flags、Eval 判定构成、官方两级聚合汇总指标,展开到 Eval 携带全部 Attempt locator", async () => {
     const s = snap({
       experimentId: "group/codex-e2b--mempal",
       agent: "codex",
@@ -181,8 +181,6 @@ describe("ExperimentList.data", () => {
     expect(evalA.verdict).toBe("passed"); // 任一轮通过 → 该题通过
     expect(evalA.attempts.map((a) => a.attempt)).toEqual([0, 1]);
     expect(evalA.attempts[0]!.locator).toMatch(/^@1[0-9a-z]{7}$/);
-    expect(evalA.attempts[0]!.capabilities.execution).toBe(true); // attempt0 的 hasEvents:true
-    expect(evalA.attempts[1]!.capabilities.execution).toBe(false);
     // 每次 attempt 独立编码,即使同一道题也不撞车
     expect(evalA.attempts[0]!.locator).not.toBe(evalA.attempts[1]!.locator);
 
@@ -243,7 +241,7 @@ describe("EvalList.data", () => {
 });
 
 describe("AttemptList.data", () => {
-  it("每项一个 Attempt:身份、判定、assertions、耗时、成本、locator 与证据能力标记", async () => {
+  it("每项一个 Attempt:身份、判定、assertions、耗时、成本与 locator", async () => {
     const s = snap({
       experimentId: "exp/x",
       results: [
@@ -270,9 +268,6 @@ describe("AttemptList.data", () => {
     });
     expect(item.assertions).toHaveLength(1);
     expect(item.locator).toMatch(/^@1[0-9a-z]{7}$/);
-    // 证据能力:cheap 计算,不调用完整 loadAttemptEvidence——execution 来自 hasEvents,
-    // eval 来自 hasSources,diff 懒加载(fixture 的 diff() 恒返回 null → false)。
-    expect(item.capabilities).toEqual({ eval: true, execution: true, timing: false, diff: false });
   });
 
   it("redact 作用于 error 与断言 detail/evidence,不作用于身份字段", async () => {

package/src/report/text/faces.ts

@@ -9,7 +9,6 @@ import type {
   AttemptListItem,
   DeltaData,
   EvalListItem,
-  ExperimentListEvalRow,
   ExperimentListItem,
   GroupSummaryData,
   LineData,
@@ -24,7 +23,6 @@ import type { TextContext } from "../tree.ts";
 import type { TableColumn, TableRow } from "../primitives.tsx";
 import {
   attemptItemReason,
-  capabilityBadge,
   formatDurationMs,
   formatMetricValue,
   formatPlainNumber,
@@ -325,11 +323,6 @@ export function scatterText(data: ScatterData, ctx: TextContext): string {
   if (drawable.length === 0) {
     return [localeText(locale, "scatter.noData", axes), ...footnotes].join("\n");
   }
-  // 恰好 1 个可画点:成本 × 通过率的比较至少要两个实验,单点不成图。
-  if (drawable.length === 1) {
-    return [localeText(locale, "scatter.needTwo", axes), ...footnotes].join("\n");
-  }
-
   // 点太密排不下时降级为坐标表,不硬挤
   if (drawable.length > POINT_MARKS.length || ctx.width < 44) {
     const table = renderCoordinateTable(
@@ -451,45 +444,93 @@ export function deltaText(data: DeltaData, ctx: TextContext): string {
 
 // ───────────────────────── 实体列表(ExperimentList / EvalList / AttemptList)─────────────────────────
 //
-// 三面共用的紧凑标记:`locator✓[E,X,⏱]`(判定符紧跟 locator,证据能力方括号紧跟判定符,
-// 中间不留空格)——docs-site/zh/guides/report-components.mdx「终端输出形成反馈闭环」定的形态。
+// 三面共用的紧凑标记:`locator✓`(判定符紧跟 locator,中间不留空格)。
 // ExperimentList / EvalList 逐 attempt 只列这一个标记 + 各自的原因/耗时摘要,不重复整段
 // niceeval show 命令;要看某个 attempt 的完整证据,agent 自己拼 `niceeval show <locator>`——
 // 命令模板只在 AttemptList(叶子层)展示完整断言明细时才值得,不在中间层重复。
 
-function locatorBadge(item: { locator: string; verdict: AttemptListItem["verdict"]; capabilities: AttemptListItem["capabilities"] }): string {
-  return `${item.locator}${verdictMark(item.verdict)}${capabilityBadge(item.capabilities)}`;
+function locatorBadge(item: { locator: string; verdict: AttemptListItem["verdict"] }): string {
+  return `${item.locator}${verdictMark(item.verdict)}`;
 }
 
 // ── ExperimentList ──
 
-function experimentListEvalLine(row: ExperimentListEvalRow): string {
-  const badges = row.attempts.map(locatorBadge).join(" ");
-  const trailer =
-    row.verdict === "passed"
-      ? [formatDurationMs(row.duration.value ?? 0), row.cost.value === null ? undefined : formatUSD(row.cost.value)]
-          .filter((s): s is string => s !== undefined)
-          .join(" · ")
-      : (row.reason ?? "");
-  return `  ${verdictMark(row.verdict)} ${row.evalId}   ${badges}   ${trailer}`;
+function experimentSummaryTable(items: ExperimentListItem[], ctx: TextContext): string {
+  const locale = ctx.locale;
+  const compact = ctx.width < 100;
+  const columns: TableColumn[] = [
+    { key: "experiment", header: compact && locale === "en" ? "Exp." : localeText(locale, "experimentList.experiment") },
+    { key: "model", header: localeText(locale, "table.model") },
+    { key: "agent", header: localeText(locale, "table.agent") },
+    { key: "duration", header: compact && locale === "en" ? "Avg" : localeText(locale, "experimentList.avgDuration"), align: "right" },
+    { key: "passRate", header: compact && locale === "en" ? "Pass" : localeText(locale, "experimentList.passRate"), align: "right" },
+    { key: "result", header: localeText(locale, "experimentList.result") },
+    { key: "tokens", header: localeText(locale, "experimentList.tokens"), align: "right" },
+    { key: "cost", header: compact && locale === "en" ? "Cost" : localeText(locale, "experimentList.estimatedCost"), align: "right" },
+  ];
+  const rows: TableRow[] = items.map((item) => ({
+    key: item.experimentId,
+    cells: {
+      experiment: item.experimentId,
+      model: item.model ?? localeText(locale, "experimentList.defaultModel"),
+      agent: item.agent,
+      duration: cellText(item.duration),
+      passRate: cellText(item.passRate),
+      result: verdictTallyText(item.verdicts, locale),
+      tokens: cellText(item.tokens),
+      cost: cellText(item.cost),
+    },
+  }));
+  const metadata = items.flatMap((item) =>
+    wrapDisplay(
+      `${item.experimentId}: ${localeText(locale, "overview.evalsCount", { n: item.evals })} · ${localeText(locale, "overview.attemptsCount", { n: item.attempts })} · ${item.lastRunAt}`,
+      Math.max(8, ctx.width - 2),
+    ).map((line) => `  ${line}`),
+  );
+  return [renderTableText({ columns, rows, locale }, ctx), metadata.join("\n")].join("\n");
 }
 
-export function experimentListText(items: ExperimentListItem[], ctx: TextContext): string {
+function experimentDetailTable(item: ExperimentListItem, ctx: TextContext): string {
   const locale = ctx.locale;
-  if (items.length === 0) return localeText(locale, "attemptList.empty");
-  const blocks = items.map((item) => {
-    const identity = item.model ? `${item.experimentId} · ${item.agent} · ${item.model}` : `${item.experimentId} · ${item.agent}`;
-    const summary = [
-      `${localeText(locale, "overview.passRate")} ${cellText(item.passRate)}`,
-      verdictTallyText(item.verdicts, locale),
-      localeText(locale, "overview.attemptsCount", { n: item.attempts }),
-      formatDurationMs(item.duration.value ?? 0),
-      item.cost.value === null ? missingText(locale) : formatUSD(item.cost.value),
-    ].join(" · ");
-    const evalLines = item.evalRows.map(experimentListEvalLine);
-    return [identity, `  ${summary}`, ...evalLines].join("\n");
+  const columns: TableColumn[] = [
+    { key: "status", header: localeText(locale, "experimentList.status") },
+    { key: "entity", header: localeText(locale, "experimentList.evalAttempt") },
+    { key: "result", header: localeText(locale, "experimentList.result") },
+    { key: "duration", header: localeText(locale, "experimentList.duration"), align: "right" },
+    { key: "cost", header: localeText(locale, "experimentList.cost"), align: "right" },
+  ];
+  const rows: TableRow[] = item.evalRows.flatMap((row) => {
+    const parent: TableRow = {
+      key: row.evalId,
+      cells: {
+        status: `${verdictMark(row.verdict)} ${localeText(locale, `verdict.${row.verdict}`)}`,
+        entity: row.evalId,
+        result: "",
+        duration: "",
+        cost: "",
+      },
+    };
+    const attempts: TableRow[] = row.attempts.map((attempt, index) => ({
+      key: attempt.locator,
+      cells: {
+        status: `  ${verdictMark(attempt.verdict)}`,
+        entity: `${index === row.attempts.length - 1 ? "└─" : "├─"} ${attempt.locator}`,
+        result: attemptItemReason(attempt) ?? MISSING_MARK,
+        duration: attempt.verdict === "skipped" && attempt.durationMs === 0 ? null : formatDurationMs(attempt.durationMs),
+        cost: attempt.costUSD === undefined ? null : formatUSD(attempt.costUSD),
+      },
+    }));
+    return [parent, ...attempts];
   });
-  return blocks.join("\n\n");
+  const flags = item.flags && Object.keys(item.flags).length > 0
+    ? `${localeText(locale, "experimentList.flags")} ${Object.entries(item.flags).map(([key, value]) => `${key}=${String(value)}`).join(" · ")}`
+    : undefined;
+  return [item.experimentId, flags, renderTableText({ columns, rows, locale }, ctx)].filter(Boolean).join("\n");
+}
+
+export function experimentListText(items: ExperimentListItem[], ctx: TextContext): string {
+  if (items.length === 0) return localeText(ctx.locale, "attemptList.empty");
+  return [experimentSummaryTable(items, ctx), ...items.map((item) => experimentDetailTable(item, ctx))].join("\n\n");
 }
 
 // ── EvalList ──
@@ -518,7 +559,7 @@ export function evalListText(items: EvalListItem[], ctx: TextContext): string {
 
 // ── AttemptList ──
 
-/** 一个 AttemptListItem 的完整 text 卡片:判定符 + locator + 身份 + 耗时/成本 + 证据能力,
+/** 一个 AttemptListItem 的完整 text 卡片:判定符 + locator + 身份 + 耗时/成本,
  * 然后逐条断言(gate 与 soft 都列,与 web 面的 AttemptRow 同一份材料)。 */
 function attemptListItemText(item: AttemptListItem, ctx: TextContext, locale: ReportLocale): string {
   const head = [
@@ -527,7 +568,6 @@ function attemptListItemText(item: AttemptListItem, ctx: TextContext, locale: Re
     item.experimentId,
     formatDurationMs(item.durationMs),
     ...(item.costUSD !== undefined ? [formatUSD(item.costUSD)] : []),
-    ...(capabilityBadge(item.capabilities) ? [capabilityBadge(item.capabilities)] : []),
   ].join(" · ");
   const lines = [head];
   if (item.error) {

package/src/report/types.ts

@@ -4,12 +4,11 @@
 
 import type { AttemptHandle, SelectionWarning } from "../results/types.ts";
 import type { AttemptLocator } from "../results/locator.ts";
-import type { AttemptEvidenceCapabilities } from "../results/attempt-evidence.ts";
 import type { AssertionResult, Verdict } from "../types.ts";
 import type { LocalizedLabel, ReportLocale } from "./locale.ts";
 
 export type { SelectionWarning };
-export type { AttemptLocator, AttemptEvidenceCapabilities };
+export type { AttemptLocator };
 export type { LocalizedLabel, ReportLocale };
 
 // ───────────────────────── 指标与聚合 ─────────────────────────
@@ -336,10 +335,8 @@ export interface DeltaData<K extends string = string> {
 // `EvalListItem.attempts` 的元素,报告作者可以直接把这些嵌套数组喂给 `<AttemptList items={...} />`。
 
 /**
- * `AttemptList` 一项 = 一个 Attempt:身份、判定、断言、error、耗时、成本、locator,外加
- * 证据能力标记(有没有保存的 Eval 源码 / 执行事件 / OTel 计时 / diff——与 `AttemptEvidence.capabilities`
- * 同一个类型,不重新发明四个布尔位的含义)。`ExperimentList` / `EvalList` 的下钻数组复用同一个
- * 类型,不是各自的精简版。
+ * `AttemptList` 一项 = 一个 Attempt:身份、判定、断言、error、耗时、成本和 locator。
+ * `ExperimentList` / `EvalList` 的下钻数组复用同一个类型,不是各自的精简版。
  */
 export interface AttemptListItem {
   evalId: string;
@@ -352,15 +349,6 @@ export interface AttemptListItem {
   durationMs: number;
   costUSD?: number;
   locator: AttemptLocator;
-  /**
-   * 证据能力标记:`eval`(运行时 Eval 源码已保存)、`execution`(执行事件非空)、
-   * `timing`(执行事件 + 这次运行接入过 OTel)、`diff`(工作区有文件改动)。逐位定义与
-   * `AttemptEvidence.capabilities` 完全一致——这里复用同一个类型,不重复声明四个布尔位的门槛。
-   * 计算函数从 `EvalResult` 上已有的 `hasEvents` / `hasSources` / `hasTrace` 摘要位与一次
-   * `attempt.diff()` 懒加载算出,不对每个 item 调用完整的 `loadAttemptEvidence`
-   * (那还会额外装配 Eval 源码标注与 ExecutionTree,这里只要四个布尔位)。
-   */
-  capabilities: AttemptEvidenceCapabilities;
 }
 
 /**

package/src/results/host-equivalence.test.ts

@@ -418,7 +418,7 @@ describe("宿主接线 · show text 面与 view web 面反映同一批事实", (
       expect(face).toContain("compare/bub");
       expect(face).toContain("q2"); // 周一补齐的失败 eval,若宿主回退 results.latest() 就会消失
     }
-    expect(text).toContain("1 passed · 1 failed");
+    expect(text).toContain("1 passed / 1 failed");
     expect(html).toContain("50%");
     // 现刻水位覆盖齐全 → 两面都不出 partial-coverage(text: "verdicts cover"; html: data-kind)。
     expect(hasPartialCoverageText(text)).toBe(false);
@@ -493,7 +493,7 @@ function hasPartialCoverageHtml(html: string): boolean {
   return html.includes('data-kind="partial-coverage"');
 }
 
-describe("默认宿主事实对照 · show 索引与 view 报告同口径", () => {
+  describe("默认宿主事实对照 · show / view 渲染同一报告的两面", () => {
   it("通过计数一致:同一 fixture 两面反映 3 过 1 败", async () => {
     const root = await makeRoot();
     await writeSnapshot(root, "2026-07-01T08-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-01T08:00:00.000Z" }, [
@@ -504,7 +504,7 @@ describe("默认宿主事实对照 · show 索引与 view 报告同口径", () =
     ]);
     const text = await showText(root, []);
     const html = await viewHtml(root);
-    expect(text).toContain("3 passed · 1 failed");
+    expect(text).toContain("3 passed / 1 failed");
     expect(html).toContain("75%");
     for (const face of [text, html]) expect(face).toContain("compare/bub");
     // 无残缺:两面都不出 partial-coverage,布尔一致(某面偷偷补/漏警告即失配)。

package/src/show/index.ts

@@ -2,7 +2,7 @@
 // 宿主组合语义:docs/feature/reports/architecture.md「Selection 是计算入口」)。
 //
 // 位置参数 = eval id 前缀,或 `@<locator>`(精确指名单个 attempt,见 results/locator.ts):
-//   裸跑 / 前缀       show 专用的紧凑 attempt 索引(单 eval 前缀仍进入详情)
+//   裸跑 / 多 eval 前缀  默认 ExperimentComparison 的 text 面(单 eval 前缀仍进入详情)
 //   恰好一个 eval     单 eval 详情(attempt / 断言明细,宿主本体)
 //   @<locator>        精确 attempt:无证据 flag → 紧凑全景;带 flag → 对应证据切面
 //   --eval / --execution / --diff[=路径]   证据切面(宿主本体):出现即走证据室,不渲染报告槽
@@ -11,7 +11,7 @@
 //   --run <目录>     结果根换成该目录;--experiment Selection 只留该实验
 //
 // 数据全部走 niceeval/results 的读取面(openResults + 合成 Selection + loadAttemptEvidence),
-// 不自己爬目录、不自己重算证据 capability。
+// 不自己爬目录；证据可用性只由 loadAttemptEvidence 在单 Attempt 页面计算。
 
 import { existsSync } from "node:fs";
 import { join, resolve } from "node:path";
@@ -24,7 +24,7 @@ import {
   MalformedLocatorError,
 } from "../results/index.ts";
 // report.ts / load.ts have no JSX of their own, but ReportDefinition / ReportLoadError must
-// come from the SAME module instance CostPassRateComparison (built-ins, .tsx) is built
+// come from the SAME module instance ExperimentComparison (built-ins, .tsx) is built
 // against — `unique symbol` branding and `instanceof` are keyed by declaration site, so a raw
 // src copy and the compiled dist copy of "the same" ReportDefinition are, to TypeScript and to
 // `instanceof`, two different types. The package-owned report runtime ships as precompiled
@@ -32,6 +32,7 @@ import {
 // when niceeval is linked in — see tsconfig.report-build.json); show pulls all of it from
 // there, not just the .tsx-touching pieces.
 import { renderReportToText } from "../../dist/report/report.js";
+import { ExperimentComparison } from "../../dist/report/built-ins/index.js";
 import { ReportLoadError, loadReportFile } from "../../dist/report/load.js";
 import { detectLocale, t } from "../i18n/index.ts";
 import { foldEvalVerdict } from "../shared/verdict.ts";
@@ -43,7 +44,6 @@ import {
   attemptIndexLine,
   attemptOverviewText,
   attemptsOfEval,
-  cheapCapabilities,
   diffText,
   evalDetailText,
   evalHistoryText,
@@ -52,7 +52,6 @@ import {
   experimentHistoryText,
   pickDetailAttempt,
   skippedRunsText,
-  showIndexText,
   verdictReasonLine,
 } from "./render.ts";
 
@@ -202,7 +201,7 @@ async function show(
   // 证据切面是宿主本体:出现即走证据室,不渲染报告槽(与默认报告同规则)。
   if (evidence) {
     if (matchedEvalIds.length !== 1) {
-      // 撞多个 eval 时不止说「有几个」,直接给紧凑索引(locator + capability 字母 + 失败原因)
+      // 撞多个 eval 时不止说「有几个」,直接给紧凑索引(locator + 失败原因)
       // 让 agent 一步摘到 `@<locator>`,不必再跑一轮 `show <eval id>` 才知道选谁。
       const index = matchedEvalIds
         .map((evalId) => {
@@ -213,7 +212,6 @@ async function show(
             evalId,
             verdict,
             locator: rep?.locator,
-            capabilities: rep ? cheapCapabilities(rep.result) : { eval: false, execution: false, timing: false, diff: false },
             reason: rep ? verdictReasonLine(rep.result) : undefined,
           });
         })
@@ -277,16 +275,11 @@ async function show(
     return;
   }
 
-  // 裸 show 是终端诊断索引(docs/feature/reports/show.md),不是 web 默认分析报告的 text 面。
-  // --report 才进入可替换报告槽；view 仍可选择 CostPassRateComparison 作为自己的默认面。
-  // locale = CLI 界面语言(NICEEVAL_LANG / LC_* / LANG 检测):报告 chrome 文案跟随
+  // 裸 show 与裸 view 选择同一个普通默认 definition,这里只是渲染 text 面。
+  // --report 替换同一个报告槽。locale = CLI 界面语言(NICEEVAL_LANG / LC_* / LANG 检测):报告 chrome 文案跟随
   // 终端语言(docs/feature/reports/library.md「locale:渲染面的语言」);Locale 与 ReportLocale 同为
   // "en" | "zh-CN",直接传递。
-  if (flags.report === undefined) {
-    io.out(showIndexText(selection, io.width) + "\n");
-    return;
-  }
-  const definition = await loadReportFile(cwd, flags.report);
+  const definition = flags.report === undefined ? ExperimentComparison : await loadReportFile(cwd, flags.report);
   // attemptCommand 留给 renderReportToText 的默认值:AttemptLocator 已经是可直接 `niceeval show
   // @<locator>` 的真实 CLI 语法,不需要再反查 eval id 拼一条近似命令(见 tree.ts 的默认实现)。
   const text = await renderReportToText(definition, { selection, results }, { width: io.width, locale: detectLocale() });