niceeval 0.9.0 → 0.10.0

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 = "—";
 
@@ -113,7 +114,7 @@ export function attemptsOfEval(snapshots: Snapshot[], evalId: string): AttemptHa
 /**
  * 详情块 / eval-id 前缀证据切面默认挑最新一次失败的 attempt;没有失败挑最新一次。
  * 精确选某一次 attempt 走 `@<locator>`(`resolveLocator`),不再有数字 `--attempt`——
- * --experiment 已在 Selection 合成时收窄。
+ * --exp 已在 Selection 合成时收窄。
  */
 export function pickDetailAttempt(attempts: AttemptHandle[]): AttemptHandle | undefined {
   if (attempts.length === 0) return undefined;
@@ -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 {

package/src/show/show.test.ts

@@ -1,5 +1,5 @@
 // cases: docs/engineering/unit-tests/reports/cases.md
-// niceeval show 终端宿主的测试(行为规范:docs-site/zh/how-to/viewing-results.mdx;
+// niceeval show 终端宿主的测试(行为规范:docs-site/zh/tutorials/viewing-results.mdx;
 // 组合语义:docs/feature/reports/architecture.md「Selection 是计算入口」)。覆盖:
 // - 榜单合成口径:每 experiment × eval 取最新判定,局部重跑从更早快照补齐,头部标注合成自几个快照;
 // - 前缀过滤收窄 Selection,覆盖警告分母 = 已知并集 ∩ 范围;
@@ -22,7 +22,7 @@ import { RESULTS_FORMAT, RESULTS_SCHEMA_VERSION, type EvalResult, type TimingNod
 import { selectCurrentResults } from "../results/select.ts";
 import { attemptHistory } from "./compose.ts";
 import { runShow, type ShowFlags } from "./index.ts";
-import { stringWidth } from "../report/text/layout.ts";
+import { stringWidth } from "../report/model/text-layout.ts";
 
 // ───────────────────────── fixture 工具 ─────────────────────────
 
@@ -152,12 +152,12 @@ describe("默认报告:跨快照合成的现刻水位(ExperimentComparison text
     const root = await seedComposedRoot();
     const { out, code } = await show(root, []);
     expect(code).toBe(0);
-    expect(out).toContain("No data to plot Cost × End-to-end pass rate");
+    expect(out).toContain("No data to plot Cost × Pass rate");
     expect(out).toContain("1 point missing data");
     expect(out).not.toContain("needed to compare");
     expect(out).not.toContain("COMPARISON");
     expect(out).toMatch(/\bbub\s+default\s+bub\s+1s\s+50%/);
-    expect(out).toMatch(/1 passed[\s\S]*?\/ 1\s+failed/);
+    expect(out).toMatch(/1 passed[\s\S]*?· 1\s+failed/);
     expect(out).toMatch(/✗ failed\s+fixtures\/button[\s\S]*└─ @1[0-9a-z]{7}[\s\S]*fileChanged/);
     expect(out).toMatch(/✓ passed\s+weather\/brooklyn[\s\S]*└─ @1[0-9a-z]{7}/);
     expect(out).not.toMatch(/\[[EXD⏱,]+\]/);
@@ -220,8 +220,8 @@ describe("默认报告:跨快照合成的现刻水位(ExperimentComparison text
     try {
       const { out, code } = await show(root, []);
       expect(code).toBe(0);
-      expect(out).toContain("成本 × 端到端成功率 没有可绘制的数据");
-      expect(out).toContain("1 通过 / 1 失败");
+      expect(out).toContain("成本 × 通过率 没有可绘制的数据");
+      expect(out).toContain("1 通过 · 1 失败");
       expect(out).toMatch(/\bbub\s+默认\s+bub\s+1s\s+50%/);
       // 页首 Hero 与尾部页索引同样跟随 locale(内置文案与页名)。
       expect(out.split("\n")[0]).toBe("Eval 运行结果");
@@ -259,7 +259,10 @@ describe("默认报告:跨快照合成的现刻水位(ExperimentComparison text
     expect(code).toBe(0);
     // 成本轴(better: lower)反向,「更好」恒指向右上;两轴都声明 better → 提示在场
     expect(out).toContain("better → upper right");
-    expect(out).toContain("A compare/a   B compare/b");
+    // 无 line 声明:按 agent 归类、不连线;图例一行一个 series(显示键字典序),标记按图例顺序分配
+    expect(out).toContain("grouped by agent");
+    expect(out).toContain("claude  A compare/b");
+    expect(out).toContain("codex   B compare/a");
     expect(out).toMatch(/\bb\s+large\s+claude\s+1s\s+100%/);
     expect(out).toMatch(/\ba\s+mini\s+codex\s+1s\s+50%/);
     expect(out).toMatch(/✓ passed\s+q1[\s\S]*└─ @1[0-9a-z]{7}/);
@@ -480,12 +483,23 @@ describe("--report 装载", () => {
         "    locator: ctx.scope.snapshots[0].evals[0].attempts[0].locator,",
         "  },",
         "});",
+        "const AttemptPage = () => null;",
+        "AttemptPage[FACES] = {",
+        "  resolve: (props, ctx) => ({ locator: ctx.page.locator, verdict: ctx.page.evidence.result.verdict }),",
+        "  web: () => null,",
+        "  text: (props) => `CUSTOM ATTEMPT PAGE · ${props.locator} · ${props.verdict}`,",
+        "};",
         "const definition = {",
         '  kind: "report",',
         "  links: [],",
         "  scripts: [],",
         "  styles: [],",
-        '  pages: [{ id: "report", title: "Report", content: { type: Wrapper, props: {} } }],',
+        "  pages: [",
+        '    { id: "report", title: "Report", input: "scope", navigation: true, content: { type: Wrapper, props: {} } },',
+        // 报告需要声明一张 attempt-input page,attemptCommand 生成器才存在(architecture.md
+        // 「Attempt 详情是一张参数化 page」);navigation:false 不进「其余页」索引,单页断言不受影响。
+        '    { id: "attempt", title: "Attempt", input: "attempt", navigation: false, content: { type: AttemptPage, props: {} } },',
+        "  ],",
         "};",
         "Object.defineProperty(definition, DEFINITION, { value: true });",
         "export default definition;",
@@ -592,7 +606,7 @@ describe("--report 装载", () => {
     expect(out).toContain("CUSTOM weather/brooklyn ·");
   });
 
-  it("--experiment 让 Selection 只留该实验", async () => {
+  it("--exp 让 Selection 只留该实验", async () => {
     const root = await seedComposedRoot();
     await writeSnapshot(
       root,
@@ -660,6 +674,75 @@ describe("--report 装载", () => {
     expect(builtin.err).toContain('page "typo" not found in the built-in report');
     expect(builtin.err).toContain("Available pages: report, attempts, traces");
   });
+
+  /** 只有一张 scope-input page,没有声明 attempt-input page。 */
+  async function writeReportFileNoAttemptPage(dir: string): Promise<string> {
+    const path = join(dir, "no-attempt-page.mjs");
+    await writeFile(
+      path,
+      [
+        'const FACES = Symbol.for("niceeval.report.faces");',
+        'const DEFINITION = Symbol.for("niceeval.report.definition");',
+        "const Overview = () => null;",
+        "Overview[FACES] = { web: () => null, text: () => \"OVERVIEW\" };",
+        "const definition = {",
+        '  kind: "report",',
+        "  links: [],",
+        "  scripts: [],",
+        "  styles: [],",
+        '  pages: [{ id: "report", title: "Report", input: "scope", navigation: true, content: { type: Overview, props: {} } }],',
+        "};",
+        "Object.defineProperty(definition, DEFINITION, { value: true });",
+        "export default definition;",
+        "",
+      ].join("\n"),
+      "utf-8",
+    );
+    return path;
+  }
+
+  it("show @<locator> --report <file>:选中该定义自己声明的 attempt-input page,不是内建固定首页(docs/engineering/unit-tests/reports/cases.md 第 202 行)", async () => {
+    const root = await seedComposedRoot();
+    const report = await writeReportFile(root);
+    const results = await openResults(root);
+    const evals = results.experiments[0]!.snapshots.flatMap((s) => s.evals);
+    const locator = evals.find((e) => e.id === "weather/brooklyn")!.attempts[0]!.locator!;
+
+    const { out, code } = await show(root, [locator], { report });
+    expect(code).toBe(0);
+    expect(out).toBe(`CUSTOM ATTEMPT PAGE · ${locator} · passed\n`);
+    // 不是内建 standard 的固定首页文案(AttemptSummary/AttemptAssessment 等)
+    expect(out).not.toContain("niceeval show");
+  });
+
+  it("自定义报告没有 attempt-input page 时,裸 show @<locator> --report <file> 报完整用户反馈,指引三种解决路径,不回退到内建详情(第 204 行)", async () => {
+    const root = await seedComposedRoot();
+    const report = await writeReportFileNoAttemptPage(root);
+    const results = await openResults(root);
+    const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
+
+    const { err, code } = await show(root, [locator], { report });
+    expect(code).toBe(1);
+    expect(err).toContain(report);
+    expect(err).toContain("has no attempt-input page");
+    expect(err).toContain("extends: standard");
+    expect(err).toContain("standardAttemptPage");
+    expect(err).toContain('input: "attempt"');
+    // 不静默回退渲染内建 standard 的详情页(那会让用户以为自定义报告本来就有这页)
+    expect(err).not.toContain("Eval Results");
+  });
+
+  it("--source/--execution/--diff 直接读 AttemptEvidence,不经 page content:没有 attempt page 的自定义报告下仍正常工作(第 203 行)", async () => {
+    const root = await seedComposedRoot();
+    const report = await writeReportFileNoAttemptPage(root);
+    const results = await openResults(root);
+    const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
+
+    const { out, code } = await show(root, [locator], { report, diff: true });
+    expect(code).toBe(0);
+    expect(out).toContain(locator);
+    expect(out).not.toContain("has no attempt-input page");
+  });
 });
 
 // ───────────────────────── 证据切面:--execution ─────────────────────────
@@ -812,7 +895,7 @@ describe("show @<locator>", () => {
     const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
     const { out } = await show(root, [locator]);
     expect(out).toContain("gate · Issue 15193");
-    expect(out).toContain("assertion: equals(4)");
+    expect(out).toContain("equals(4)");
     expect(out).toContain("expected: 4");
     expect(out).toContain("received: 1");
     expect(out).toContain("source: evals/manager.eval.ts:40:11");
@@ -829,10 +912,13 @@ describe("show @<locator>", () => {
 
     const { out, code } = await show(root, [locator]);
     expect(code).toBe(0);
-    expect(out).toContain(`${locator} · weather/brooklyn · compare/bub · passed`);
-    expect(out).toContain("attempt 1 · ");
-    expect(out).toContain("execution: unavailable (no events recorded for this attempt)");
-    expect(out).toContain("changes: diff unavailable");
+    // 首行不做整段精确匹配:startedAt 按本机时区格式化,时间部分跨时区不稳定,只断言与时区无关的片段。
+    expect(out).toContain(`${locator} · weather/brooklyn · compare/bub · attempt 1 · ✓ passed`);
+    expect(out).toContain("· 1.0s");
+    // 没有 events/diff 的组件零输出(不是"execution: unavailable"这类占位文案)——
+    // 与「Attempt 详情组件族:非空/空证据矩阵」的单元测试同一条契约,这里是集成层确认。
+    expect(out).not.toContain("execution:");
+    expect(out).not.toContain("changes:");
     expect(out).not.toContain("evidence:");
     expect(out).not.toContain("available:");
   });
@@ -857,22 +943,22 @@ describe("show @<locator>", () => {
 
     const { out, code } = await show(root, [locator]);
     expect(code).toBe(0);
-    expect(out).toContain(`${locator} · agent-029 · compare/claude-e2b · errored`);
-    // 结构化 error 块:phase 直接展示闭集点分名,code/message/cause 各一行(见 docs/feature/reports/show.md)
-    expect(out).toContain("error:");
+    // 同上:跳过 startedAt 的时区敏感部分,只断言与时区无关的片段。
+    expect(out).toContain(`${locator} · agent-029 · compare/claude-e2b · attempt 1 · ! errored`);
+    expect(out).toContain("· 1.0s");
+    // 结构化 error 块:code 折进 error: 首行,phase/message/cause 各一行(AttemptError)
+    expect(out).toContain("error: sandbox-rate-limit");
     expect(out).toContain("phase: sandbox.create");
-    expect(out).toContain("code: sandbox-rate-limit");
     expect(out).toContain("message: E2B sandbox allocation failed after 5 attempts");
     expect(out).toContain("cause: RateLimitError · too many concurrent sandboxes");
-    // attempt 级 diagnostics 块(与 verdict 独立)
-    expect(out).toContain("diagnostics:");
-    expect(out).toContain("warning · sandbox.teardown · teardown-failed");
-    expect(out).toContain("container stop timed out");
-    // errored 且无断言:不打印空的 assertions 汇总行
+    // attempt 级 diagnostics 按 lifecycle phase 分组(AttemptDiagnostics),标题即 phase 名
+    expect(out).toContain("sandbox.teardown:");
+    expect(out).toContain("warning · teardown-failed — container stop timed out");
+    // errored 且无断言:AttemptAssertions 零输出,不打印空的 assertions 汇总行
     expect(out).not.toContain("assertions:");
   });
 
-  it("available 只逐行列出实际存在的证据命令,不打印能力字母或合并式 next", async () => {
+  it("证据下钻命令散落在各自区块行尾(带 locator),不存在的能力不生成假命令,也没有合并式 available/next 尾块", async () => {
     const root = await makeRoot();
     const dir = await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
       res("weather/brooklyn", "failed", { hasEvents: true }),
@@ -889,9 +975,14 @@ describe("show @<locator>", () => {
 
     const { out, code } = await show(root, [locator]);
     expect(code).toBe(0);
-    expect(out).toContain(`available:\n  niceeval show ${locator} --source\n  niceeval show ${locator} --execution`);
+    // AttemptSource / AttemptConversation 各自行尾带命令(不是集中列在一个 available: 块里)
+    expect(out).toContain(`niceeval show ${locator} --source`);
+    expect(out).toContain(`niceeval show ${locator} --execution`);
+    // 没有 diff/timing 能力(fixture 没写 diff.json,也没有 phases)时不生成假命令
     expect(out).not.toContain(`niceeval show ${locator} --diff`);
+    expect(out).not.toContain(`niceeval show ${locator} --timing`);
     expect(out).not.toContain("evidence:");
+    expect(out).not.toContain("available:");
     expect(out).not.toContain("next:");
   });
 
@@ -1054,7 +1145,7 @@ describe("show @<locator>", () => {
     expect(out).not.toContain("git show ×");
   });
 
-  it("attempt 首页 timing 只列大头,但保留变慢的 telemetry 阶段", async () => {
+  it("attempt 首页 timing 逐阶段一行,收尾段单独归拢(AttemptTimeline 不做单行宽度收窄,--timing 才是完整分解)", async () => {
     const root = await makeRoot();
     const dir = await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
       res("weather/brooklyn", "passed", {
@@ -1080,15 +1171,17 @@ describe("show @<locator>", () => {
     const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
 
     const { out } = await show(root, [locator]);
-    const timingLine = out.split("\n").find((line) => line.startsWith("timing:"));
-    expect(out).toContain("1 AI message");
-    expect(timingLine).toContain("sandbox.queue 0ms");
-    expect(timingLine).toContain("sandbox.create 3.0s");
-    expect(timingLine).toContain("eval.run 45.0s");
-    expect(timingLine).toContain("telemetry.collect 35.3s");
-    expect(timingLine).toContain("teardown +8.5s");
-    expect(timingLine).not.toContain("workspace.baseline");
-    expect(timingLine).not.toContain("telemetry.configure");
+    // 每个主链阶段各占一行(不再是旧格式的单行拼接,那条规则本身是为单行可读服务的,
+    // 多行块没有这个约束了),收尾段(sandbox.stop)单独归拢在 teardown: 之下。
+    expect(out).toContain("timing: 1m 24s");
+    expect(out).toContain("sandbox.queue 0ms");
+    expect(out).toContain("sandbox.create 3.0s");
+    expect(out).toContain("workspace.baseline 310ms");
+    expect(out).toContain("telemetry.configure 358ms");
+    expect(out).toContain("eval.run 45.0s");
+    expect(out).toContain("telemetry.collect 35.3s");
+    expect(out).toContain("teardown:");
+    expect(out).toContain("sandbox.stop 8.5s");
   });
 
   it("语法不对的 locator 报「not a valid attempt locator」,退出码 1,不崩", async () => {

package/src/view/app/App.test.tsx

@@ -22,11 +22,11 @@ const reportPages = {
 };
 
 function dataWithShell(report: ViewData["report"]): ViewData {
-  return { composedRuns: 1, snapshots: [], ...(report !== undefined ? { report } : {}) };
+  return { composedRuns: 1, ...(report !== undefined ? { report } : {}) };
 }
 
-describe("外壳:宿主无品牌位与 hero,导航只有报告页", () => {
-  it("宿主导航壳 DOM 无品牌节点、无 hero 区:品牌与 hero 是页内组件,不归宿主", () => {
+describe("外壳:宿主恒有 NiceEval 品牌位、无 hero,导航只有报告页", () => {
+  it("宿主页头有恒定品牌字标、无 hero 区:hero 是页内组件不归宿主,品牌归宿主", () => {
     const html = renderToStaticMarkup(
       <App
         data={dataWithShell({
@@ -38,11 +38,12 @@ describe("外壳:宿主无品牌位与 hero,导航只有报告页", () => {
         reportPages={reportPages}
       />,
     );
-    // 宿主 DOM 无品牌节点:没有字标、没有 Powered by、没有任何官网品牌链接。
-    expect(html).not.toContain('class="brand"');
-    expect(html).not.toContain("NiceEval");
+    // 宿主页头恒有 NiceEval 品牌字标,外链官网;报告定义不能覆盖或移除。
+    expect(html).toContain('class="brand"');
+    expect(html).toContain("NiceEval");
+    expect(html).toContain("niceeval.com");
+    // 品牌位是页头字标,不是 hero 的 Powered by 行——后者是页内组件,不落宿主壳。
     expect(html).not.toContain("Powered by");
-    expect(html).not.toContain("niceeval.com");
     // 宿主无 hero 区:标题只落浏览器 <title>(useEffect,静态渲染不执行),不落任何宿主节点。
     expect(html).not.toContain('class="hero"');
     expect(html).not.toContain("<h1");