niceeval 0.9.1 → 0.10.0

package/src/shared/aggregate.test.ts

@@ -0,0 +1,52 @@
+// cases: docs/engineering/unit-tests/experiments-runner/cases.md
+import { describe, expect, it } from "vitest";
+import { matchExperimentSelector } from "./aggregate.ts";
+
+describe("matchExperimentSelector", () => {
+  const ids = [
+    "compare/bub-gpt-5.6-luna",
+    "compare/codex-gpt-5.6-luna",
+    "compare/codex-gpt-5.6-luna--agents-md",
+    "compare/codex-gpt-5.6-luna--mempal",
+    "dev-e2b/bub-e2b",
+    "dev/bub-gpt-5.4-mini",
+    "top-level",
+  ];
+
+  it("selects an entire group by exact directory match", () => {
+    expect(matchExperimentSelector(ids, "compare")).toEqual([
+      "compare/bub-gpt-5.6-luna",
+      "compare/codex-gpt-5.6-luna",
+      "compare/codex-gpt-5.6-luna--agents-md",
+      "compare/codex-gpt-5.6-luna--mempal",
+    ]);
+  });
+
+  it("an exact id match wins even when it is a prefix of sibling variants", () => {
+    expect(matchExperimentSelector(ids, "compare/codex-gpt-5.6-luna")).toEqual([
+      "compare/codex-gpt-5.6-luna",
+    ]);
+  });
+
+  it("falls back to a filename prefix within an exact-matched directory to select a family", () => {
+    expect(matchExperimentSelector(ids, "compare/codex")).toEqual([
+      "compare/codex-gpt-5.6-luna",
+      "compare/codex-gpt-5.6-luna--agents-md",
+      "compare/codex-gpt-5.6-luna--mempal",
+    ]);
+  });
+
+  it("requires an exact directory segment; a group-like prefix does not leak across groups", () => {
+    expect(matchExperimentSelector(ids, "dev")).toEqual(["dev/bub-gpt-5.4-mini"]);
+    expect(matchExperimentSelector(ids, "dev")).not.toContain("dev-e2b/bub-e2b");
+  });
+
+  it("matches a top-level (groupless) experiment by exact id", () => {
+    expect(matchExperimentSelector(ids, "top-level")).toEqual(["top-level"]);
+  });
+
+  it("returns empty when nothing matches", () => {
+    expect(matchExperimentSelector(ids, "dev-e3b")).toEqual([]);
+    expect(matchExperimentSelector(ids, "compare/nope")).toEqual([]);
+  });
+});

package/src/shared/aggregate.ts

@@ -55,6 +55,35 @@ export function evalPrefixPredicate(evals?: string | string[]): (id: string) =>
   return (id) => prefixes.some((prefix) => id.startsWith(prefix));
 }
 
+/**
+ * 实验选择器解析:`niceeval exp` 位置参数与 `--exp` 共用同一条规则(docs/feature/experiments/cli.md
+ * 「实验选择器怎样解析」)。按序:①精确 id 优先,即使它同时是同目录内其它文件名的前缀;
+ * ②目录段精确前缀(含更深层目录),选中该目录下全部实验;③以上都不命中且选择器形如
+ * `目录/文件名前缀` 时,目录段精确匹配后按文件名段前缀选中一族配置——目录段永远精确,不跨目录
+ * 误配,文件名段裸前缀,与 `evalPrefixPredicate` 同一条"共享前缀即家族"的逻辑。
+ */
+export function matchExperimentSelector(ids: readonly string[], selector: string): string[] {
+  const exact = ids.find((id) => id === selector);
+  if (exact !== undefined) return [exact];
+  const dirMatches = ids.filter((id) => id.startsWith(selector + "/"));
+  if (dirMatches.length > 0) return dirMatches;
+  const lastSlash = selector.lastIndexOf("/");
+  if (lastSlash === -1) return [];
+  const dir = selector.slice(0, lastSlash);
+  const namePrefix = selector.slice(lastSlash + 1);
+  if (namePrefix === "") return [];
+  return ids.filter((id) => id.startsWith(dir + "/") && id.slice(dir.length + 1).startsWith(namePrefix));
+}
+
+/**
+ * 零命中时展示的可浏览路径清单(docs/feature/experiments/cli.md「零命中」):按 id 的顶层目录段
+ * 去重、排序,带尾斜杠;没有目录段的顶层 experiment id 原样列出(它本身就是一个可选择的目标,
+ * 不是一段路径)。这是运行选择的展示投影,不是报告分组。
+ */
+export function browsableExperimentPaths(ids: readonly string[]): string[] {
+  return [...new Set(ids.map((id) => (id.includes("/") ? `${id.split("/")[0]}/` : id)))].sort();
+}
+
 /** 无 experimentId 时的兜底标签。 */
 export function fallbackExperimentLabel(result: {
   experimentId?: string;

package/src/shared/types.ts

@@ -30,7 +30,12 @@ export interface SourceArtifact {
   content: string;
 }
 
-/** 通用清理闭包(setup 返回值 / teardown 的形状),异步同步皆可,统一在 finally 里执行。 */
+/**
+ * 内部收尾闭包类型:供运行器内部的清理注册表使用(如 `postSetup` 钩子内部累积待收尾动作),
+ * 不出现在任何公开的生命周期钩子签名里——`SandboxHook`/`AgentSetup`/`AgentTeardown` 等公开
+ * 钩子一律 `void | Promise<void>`,setup 不返回值,收尾靠成对的 teardown(见
+ * docs/runner.md「环境预置不进运行器,但按顺序调它」)。
+ */
 export type Cleanup = () => Promise<void> | void;
 
 /** `ScopedFeedback.progress` 的入参:此刻正在做什么(短命状态,可被后续更新覆盖)。 */

package/src/show/command.test.ts

@@ -0,0 +1,34 @@
+// cases: docs/engineering/unit-tests/reports/cases.md
+// 「外壳、页面与 Tabs」——show 专属的索引命令拼装(showCommand)与其余页索引文案(otherPagesText)。
+
+import { describe, expect, it } from "vitest";
+import { showCommand } from "./command.ts";
+import { otherPagesText } from "./render.ts";
+
+describe("其余页索引与索引命令上下文", () => {
+  const pages = [
+    { id: "overview", title: { en: "Overview", "zh-CN": "总览" } },
+    { id: "exam", title: { en: "Exam", "zh-CN": "成绩单" } },
+  ];
+
+  it("只列未渲染的页,索引命令保留当前 --results / --report 与位置参数,复制即可复现下一层视图", () => {
+    // 渲染的是 overview,其余页索引只含 exam 一行——与「渲染初始页 + 尾部附其余页索引」
+    // 的行为一致(docs/feature/reports/show/reports.md Case 2)。
+    const text = otherPagesText({
+      otherPages: pages.filter((p) => p.id !== "overview"),
+      command: { patterns: [], results: "tmp/published-results", report: "reports/site.tsx" },
+      locale: "zh-CN",
+    });
+    expect(text).toContain("其余页：");
+    expect(text).toContain("niceeval show --results tmp/published-results --report reports/site.tsx --page exam");
+    expect(text).toContain("成绩单");
+    expect(text).not.toContain("总览");
+    expect(text).not.toContain("--page overview");
+  });
+
+  it("showCommand 按序携带位置参数与 --exp / --results / --report / --page", () => {
+    expect(
+      showCommand({ patterns: ["memory/swelancer"], experiment: "dev-e2b", report: "reports/site.tsx", page: "exam" }),
+    ).toBe("niceeval show memory/swelancer --exp dev-e2b --report reports/site.tsx --page exam");
+  });
+});

package/src/show/command.ts

@@ -0,0 +1,16 @@
+// show 专属的可复制命令拼装:页索引 / 组索引共用的携带规则(docs/feature/reports/show/reports.md
+// 「索引命令携带完整上下文」)。这不是两宿主共用的报告 runtime 部分——view 走网页路由,不生成
+// 终端命令——所以住在 show 自己这一侧,不进 src/report/runtime/host.ts。
+
+import type { HostCommandContext } from "../report/runtime/host.ts";
+
+/** 按上下文拼一条可复制的 show 命令(页索引 / 组索引共用的携带规则)。 */
+export function showCommand(ctx: HostCommandContext, extra: string[] = []): string {
+  const parts = ["niceeval show", ...ctx.patterns];
+  if (ctx.experiment !== undefined) parts.push(`--exp ${ctx.experiment}`);
+  if (ctx.results !== undefined) parts.push(`--results ${ctx.results}`);
+  if (ctx.report !== undefined) parts.push(`--report ${ctx.report}`);
+  if (ctx.page !== undefined) parts.push(`--page ${ctx.page}`);
+  parts.push(...extra);
+  return parts.join(" ");
+}

package/src/show/compose.ts

@@ -6,7 +6,7 @@
 // 现刻水位 Scope(两个宿主共用)住在 ../results/select.ts;本文件只留 show 独有的时间轴计算。
 // 数据只消费 niceeval/results 的读取面。
 
-import { attemptCostUSD } from "../report/metrics.ts";
+import { attemptCostUSD } from "../report/model/metrics.ts";
 import { compactAssertionSummary, primaryAssertionSummary, summaryText } from "../scoring/display.ts";
 import type { EvalResult, Verdict } from "../types.ts";
 import type { AttemptHandle, Experiment } from "../results/index.ts";

package/src/show/index.ts

@@ -4,7 +4,8 @@
 // 位置参数 = eval id 前缀,或 `@<locator>`(精确指名单个 attempt,见 results/locator.ts):
 //   裸跑 / 多 eval 前缀  内建报告(niceeval/report/built-in 默认导出)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
+//   @<locator>        精确 attempt:无证据 flag → 当前 report 的 attempt-input page(内建 standard
+//                     或 --report 自定义);带 flag → 对应证据切面(宿主本体,不经报告管线)
 //   --source / --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
 //   --history        执行时间轴(逐 experimentId + evalId 分节),与 --report 互斥
 //   --report <文件>  整槽换成用户报告;位置前缀 / --results / --exp 先收窄 Scope 再注入
@@ -28,26 +29,25 @@ import {
 // against — `instanceof` is keyed by declaration site, so a raw src copy and the compiled
 // dist copy of "the same" class are two different types. The package-owned report runtime
 // ships as precompiled ESM (dist/report/**, built by `pnpm run build:report`); all report
-// loading/rendering goes through ./report-host.ts (the single contact surface with it).
-import { ReportLoadError } from "../../dist/report/load.js";
+// loading/rendering goes through ../report/runtime/host.ts (the shared contact surface).
+import { ReportLoadError } from "../../dist/report/runtime/load.js";
 import { detectLocale, t } from "../i18n/index.ts";
 import { foldEvalVerdict } from "../shared/verdict.ts";
 import { selectCurrentResults, filterExperiments } from "../results/select.ts";
 import { evalPrefixPredicate } from "../shared/aggregate.ts";
 import { attemptHistory } from "./compose.ts";
 import {
+  buildHostReportMeta,
   HostReportError,
   loadHostReport,
-  reportMetaFor,
   renderHostPageText,
   type HostCommandContext,
-} from "./report-host.ts";
+} from "../report/runtime/host.ts";
 import {
   attemptArtifactsPath,
   attemptEvidenceHeader,
   attemptHistoryText,
   attemptIndexLine,
-  attemptOverviewText,
   attemptsOfEval,
   diffText,
   evalDetailText,
@@ -96,8 +96,9 @@ function clampWidth(columns: number | undefined): number {
   return Math.max(40, Math.min(columns as number, 160));
 }
 
-// --report 的装载与规范化住在 ./report-host.ts(两个宿主共用的唯一联系面)。
-export { loadHostReport, normalizeHostReport, resolveReportTitle, localizeText } from "./report-host.ts";
+// --report 的装载住在 ../report/runtime/host.ts(两个宿主共用的中性联系面);规范化本身是
+// `defineReport` 自己的职责,不在宿主层重复。
+export { loadHostReport, localizeText } from "../report/runtime/host.ts";
 
 export async function runShow(
   cwd: string,
@@ -163,8 +164,7 @@ async function show(
   // `@<locator>` 位置参数:身份直达单个 attempt,与 eval id 前缀匹配完全不同的语义
   // (`@` 打头对 eval id 天然无歧义,见 locator.ts),必须在下面的前缀匹配逻辑之前分流掉,
   // 不然 "@1x7f3q" 会被当成一个谁都匹配不到的 eval id 前缀,报「no eval match」这种文不对题的
-  // 错误。这一步只解析并渲染出「当前 show 对单个已解析 attempt 能渲染的东西」(单 eval 详情 /
-  // 三个证据切面)——真正的 `--source`/`--execution`/`--diff` 统一 attempt 全景是后续阶段。
+  // 错误。
   const locatorArg = patterns.find((p) => p.startsWith(ATTEMPT_LOCATOR_PREFIX));
   if (locatorArg !== undefined) {
     if (patterns.length !== 1) {
@@ -199,7 +199,35 @@ async function show(
       io.out(blocks.join("\n\n") + "\n");
       return;
     }
-    io.out(attemptOverviewText(attemptEvidence, { header, artifactPath, width: io.width }) + "\n");
+    // 无证据 flag:选中当前 report definition 里唯一的 attempt-input page,注入这份 evidence,
+    // 走与其它 page 完全相同的 resolve → validate → render 管线(docs/feature/reports/show/attempt.md;
+    // docs/feature/reports/library/attempt-detail.md「在 show 与 view 怎样渲染」)。不带 --report
+    // 时装载内建 standard,其中就带这张 page;--report 指向的自定义报告没有声明 attempt-input page
+    // 时报完整用户反馈,不回退到内建详情(三条解决路径都在错误文案里给出)。
+    const report = await loadHostReport(cwd, flags.report);
+    const attemptPage = report.pages.find((p) => p.input === "attempt");
+    if (attemptPage === undefined) {
+      const sourceLabel = flags.report ?? "the built-in report";
+      throw new ShowError(
+        `error: ${sourceLabel} has no attempt-input page — "${locatorArg}" cannot be opened without one. ` +
+          `Add one: use \`extends: standard\` (inherits its attempt page), import { standardAttemptPage } from ` +
+          `"niceeval/report/built-in" and add it to your pages list, or declare your own \`input: "attempt"\` page.\n`,
+      );
+    }
+    const locale = detectLocale();
+    const selection = selectCurrentResults(results, {});
+    const meta = await buildHostReportMeta(report, selection);
+    const text = await renderHostPageText(
+      attemptPage,
+      {
+        scope: selection,
+        results,
+        report: meta,
+        page: { id: attemptPage.id, input: "attempt", locator: attempt.locator!, evidence: attemptEvidence },
+      },
+      { width: io.width, locale },
+    );
+    io.out(text + "\n");
     return;
   }
 
@@ -322,16 +350,23 @@ async function show(
   };
   const sourceLabel = flags.report ?? "the built-in report";
 
-  // 初始页 = --page 指定的页,缺省第一页(docs/feature/reports/show/reports.md Case 2);
-  // 本地宿主只 resolve 被打开的这一页——其余页只留 id / title,不触发取数(见 shell.md
-  // 「行为约束」「本地宿主只 resolve 被打开的页」)。
-  let page = report.pages[0];
+  // 初始页 = --page 指定的页,缺省第一张可导航页(docs/feature/reports/show/reports.md
+  // Case 2);本地宿主只 resolve 被打开的这一页——其余页只留 id / title,不触发取数(见
+  // shell.md「行为约束」「本地宿主只 resolve 被打开的页」)。navigation:false 的页(参数化
+  // attempt 详情)不参与缺省选择,也不能被 --page 直接打开——没有 locator 不能拿 Scope 强行
+  // resolve(architecture.md「Attempt 详情是一张参数化 page」)。
+  let page = report.pages.find((p) => p.navigation !== false) ?? report.pages[0];
   if (flags.page !== undefined) {
     const hit = report.pages.find((p) => p.id === flags.page);
     if (!hit) {
       // 用法错误:列出可用页 id(docs/feature/reports/show/reports.md Case 1/2 的报错样例)。
       throw new ShowError(
-        `error: page "${flags.page}" not found in ${sourceLabel}. Available pages: ${report.pages.map((p) => p.id).join(", ")}\n`,
+        `error: page "${flags.page}" not found in ${sourceLabel}. Available pages: ${report.pages.filter((p) => p.navigation !== false).map((p) => p.id).join(", ")}\n`,
+      );
+    }
+    if (hit.input === "attempt") {
+      throw new ShowError(
+        `error: page "${hit.id}" in ${sourceLabel} is an attempt-input page and needs a locator — it cannot be opened with --page directly. Use niceeval show @<locator> instead.\n`,
       );
     }
     page = hit;
@@ -339,10 +374,10 @@ async function show(
 
   // attemptCommand 留给渲染管线的默认值: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。
-  const meta = reportMetaFor(report, page.id, selection.snapshots);
+  const meta = await buildHostReportMeta(report, selection);
   const text = await renderHostPageText(
     page,
-    { scope: selection, results, report: meta },
+    { scope: selection, results, report: meta, page: { id: page.id, input: "scope" } },
     {
       width: io.width,
       locale,
@@ -350,8 +385,9 @@ async function show(
     },
   );
 
-  // 页数大于一时尾部附「其余页」索引(只列未渲染的页,不倾倒内容);单页定义没有这段。
-  const remaining = report.pages.filter((p) => p.id !== page.id);
+  // 页数大于一时尾部附「其余页」索引(只列未渲染、且可导航的页,不倾倒内容);单页定义
+  // 没有这段;隐藏的 attempt page 不出现在「其余页」里。
+  const remaining = report.pages.filter((p) => p.id !== page.id && p.navigation !== false);
   if (remaining.length === 0) {
     io.out(text + "\n");
     return;

package/src/show/render.ts

@@ -1,10 +1,10 @@
 // show 的文本渲染:单 eval 详情、--history 时间轴、三个证据切面(transcript / trace / diff)。
-// 输出形态照 docs-site/zh/how-to/viewing-results.mdx 的示例块;长内容一律截断,
+// 输出形态照 docs-site/zh/tutorials/viewing-results.mdx 的示例块;长内容一律截断,
 // 但截断永远如实标注剩余数量和原始 artifact 路径 —— 输出对上下文窗口友好,事实源留在盘上。
 // 全部纯函数(时间经 now 显式传入),证据数据由调用方 await 好了递进来。
 
 import { join, relative } from "node:path";
-import type { AssertionResult, DiffData, EvalResult, TimingNode, TraceSpan, Verdict } from "../types.ts";
+import type { AssertionResult, DiffData, EvalResult, LocalizedText, TimingNode, TraceSpan, Verdict } from "../types.ts";
 import type { AttemptEvidence, AttemptHandle, Snapshot } from "../results/index.ts";
 import type { AnnotatedSourceLine, SendAnnotation } from "../results/index.ts";
 import { groupIncompatibleVersionSkips } from "../results/index.ts";
@@ -12,11 +12,12 @@ import type { SkippedDir } from "../results/index.ts";
 import type { ExecutionNode } from "../o11y/execution-tree.ts";
 import { foldEvalVerdict } from "../shared/verdict.ts";
 import { summaryText } from "../scoring/display.ts";
-import { attemptCostUSD } from "../report/metrics.ts";
-import { formatDurationMs, formatMetricValue, formatPlainNumber, formatUSD } from "../report/format.ts";
-import { indentBlock, padDisplay, renderAlignedRows, wrapDisplay } from "../report/text/layout.ts";
+import { attemptCostUSD } from "../report/model/metrics.ts";
+import { formatDurationMs, formatMetricValue, formatPlainNumber, formatUSD } from "../report/model/format.ts";
+import { indentBlock, padDisplay, renderAlignedRows, wrapDisplay } from "../report/model/text-layout.ts";
 import type { AttemptHistoryRow } from "./compose.ts";
-import { localizeText, showCommand, type HostCommandContext, type HostReport } from "./report-host.ts";
+import { localizeText, type HostCommandContext } from "../report/runtime/host.ts";
+import { showCommand } from "./command.ts";
 
 const MISSING = "—";
 
@@ -158,74 +159,6 @@ export function assertionLine(a: AssertionResult): string {
   return head;
 }
 
-/** 字段行:首行 `<label>: <值首行>`,值的其余行(如 CommandResult 的 output tail)缩进原样展开。 */
-function fieldLines(label: string, value: string): string[] {
-  const [first, ...rest] = value.split("\n");
-  return [`    ${label}: ${first ?? ""}`, ...rest.map((line) => `      ${line}`)];
-}
-
-/** 每条的通用行组(见 docs/feature/scoring/library/display.md「通用渲染规则」):
- *  首行 `severity · <标题>`(有分组时标题是分组路径,随后 assertion: 行给检查方式),
- *  之后按有则显示的顺序列 expected / received / score / reason / source。 */
-function assertionRecordLines(a: AssertionResult): string[] {
-  const group = groupTitle(a);
-  const optional = a.optional ? "optional · " : "";
-  const title = group ?? a.name;
-  const lines: string[] = [];
-  const scoreSuffix =
-    a.outcome !== "unavailable" && (a.severity === "soft" || a.threshold !== undefined)
-      ? `   ${scoreText(a.score)}${a.threshold !== undefined ? ` / ${scoreText(a.threshold)}` : ""}`
-      : "";
-  lines.push(`  ${a.severity} · ${optional}${title}${scoreSuffix}`);
-  const detailLine = a.detail ?? (group ? a.name : undefined);
-  if (detailLine !== undefined && detailLine !== title) lines.push(`    assertion: ${detailLine}`);
-  if (a.outcome === "unavailable") {
-    lines.push(`    reason: ${a.reason}`);
-  } else {
-    if (a.expected !== undefined) lines.push(...fieldLines("expected", a.expected));
-    if (a.received !== undefined) lines.push(...fieldLines("received", a.received));
-    if (a.evidence !== undefined && a.evidence !== a.received) lines.push(...fieldLines("evidence", a.evidence));
-  }
-  if (a.loc) lines.push(`    source: ${a.loc.file}:${a.loc.line}${a.loc.column ? `:${a.loc.column}` : ""}`);
-  return lines;
-}
-
-/**
- * 默认 Attempt 页的首要诊断:按结果分节,只逐条列出需要看的(见 display.md)——
- * `failures:`(gate 失败,含 --strict 下改判的 soft)、`soft below threshold:`、
- * `scores:`(无阈值 judge 的纯打分)、`unavailable:`(证据评不了的,带 reason)。
- * 全部通过时这些节整体省略(返回 undefined),只留计数行。
- */
-export function failureDiagnostics(assertions: AssertionResult[], width: number, strict?: boolean): string | undefined {
-  const failures = assertions.filter(
-    (a) => a.outcome === "failed" && (a.severity === "gate" || strict),
-  );
-  const softBelow = assertions.filter(
-    (a) => a.outcome === "failed" && a.severity === "soft" && !strict,
-  );
-  const scores = assertions.filter(
-    (a) => a.outcome === "passed" && a.severity === "soft" && a.threshold === undefined && a.score < 1,
-  );
-  const unavailableOnes = assertions.filter((a) => a.outcome === "unavailable");
-  const lines: string[] = [];
-  const section = (title: string, items: AssertionResult[]) => {
-    if (items.length === 0) return;
-    if (lines.length > 0) lines.push("");
-    lines.push(title);
-    for (const a of items) lines.push(...assertionRecordLines(a));
-  };
-  section("failures:", failures);
-  section("soft below threshold:", softBelow);
-  section("scores:", scores);
-  section("unavailable:", unavailableOnes);
-  if (lines.length === 0) return undefined;
-  return lines.flatMap((line) => {
-    if (line === "") return [line];
-    const indent = line.length - line.trimStart().length;
-    return wrapDisplay(line.trimStart(), Math.max(20, width - indent)).map((part) => `${" ".repeat(indent)}${part}`);
-  }).join("\n");
-}
-
 /** 详情块头部:`attempt 3 · compare/codex-gpt-5.4 · failed · 41s · 12.3k tokens · $0.04`。 */
 export function attemptHeader(attempt: AttemptHandle): string {
   const r = attempt.result;
@@ -242,12 +175,13 @@ export function attemptHeader(attempt: AttemptHandle): string {
 }
 
 // ───────────────────────── AttemptEvidence 共用 ─────────────────────────
-// evalSourceText / executionText / attemptOverviewText(--eval / --execution / 默认全景)
-// 与 evalDetailText 的紧凑索引列共用的小件:locator 头、失败原因、
-// 断言计票摘要。三个证据 renderer 与全景面都只消费同一份 AttemptEvidence,不各自读
-// artifact 或重新判定 capability(loadAttemptEvidence 已经算好)。
+// evalSourceText / executionText / timingText / diffText(--source / --execution / --timing / --diff
+// 四个证据切面)与 evalDetailText 的紧凑索引列共用的小件:locator 头、失败原因、断言计票摘要。
+// 四个证据 renderer 都只消费同一份 AttemptEvidence,不各自读 artifact 或重新判定 capability
+// (loadAttemptEvidence 已经算好)。无证据 flag 的默认页改由 niceeval/report 的 attempt-input
+// page 管线渲染(docs/feature/reports/show/attempt.md),不在这里。
 
-/** `--eval` / `--execution` / 全景块的头行:`@<locator> · <evalId> · <experimentId> · <verdict>`。 */
+/** 证据切面的头行:`@<locator> · <evalId> · <experimentId> · <verdict>`。 */
 export function attemptEvidenceHeader(evidence: AttemptEvidence): string {
   return [evidence.locator, evidence.identity.evalId, evidence.identity.experimentId, evidence.result.verdict].join(" · ");
 }
@@ -427,7 +361,7 @@ export function attemptHistoryText(opts: {
  * 拼接这段(单页定义没有「其余页」段);`otherPages` 不含被渲染的那一页。
  */
 export function otherPagesText(opts: {
-  otherPages: { id: string; title: HostReport["pages"][number]["title"] }[];
+  otherPages: { id: string; title: LocalizedText }[];
   command: HostCommandContext;
   locale: string;
 }): string {
@@ -619,7 +553,7 @@ function execBody(
 /**
  * action / subagent 节点渲染成两行(call + result)——节点模型把 call+result 合并成一个
  * ExecutionNode(按 callId),但分两行读更符合「先看调用、再看结果」的阅读顺序,
- * 与 docs-site/zh/how-to/agent-feedback-loop.mdx 的示例一致。
+ * 与 docs-site/zh/tutorials/agent-feedback-loop.mdx 的示例一致。
  */
 function executionNodeLines(node: ExecutionNode, originMs: number, timingAvailable: boolean, width: number): string[] {
   const time = node.kind !== "telemetry" && node.span ? relSeconds(node.span.startMs, originMs) : undefined;
@@ -753,10 +687,6 @@ export function executionText(
   return `${header}\n\n${lines.join("\n")}\n\n${tail.join("\n")}`;
 }
 
-// ───────────────────────── 默认全景:AttemptEvidence 紧凑总览 ─────────────────────────
-
-const MAX_OVERVIEW_DIFF_NAMES = 5;
-
 /** net 效果的单字母标记(A/M/D;none = 动过但净无变化,标 ±)。 */
 function netLetter(net: string): string {
   switch (net) {
@@ -771,15 +701,6 @@ function netLetter(net: string): string {
   }
 }
 
-function overviewDiffLine(diff: DiffData): string {
-  const names = Object.entries(diff.files)
-    .sort(([a], [b]) => a.localeCompare(b))
-    .map(([p, f]) => `${netLetter(f.net)} ${p}`);
-  const shown = names.slice(0, MAX_OVERVIEW_DIFF_NAMES);
-  const more = names.length > shown.length ? ` · +${names.length - shown.length} more` : "";
-  return `changes: ${names.length} ${names.length === 1 ? "file" : "files"} changed by agent · ${shown.join(" · ")}${more}`;
-}
-
 /** 有界行 diff(公共前后缀修剪):对单区域编辑精确,复杂编辑给出上界近似。 */
 function lineDelta(before: string | undefined, after: string | undefined): { adds: number; dels: number } {
   const a = before === undefined ? [] : before.split("\n");
@@ -791,131 +712,6 @@ function lineDelta(before: string | undefined, after: string | undefined): { add
   return { adds: b.length - prefix - suffix, dels: a.length - prefix - suffix };
 }
 
-/**
- * `niceeval show @<locator>` 不带证据 flag 时的默认面:紧凑全景,只给摘要——Eval 断言计票、
- * 执行事件计数、可选 OTel 时间指示、工作区 diff 摘要,不复现 `--eval` 的完整源码、
- * `--execution` 的完整事件流或 `--diff` 的完整文件列表(那些内容各自的证据 flag 才给)。
- */
-/** lifecycle operation → 首页可读标签:点换空格,与 docs/feature/reports/show.md 的
- *  `phase: sandbox provision` 字面一致(不引入第二套本地化标签,show 首页用英文原样呈现)。 */
-function operationWords(operation: string): string {
-  return operation.replace(/\./g, " ");
-}
-
-/**
- * errored attempt 的结构化 `error:` 块(见 docs/feature/reports/show.md「errored attempt 的首页」)。
- * 先展开 phase(= operation 的可读形态)/ code / message / cause,stack 放在块后、保持原始换行。
- * 字段来自结构化 `AttemptError`;非 errored(`r.error === undefined`)返回 undefined。
- */
-function renderErrorBlock(r: EvalResult): string | undefined {
-  const err = r.error;
-  if (err === undefined) return undefined;
-  const lines = ["error:", `  phase: ${err.phase}`, `  code: ${err.code}`, `  message: ${err.message}`];
-  if (err.cause) {
-    const c = err.cause;
-    const causeText = c.name ? `${c.name} · ${c.message}` : c.message;
-    lines.push(`  cause: ${causeText}`);
-  }
-  // stack 放在 error 块之后、保持原始换行(见 docs);没有 stack(如 timeout / turn-failed)就不加空块。
-  if (err.stack && err.stack.trim() !== "") return `${lines.join("\n")}\n\n${err.stack.replace(/\n+$/, "")}`;
-  return lines.join("\n");
-}
-
-/**
- * attempt 级诊断块(见 docs/feature/reports/show.md「diagnostics」)。每条一行标头
- * `<level> · <operation> · <code>`,message 缩进在下一行;`count > 1` 时补 `(N occurrences)`。
- * 诊断的 level 与 verdict 无关 —— passed / failed / errored 都可能带一条 cleanup / teardown
- * warning。没有诊断返回 undefined。
- */
-function renderAttemptDiagnostics(r: EvalResult): string | undefined {
-  if (!r.diagnostics || r.diagnostics.length === 0) return undefined;
-  const lines = ["diagnostics:"];
-  for (const d of r.diagnostics) {
-    lines.push(`  ${d.level} · ${d.phase} · ${d.code}`);
-    const occ = d.count && d.count > 1 ? ` (${d.count} occurrences)` : "";
-    lines.push(`    ${d.message}${occ}`);
-  }
-  return lines.join("\n");
-}
-
-export function attemptOverviewText(
-  evidence: AttemptEvidence,
-  opts: { header: string; artifactPath?: string; width: number },
-): string {
-  const { header, artifactPath } = opts;
-  const r = evidence.result;
-
-  const metaParts = [`snapshot ${evidence.identity.snapshotStartedAt}`, `attempt ${evidence.identity.attempt + 1}`, formatDurationMs(r.durationMs)];
-  if (r.usage) metaParts.push(`${formatMetricValue(r.usage.inputTokens + r.usage.outputTokens)} tokens`);
-  const cost = attemptCostUSD(r);
-  if (cost !== null) metaParts.push(formatUSD(cost));
-
-  const blocks: string[] = [[header, metaParts.join(" · ")].join("\n")];
-
-  // errored attempt 的首页不靠 trace 就要能解释基础设施错误(见 docs/feature/reports/show.md
-  // 「errored attempt 的首页」):先展开结构化 error(phase/operation/code/message/cause + stack),
-  // 断言块对没有断言的 errored attempt 省略(它在评分之前就挂了)。
-  const errorBlock = renderErrorBlock(r);
-  if (errorBlock) blocks.push(errorBlock);
-
-  // errored 且没有任何断言时不打印空的 "assertions: 0 passed" —— 主因已经在 error 块里说清楚了。
-  if (r.assertions.length > 0 || r.error === undefined) {
-    blocks.push(
-      [
-        assertionSummaryLine(r.assertions),
-        evidence.evalSource
-          ? `eval source: ${evidence.evalSource.sourcePath} · sha256:${evidence.evalSource.sourceSha256.slice(0, 8)}…`
-          : "eval source: unavailable (not captured for this attempt)",
-      ].join("\n"),
-    );
-  }
-
-  const diagnostics = failureDiagnostics(r.assertions, opts.width);
-  if (diagnostics) blocks.push(diagnostics);
-
-  // attempt 级诊断(teardown/cleanup 等,与 verdict 独立;passed/failed/errored 都可能有)。
-  const attemptDiag = renderAttemptDiagnostics(r);
-  if (attemptDiag) blocks.push(attemptDiag);
-
-  if (evidence.execution) {
-    const nodes = evidence.execution.nodes.filter((node) => node.kind !== "telemetry");
-    const skillLoads = nodes.filter((n) => n.kind === "skill.loaded").length;
-    const toolCalls = nodes.filter((n) => n.kind === "action").length;
-    const aiMessages = nodes.filter((n) => n.kind === "message" && n.role === "assistant").length;
-    const counted = (count: number, singular: string, plural = `${singular}s`) =>
-      `${count} ${count === 1 ? singular : plural}`;
-    const execLines = [
-      `execution: ${counted(nodes.length, "event")} · ${counted(skillLoads, "skill load")} · ${counted(toolCalls, "tool call")} · ${counted(aiMessages, "AI message")}`,
-    ];
-    const timingLine = overviewTimingLine(r);
-    if (timingLine) execLines.push(timingLine);
-    blocks.push(execLines.join("\n"));
-  } else {
-    blocks.push("execution: unavailable (no events recorded for this attempt)");
-  }
-
-  if (evidence.diff && evidence.capabilities.diff) {
-    blocks.push(overviewDiffLine(evidence.diff));
-  } else if (evidence.diff) {
-    blocks.push("changes: diff unavailable · this attempt did not produce workspace file changes");
-  } else {
-    blocks.push("changes: diff unavailable · no workspace diff was recorded for this attempt");
-  }
-
-  const tail: string[] = [];
-  if (artifactPath) tail.push(`artifacts: ${artifactPath}/`);
-  const available = [
-    evidence.capabilities.source ? `niceeval show ${evidence.locator} --source` : undefined,
-    evidence.capabilities.execution ? `niceeval show ${evidence.locator} --execution` : undefined,
-    evidence.capabilities.timing ? `niceeval show ${evidence.locator} --timing` : undefined,
-    evidence.capabilities.diff ? `niceeval show ${evidence.locator} --diff` : undefined,
-  ].filter((command): command is string => command !== undefined);
-  if (available.length > 0) tail.push(`available:\n${available.map((command) => `  ${command}`).join("\n")}`);
-  if (tail.length > 0) blocks.push(tail.join("\n"));
-
-  return blocks.join("\n\n");
-}
-
 // ───────────────────────── 证据切面:diff ─────────────────────────
 
 const MAX_DIFF_LINES = 200;
@@ -1013,26 +809,6 @@ function windowHunk(c: { status: string; before?: string; after?: string }): str
 
 const CLOSING_PHASE_NAMES = new Set(["eval.teardown", "agent.teardown", "sandbox.teardown", "sandbox.suspend", "sandbox.stop"]);
 
-// 首页只保留能回答「大头在哪」的阶段。分类账锚点与 telemetry bookkeeping 在很短时
-// 留给 --timing 完整树；一旦自身变成慢点或失败，仍必须抬到首页，不能被静默藏掉。
-const OVERVIEW_DETAIL_PHASE_NAMES = new Set(["workspace.baseline", "telemetry.configure", "telemetry.collect"]);
-const OVERVIEW_DETAIL_THRESHOLD_MS = 1_000;
-
-/** phases 主链摘要行(首页 `timing:`):各主链阶段耗时点隔,收尾段合计以 `teardown +N` 单列。 */
-export function overviewTimingLine(r: EvalResult): string | undefined {
-  if (!r.phases || r.phases.length === 0) return undefined;
-  const main = r.phases.filter(
-    (p) =>
-      !CLOSING_PHASE_NAMES.has(p.name) &&
-      (p.failed === true || !OVERVIEW_DETAIL_PHASE_NAMES.has(p.name) || p.durationMs >= OVERVIEW_DETAIL_THRESHOLD_MS),
-  );
-  const closing = r.phases.filter((p) => CLOSING_PHASE_NAMES.has(p.name));
-  const parts = main.map((p) => `${p.name} ${formatDurationMs(p.durationMs)}${p.failed ? " ✗" : ""}`);
-  const closingMs = closing.reduce((sum, p) => sum + p.durationMs, 0);
-  if (closingMs > 0) parts.push(`teardown +${formatDurationMs(closingMs)}`);
-  return `timing: ${parts.join(" · ")}`;
-}
-
 const TIMING_DETAIL_NODE_BUDGET = 80;
 
 interface DiagnosticTimingNode {