niceeval 0.5.4 → 0.6.0

package/src/show/show.test.ts

@@ -1,26 +1,27 @@
 // niceeval show 终端宿主的测试(行为规范:docs-site/zh/guides/viewing-results.mdx;
-// 组合语义:docs/reports.md「宿主输入的组合语义」)。覆盖:
-// - 榜单合成口径:每 experiment × eval 取最新判决,局部重跑从更早 run 补齐,头部标注合成自几个 run;
-// - 前缀过滤收窄选集,覆盖警告分母 = 已知并集 ∩ 范围;
+// 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;
 // - --history 时间轴只列真实执行,resume 携带的复印件不占行;
-// - --report 装载(合法 / 非法默认导出 / 文件缺失)、位置前缀收窄注入选集、attemptCommand 下钻;
+// - --report 装载(合法 / 非法默认导出 / 文件缺失)、位置前缀收窄注入 Selection、attemptCommand 下钻;
 // - 互斥:--history 与 --report;
-// - 单 eval 详情与 --transcript 证据切面的输出形态。
+// - 单 eval 详情、`@<locator>` 精确定位与 --eval / --execution / --diff 证据切面的输出形态。
+//
+// fixture 直接写新布局(<expDir>/<snapDir>/snapshot.json + <evalId>/a<n>/result.json),
+// 依据是 docs/feature/results/architecture.md 的稳定磁盘契约,不经 writer 运行时 API(避免与并行重写的
+// niceeval/results 写入面签名耦合)。
 
-import { afterEach, describe, expect, it } from "vitest";
+import { afterAll, afterEach, beforeAll, describe, expect, it } from "vitest";
 import { mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
-import {
-  RESULTS_FORMAT,
-  RESULTS_SCHEMA_VERSION,
-  createRunWriter,
-  openResults,
-  type EvalResult,
-  type RunSummary,
-} from "../results/index.ts";
-import { composeShowSelection, evalHistory } from "./compose.ts";
+import { createHash } from "node:crypto";
+import { openResults } from "../results/index.ts";
+import { RESULTS_FORMAT, RESULTS_SCHEMA_VERSION, type EvalResult, type Verdict } from "../types.ts";
+import { selectCurrentResults } from "../results/select.ts";
+import { evalHistory } from "./compose.ts";
 import { runShow, type ShowFlags } from "./index.ts";
+import { stringWidth } from "../report/text/layout.ts";
 
 // ───────────────────────── fixture 工具 ─────────────────────────
 
@@ -34,40 +35,72 @@ afterEach(async () => {
   await Promise.all(roots.splice(0).map((r) => rm(r, { recursive: true, force: true })));
 });
 
-function res(over: Partial<EvalResult> & Pick<EvalResult, "id">): EvalResult {
-  return {
-    agent: "bub",
-    outcome: "passed",
-    attempt: 0,
-    durationMs: 1000,
-    assertions: [],
-    ...over,
-  };
+// show 的报告 chrome 跟随 CLI 界面语言(detectLocale);本文件的断言按英文写,
+// 固定 en 让用例不随宿主机 LANG 漂移;zh-CN 传递单独一条用例覆盖。
+let langBackup: string | undefined;
+beforeAll(() => {
+  langBackup = process.env.NICEEVAL_LANG;
+  process.env.NICEEVAL_LANG = "en";
+});
+afterAll(() => {
+  if (langBackup === undefined) delete process.env.NICEEVAL_LANG;
+  else process.env.NICEEVAL_LANG = langBackup;
+});
+
+/** 一条 attempt 的最小 fixture;字段照 docs/feature/results/architecture.md 的 AttemptRecord。 */
+type AttemptFixture = Pick<EvalResult, "id" | "verdict"> &
+  Partial<
+    Pick<
+      EvalResult,
+      "attempt" | "durationMs" | "assertions" | "estimatedCostUSD" | "usage" | "error" | "startedAt" | "artifactBase" | "hasEvents" | "hasTrace"
+    >
+  >;
+
+function res(id: string, verdict: Verdict, extra: Partial<AttemptFixture> = {}): AttemptFixture {
+  return { id, verdict, attempt: 0, durationMs: 1000, assertions: [], ...extra };
+}
+
+/** 实验目录名的清洗:与 docs/feature/results/architecture.md 一致(/ 与非 [\w.@-] 换成 _)。 */
+function cleanDirName(id: string): string {
+  return id.replace(/[^\w.@-]/g, "_");
+}
+
+interface SnapshotOpts {
+  experimentId: string;
+  agent?: string;
+  model?: string;
+  startedAt: string;
+  completedAt?: string;
+  knownEvalIds?: string[];
 }
 
-function summaryOf(results: EvalResult[], over: Partial<RunSummary> = {}): RunSummary {
-  const count = (o: EvalResult["outcome"]) => results.filter((r) => r.outcome === o).length;
-  return {
+/** 写一份新布局快照:snapshot.json + 各 attempt 的 result.json。返回快照目录绝对路径。 */
+async function writeSnapshot(
+  root: string,
+  snapDirName: string,
+  opts: SnapshotOpts,
+  results: AttemptFixture[],
+): Promise<string> {
+  const dir = join(root, cleanDirName(opts.experimentId), snapDirName);
+  await mkdir(dir, { recursive: true });
+  const meta = {
     format: RESULTS_FORMAT,
     schemaVersion: RESULTS_SCHEMA_VERSION,
     producer: { name: "niceeval", version: "0.4.6" },
-    agent: results[0]?.agent ?? "bub",
-    startedAt: "2026-07-08T10:00:00.000Z",
-    completedAt: "2026-07-08T10:10:00.000Z",
-    passed: count("passed"),
-    failed: count("failed"),
-    skipped: count("skipped"),
-    errored: count("errored"),
-    durationMs: 60_000,
-    results,
-    ...over,
+    experimentId: opts.experimentId,
+    agent: opts.agent ?? "bub",
+    ...(opts.model !== undefined ? { model: opts.model } : {}),
+    startedAt: opts.startedAt,
+    completedAt: opts.completedAt ?? opts.startedAt,
+    ...(opts.knownEvalIds ? { knownEvalIds: opts.knownEvalIds } : {}),
   };
-}
-
-async function writeRun(root: string, dirName: string, summary: RunSummary): Promise<void> {
-  const dir = join(root, dirName);
-  await mkdir(dir, { recursive: true });
-  await writeFile(join(dir, "summary.json"), JSON.stringify(summary, null, 2), "utf-8");
+  await writeFile(join(dir, "snapshot.json"), JSON.stringify(meta, null, 2), "utf-8");
+  for (const r of results) {
+    const attemptDir = join(dir, r.id, `a${r.attempt ?? 0}`);
+    await mkdir(attemptDir, { recursive: true });
+    await writeFile(join(attemptDir, "result.json"), JSON.stringify(r, null, 2), "utf-8");
+  }
+  return dir;
 }
 
 interface Captured {
@@ -76,89 +109,114 @@ interface Captured {
   code: number;
 }
 
-async function show(root: string, patterns: string[], flags: ShowFlags = {}): Promise<Captured> {
+async function show(root: string, patterns: string[], flags: ShowFlags = {}, width = 100): Promise<Captured> {
   let out = "";
   let err = "";
   const code = await runShow(root, patterns, { run: root, ...flags }, {
     out: (s) => (out += s),
     err: (s) => (err += s),
-    width: 100,
+    width,
     now: Date.parse("2026-07-09T10:01:00.000Z"),
   });
   return { out, err, code };
 }
 
-/** 两个 run:老 run 全量(a ✓ b ✓),新 run 只重跑 b(✗)—— 榜单该跨 run 合成。 */
+/** 两个快照:老快照全量(a ✓ b ✓),新快照只重跑 b(✗)—— 榜单该跨快照合成。 */
 async function seedComposedRoot(): Promise<string> {
   const root = await makeRoot();
-  await writeRun(
-    root,
-    "2026-07-08T10-00-00-000Z",
-    summaryOf(
-      [
-        res({ id: "weather/brooklyn", experimentId: "compare/bub", startedAt: "2026-07-08T10:00:01.000Z" }),
-        res({ id: "fixtures/button", experimentId: "compare/bub", startedAt: "2026-07-08T10:00:02.000Z" }),
-      ],
-      { startedAt: "2026-07-08T10:00:00.000Z" },
-    ),
-  );
-  await writeRun(
-    root,
-    "2026-07-09T10-00-00-000Z",
-    summaryOf(
-      [
-        res({
-          id: "fixtures/button",
-          experimentId: "compare/bub",
-          outcome: "failed",
-          startedAt: "2026-07-09T10:00:01.000Z",
-          assertions: [
-            {
-              name: 'fileChanged("src/components/Button.tsx")',
-              severity: "gate",
-              score: 0,
-              passed: false,
-              detail: "file was not modified",
-            },
-          ],
-        }),
+  await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+    res("weather/brooklyn", "passed"),
+    res("fixtures/button", "passed"),
+  ]);
+  await writeSnapshot(root, "2026-07-09T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-09T10:00:00.000Z" }, [
+    res("fixtures/button", "failed", {
+      assertions: [
+        {
+          name: 'fileChanged("src/components/Button.tsx")',
+          severity: "gate",
+          score: 0,
+          passed: false,
+          detail: "file was not modified",
+        },
       ],
-      { startedAt: "2026-07-09T10:00:00.000Z" },
-    ),
-  );
+    }),
+  ]);
   return root;
 }
 
 // ───────────────────────── 榜单合成口径 ─────────────────────────
 
-describe("榜单:跨 run 合成的现刻水位", () => {
-  it("局部重跑不撕榜单:另一题从更早 run 补齐,头部如实标注合成自 2 个 run", async () => {
+describe("榜单:跨快照合成的现刻水位(show 专用 attempt 表)", () => {
+  it("局部重跑不撕榜单:另一题从更早快照补齐,单实验不出现比较空态", async () => {
     const root = await seedComposedRoot();
     const { out, code } = await show(root, []);
     expect(code).toBe(0);
-    expect(out).toContain("Current verdicts · 1 experiment · composed from 2 runs");
-    expect(out).toContain("latest 2026-07-09T10-00-00-000Z");
-    // eval 级折叠计票:2 题里 1 题通过;两题都在(没有 partial-coverage 撕榜)
-    expect(out).toContain("1/2");
-    expect(out).toContain("50%");
-    expect(out).not.toContain("verdicts cover");
-    // Failing 清单:新判决的 fixtures/button,带失败断言与下钻命令
-    expect(out).toContain("Failing:");
-    expect(out).toContain("✗ fixtures/button");
-    expect(out).toContain('gate fileChanged("src/components/Button.tsx")');
-    expect(out).toContain("→ niceeval show fixtures/button");
-    expect(out).not.toContain("✗ weather/brooklyn");
-  });
-
-  it("合成选集:每 experiment × eval 取最新判决;compose 不产生残缺警告", async () => {
+    expect(out).not.toContain("No data to plot");
+    expect(out).not.toContain("At least 2 experiments");
+    expect(out).not.toContain("COMPARISON");
+    expect(out).toContain("EXPERIMENT  compare/bub · bub");
+    expect(out).toContain("SUMMARY     1 passed · 1 failed · 0 errored · 0 skipped · 2 attempts");
+    expect(out).toMatch(/STATUS\s+EVAL\s+ATTEMPT\s+RESULT\s+DURATION\s+COST/);
+    expect(out).toMatch(/✗ failed\s+fixtures\/button\s+@1[0-9a-z]{7}\s+file was not modified/);
+    expect(out).toMatch(/✓ passed\s+weather\/brooklyn\s+@1[0-9a-z]{7}\s+—/);
+    expect(out).not.toMatch(/@1[0-9a-z]{7}[✓✗]/);
+    expect(out).not.toContain("[E,X,");
+    expect(out).toContain("DRILL DOWN  niceeval show @<attempt>");
+  });
+
+  it("裸 show 使用稳定表头,不受自定义报告 locale chrome 影响", async () => {
+    const root = await seedComposedRoot();
+    process.env.NICEEVAL_LANG = "zh-CN";
+    try {
+      const { out, code } = await show(root, []);
+      expect(code).toBe(0);
+      expect(out).toMatch(/STATUS\s+EVAL\s+ATTEMPT\s+RESULT\s+DURATION\s+COST/);
+      expect(out).toContain("1 passed · 1 failed");
+      expect(out).not.toContain("没有可绘制的数据");
+      expect(out).toContain("compare/bub");
+    } finally {
+      process.env.NICEEVAL_LANG = "en";
+    }
+  });
+
+  it("窄终端截断长 eval 与原因,每一行都不超过终端显示宽度", async () => {
+    const root = await seedComposedRoot();
+    const { out, code } = await show(root, [], {}, 60);
+    expect(code).toBe(0);
+    expect(out).toContain("…");
+    for (const line of out.trimEnd().split("\n")) {
+      expect(stringWidth(line), line).toBeLessThanOrEqual(60);
+    }
+  });
+
+  it("多个 experiment 先显示紧凑比较表,再逐 experiment 显示 attempt 表", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/a", agent: "codex", model: "mini", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("q1", "passed", { estimatedCostUSD: 0.1 }),
+      res("q2", "failed", { estimatedCostUSD: 0.2 }),
+    ]);
+    await writeSnapshot(root, "2026-07-08T11-00-00-000Z", { experimentId: "compare/b", agent: "claude", model: "large", startedAt: "2026-07-08T11:00:00.000Z" }, [
+      res("q1", "passed", { estimatedCostUSD: 0.3 }),
+      res("q2", "passed", { estimatedCostUSD: 0.3 }),
+    ]);
+    const { out, code } = await show(root, []);
+    expect(code).toBe(0);
+    expect(out).toMatch(/COMPARISON\nEXPERIMENT\s+AGENT\s+MODEL\s+PASS\s+FAILED\s+ERROR\s+SKIP\s+DURATION\s+COST/);
+    expect(out).toContain("50% 1/2");
+    expect(out).toContain("100% 2/2");
+    expect(out.match(/^EXPERIMENT  compare\//gm)).toHaveLength(2);
+    expect(out).not.toContain("At least 2 experiments");
+  });
+
+  it("合成 Selection:每 experiment × eval 取最新判定;compose 不产生残缺警告", async () => {
     const root = await seedComposedRoot();
     const results = await openResults(root);
-    const selection = composeShowSelection(results);
+    const selection = selectCurrentResults(results);
     expect(selection.snapshots).toHaveLength(1);
     const evals = selection.snapshots[0].evals.map((e) => e.id).sort();
     expect(evals).toEqual(["fixtures/button", "weather/brooklyn"]);
     const button = selection.snapshots[0].evals.find((e) => e.id === "fixtures/button")!;
-    expect(button.attempts[0].result.outcome).toBe("failed"); // 新 run 的判决赢
+    expect(button.attempts[0].result.verdict).toBe("failed"); // 新快照的判定赢
     expect(selection.warnings).toEqual([]);
     // 对照:results.latest() 的最新快照是残缺的(这正是宿主要合成的原因)
     expect(results.latest().warnings.some((w) => w.kind === "partial-coverage")).toBe(true);
@@ -168,32 +226,28 @@ describe("榜单:跨 run 合成的现刻水位", () => {
 // ───────────────────────── 前缀过滤 ─────────────────────────
 
 describe("位置前缀收窄", () => {
-  it("前缀收窄选集覆盖的 eval;覆盖警告分母 = 已知并集 ∩ 范围", async () => {
+  it("前缀收窄 Selection 覆盖的 eval;覆盖警告分母 = 已知并集 ∩ 范围", async () => {
     const root = await makeRoot();
-    await writeRun(
+    await writeSnapshot(
       root,
       "2026-07-08T10-00-00-000Z",
-      summaryOf(
-        [
-          res({ id: "weather/brooklyn", experimentId: "compare/bub", startedAt: "2026-07-08T10:00:01.000Z" }),
-          res({ id: "algebra/quadratic", experimentId: "compare/bub", startedAt: "2026-07-08T10:00:02.000Z" }),
-        ],
-        {
-          startedAt: "2026-07-08T10:00:00.000Z",
-          // 已知并集包含一道从未落盘的题:algebra 范围外,不该刷 weather 范围的屏
-          snapshots: { "compare/bub": { knownEvalIds: ["weather/brooklyn", "weather/queens", "algebra/quadratic"] } },
-        },
-      ),
+      {
+        experimentId: "compare/bub",
+        startedAt: "2026-07-08T10:00:00.000Z",
+        // 已知并集包含一道从未落盘的题:algebra 范围外,不该刷 weather 范围的屏
+        knownEvalIds: ["weather/brooklyn", "weather/queens", "algebra/quadratic"],
+      },
+      [res("weather/brooklyn", "passed"), res("algebra/quadratic", "passed")],
     );
     const results = await openResults(root);
 
-    const weather = composeShowSelection(results, { patterns: ["weather"] });
+    const weather = selectCurrentResults(results, { patterns: ["weather"] });
     expect(weather.snapshots[0].evals.map((e) => e.id)).toEqual(["weather/brooklyn"]);
     // 分母 = {weather/brooklyn, weather/queens}:缺 queens → 1/2;algebra 的缺口不进来
     const coverage = weather.warnings.find((w) => w.kind === "partial-coverage");
     expect(coverage).toMatchObject({ covered: 1, total: 2 });
 
-    const algebra = composeShowSelection(results, { patterns: ["algebra"] });
+    const algebra = selectCurrentResults(results, { patterns: ["algebra"] });
     expect(algebra.warnings.filter((w) => w.kind === "partial-coverage")).toEqual([]);
   });
 
@@ -211,52 +265,36 @@ describe("位置前缀收窄", () => {
 describe("单 eval 详情", () => {
   it("attempt 行 + 断言明细 + artifacts 路径 + 下钻提示", async () => {
     const root = await makeRoot();
-    await writeRun(
+    await writeSnapshot(
       root,
       "2026-07-09T10-00-00-000Z",
-      summaryOf(
-        [
-          res({
-            id: "weather/brooklyn",
-            experimentId: "compare/codex",
-            agent: "codex",
-            outcome: "failed",
-            attempt: 0,
-            startedAt: "2026-07-09T10:00:01.000Z",
-            durationMs: 40_000,
-          }),
-          res({
-            id: "weather/brooklyn",
-            experimentId: "compare/codex",
-            agent: "codex",
-            outcome: "failed",
-            attempt: 1,
-            startedAt: "2026-07-09T10:00:42.000Z",
-            durationMs: 41_000,
-            usage: { inputTokens: 12_000, outputTokens: 300 },
-            estimatedCostUSD: 0.04,
-            artifactsDir: "artifacts/compare__codex__weather__brooklyn__1",
-            assertions: [
-              {
-                name: 'calledTool("get_weather")',
-                severity: "gate",
-                score: 0,
-                passed: false,
-                detail: "tool was never called",
-              },
-              { name: "succeeded()", severity: "gate", score: 1, passed: true },
-              {
-                name: 'judge("回答基于实时数据")',
-                severity: "soft",
-                score: 0.2,
-                passed: false,
-                detail: "reply invents a temperature without any tool call",
-              },
-            ],
-          }),
-        ],
-        { agent: "codex", startedAt: "2026-07-09T10:00:00.000Z" },
-      ),
+      { experimentId: "compare/codex", agent: "codex", startedAt: "2026-07-09T10:00:00.000Z" },
+      [
+        res("weather/brooklyn", "failed", { attempt: 0, durationMs: 40_000 }),
+        res("weather/brooklyn", "failed", {
+          attempt: 1,
+          durationMs: 41_000,
+          usage: { inputTokens: 12_000, outputTokens: 300 },
+          estimatedCostUSD: 0.04,
+          assertions: [
+            {
+              name: 'calledTool("get_weather")',
+              severity: "gate",
+              score: 0,
+              passed: false,
+              detail: "tool was never called",
+            },
+            { name: "succeeded()", severity: "gate", score: 1, passed: true },
+            {
+              name: 'judge("回答基于实时数据")',
+              severity: "soft",
+              score: 0.2,
+              passed: false,
+              detail: "reply invents a temperature without any tool call",
+            },
+          ],
+        }),
+      ],
     );
     const { out, code } = await show(root, ["weather/brooklyn"]);
     expect(code).toBe(0);
@@ -269,63 +307,44 @@ describe("单 eval 详情", () => {
     expect(out).toContain('✗ gate calledTool("get_weather") — tool was never called');
     expect(out).toContain("✓ gate succeeded()");
     expect(out).toContain('✗ soft judge("回答基于实时数据") — 0.2/1: reply invents a temperature');
-    expect(out).toContain("artifacts/compare__codex__weather__brooklyn__1/");
-    expect(out).toContain("next: niceeval show weather/brooklyn --transcript | --trace | --diff");
-  });
-
-  it("--attempt 指定不存在的 attempt:直说可用序号", async () => {
-    const root = await seedComposedRoot();
-    const { err, code } = await show(root, ["fixtures/button"], { attempt: 5 });
-    expect(code).toBe(1);
-    expect(err).toContain("Attempt 5 not found for fixtures/button");
+    // artifacts 路径 = <experiment-dir>/<snapshot-dir>/<evalId>/a<n>(root 相对)
+    expect(out).toContain("compare_codex/2026-07-09T10-00-00-000Z/weather/brooklyn/a1");
+    // 每 experiment 一行都带紧凑索引(locator + 失败原因),详情块再补一条精确的 attempt locator
+    expect(out).toMatch(/✗ failed\s+2 attempts.*@1[0-9a-z]{7}.*gate calledTool/);
+    expect(out).toMatch(/attempt locator: @1[0-9a-z]{7}/);
+    expect(out).toMatch(/next: niceeval show @1[0-9a-z]{7} \[--eval\|--execution\|--diff\]/);
   });
 });
 
 // ───────────────────────── --history:复印件不占行 ─────────────────────────
 
 describe("--history 时间轴", () => {
-  /** run1 真实执行;run2 resume 携带同一判决(身份键相同的复印件)+ 新题真实执行。 */
+  /** 快照1 真实执行;快照2 resume 携带同一判定(身份键相同的复印件)+ 新题真实执行。 */
   async function seedHistoryRoot(): Promise<string> {
     const root = await makeRoot();
-    const original = res({
-      id: "weather/brooklyn",
-      experimentId: "compare/bub",
-      outcome: "passed",
-      startedAt: "2026-07-07T09:00:01.000Z",
-      estimatedCostUSD: 0.03,
-      artifactsDir: "artifacts/compare__bub__weather__brooklyn__0",
-    });
-    await writeRun(
+    await writeSnapshot(
       root,
       "2026-07-07T09-00-00-000Z",
-      summaryOf([original], { startedAt: "2026-07-07T09:00:00.000Z" }),
+      { experimentId: "compare/bub", startedAt: "2026-07-07T09:00:00.000Z" },
+      [res("weather/brooklyn", "passed", { estimatedCostUSD: 0.03 })],
     );
-    // 复印件:同 id / attempt / startedAt,artifactBase 指回原 run(resume 合入的形状)
-    const { artifactsDir: _dropped, ...carriedBase } = original;
-    const carried: EvalResult = {
-      ...carriedBase,
-      artifactBase: "2026-07-07T09-00-00-000Z/artifacts/compare__bub__weather__brooklyn__0",
-    };
-    await writeRun(
+    // 复印件:同 id / attempt / startedAt(锚定原快照的 startedAt),artifactBase 指回原快照。
+    await writeSnapshot(
       root,
       "2026-07-09T10-00-00-000Z",
-      summaryOf(
-        [
-          carried,
-          res({
-            id: "weather/brooklyn",
-            experimentId: "compare/bub",
-            outcome: "failed",
-            attempt: 1,
-            startedAt: "2026-07-09T10:00:05.000Z",
-            estimatedCostUSD: 0.04,
-            assertions: [
-              { name: 'calledTool("get_weather")', severity: "gate", score: 0, passed: false },
-            ],
-          }),
-        ],
-        { startedAt: "2026-07-09T10:00:00.000Z" },
-      ),
+      { experimentId: "compare/bub", startedAt: "2026-07-09T10:00:00.000Z" },
+      [
+        res("weather/brooklyn", "passed", {
+          estimatedCostUSD: 0.03,
+          startedAt: "2026-07-07T09:00:00.000Z",
+          artifactBase: "compare_bub/2026-07-07T09-00-00-000Z/weather/brooklyn/a0",
+        }),
+        res("weather/brooklyn", "failed", {
+          attempt: 1,
+          estimatedCostUSD: 0.04,
+          assertions: [{ name: 'calledTool("get_weather")', severity: "gate", score: 0, passed: false }],
+        }),
+      ],
     );
     return root;
   }
@@ -335,18 +354,18 @@ describe("--history 时间轴", () => {
     const results = await openResults(root);
     const exp = results.experiments.find((e) => e.id === "compare/bub")!;
     const rows = evalHistory(exp, "weather/brooklyn");
-    // run2 里复印件被识别,真实执行只有:run1 的 passed + run2 的 failed(新 attempt)
+    // 快照2 里复印件被识别,真实执行只有:快照1 的 passed + 快照2 的 failed(新 attempt)
     expect(rows).toHaveLength(2);
-    expect(rows[0]).toMatchObject({ outcome: "failed", attempts: 1, costUSD: 0.04 });
+    expect(rows[0]).toMatchObject({ verdict: "failed", attempts: 1, costUSD: 0.04 });
     expect(rows[0].failedAssertion).toBe('gate calledTool("get_weather")');
-    expect(rows[1]).toMatchObject({ outcome: "passed", attempts: 1, costUSD: 0.03 });
+    expect(rows[1]).toMatchObject({ verdict: "passed", attempts: 1, costUSD: 0.03 });
 
     const { out, code } = await show(root, ["weather/brooklyn"], { history: true });
     expect(code).toBe(0);
     expect(out).toContain("compare/bub · 2 runs · passed 1/2");
     expect(out).toContain("2026-07-09T10-00");
     expect(out).toContain("2026-07-07T09-00");
-    // 复印件那份 passed 判决只出现一行(run1),不在 run2 再占一行
+    // 复印件那份 passed 判定只出现一行(快照1),不在快照2 再占一行
     expect(out.match(/✓ passed/g)).toHaveLength(1);
   });
 
@@ -355,7 +374,7 @@ describe("--history 时间轴", () => {
     const { out, code } = await show(root, [], { history: true });
     expect(code).toBe(0);
     expect(out).toContain("compare/bub · 2 runs");
-    // 每个快照一行(run2 折叠含携带的 passed 复印件,任一轮通过 → passed)
+    // 每个快照一行(快照2 折叠含携带的 passed 复印件,任一轮通过 → passed)
     expect(out.match(/1\/1 passed/g)).toHaveLength(2);
   });
 });
@@ -373,7 +392,7 @@ describe("--report 装载", () => {
         "const Custom = () => null;",
         "Custom[FACES] = {",
         "  web: () => null,",
-        "  text: (props, ctx) => `CUSTOM ${props.evals} · drill ${ctx.attemptCommand(props.ref)}`,",
+        "  text: (props, ctx) => `CUSTOM ${props.evals} · drill ${ctx.attemptCommand(props.locator)}`,",
         "};",
         "export default {",
         '  [Symbol.for("niceeval.report.definition")]: true,',
@@ -381,7 +400,7 @@ describe("--report 装载", () => {
         "    type: Custom,",
         "    props: {",
         "      evals: ctx.selection.snapshots.flatMap((s) => s.evals.map((e) => e.id)).sort().join(\",\"),",
-        "      ref: ctx.selection.snapshots[0].evals[0].attempts[0].ref,",
+        "      locator: ctx.selection.snapshots[0].evals[0].attempts[0].locator,",
         "    },",
         "  }),",
         "};",
@@ -392,7 +411,7 @@ describe("--report 装载", () => {
     return path;
   }
 
-  it("装载 + 注入选集 + attemptCommand 下钻命令", async () => {
+  it("装载 + 注入 Selection + attemptCommand 下钻命令", async () => {
     const root = await seedComposedRoot();
     const report = await writeReportFile(root);
     const { out, code } = await show(root, [], { report });
@@ -401,7 +420,7 @@ describe("--report 装载", () => {
     expect(out).toContain("drill niceeval show");
   });
 
-  it("位置前缀对 --report 生效:收窄注入选集覆盖的 eval", async () => {
+  it("位置前缀对 --report 生效:收窄注入 Selection 覆盖的 eval", async () => {
     const root = await seedComposedRoot();
     const report = await writeReportFile(root);
     const { out, code } = await show(root, ["weather"], { report });
@@ -409,18 +428,16 @@ describe("--report 装载", () => {
     expect(out).toContain("CUSTOM weather/brooklyn ·");
   });
 
-  it("--experiment 让选集只留该实验", async () => {
+  it("--experiment 让 Selection 只留该实验", async () => {
     const root = await seedComposedRoot();
-    await writeRun(
+    await writeSnapshot(
       root,
       "2026-07-09T11-00-00-000Z",
-      summaryOf(
-        [res({ id: "weather/brooklyn", experimentId: "compare/codex", agent: "codex", startedAt: "2026-07-09T11:00:01.000Z" })],
-        { agent: "codex", startedAt: "2026-07-09T11:00:00.000Z" },
-      ),
+      { experimentId: "compare/codex", agent: "codex", startedAt: "2026-07-09T11:00:00.000Z" },
+      [res("weather/brooklyn", "passed")],
     );
     const results = await openResults(root);
-    const selection = composeShowSelection(results, { experiment: "compare/codex" });
+    const selection = selectCurrentResults(results, { experiment: "compare/codex" });
     expect(selection.snapshots.map((s) => s.experimentId)).toEqual(["compare/codex"]);
   });
 
@@ -454,52 +471,254 @@ describe("--report 装载", () => {
     const { out, code } = await show(root, ["fixtures/button"], { report, diff: true });
     expect(code).toBe(0);
     expect(out).not.toContain("CUSTOM");
-    expect(out).toContain("no diff recorded"); // fixture 没有 diff 工件:如实说缺
+    expect(out).toContain("no diff recorded"); // fixture 没有 diff artifact:如实说缺
   });
 });
 
-// ───────────────────────── 证据切面:--transcript ─────────────────────────
+// ───────────────────────── 证据切面:--execution ─────────────────────────
 
-describe("--transcript", () => {
-  it("逐轮对话 + 截断标注(事件数 · 工具调用数 · 原始工件路径)", async () => {
+describe("--execution", () => {
+  it("标准事件流(消息);没有 OTel 时如实标 timing unavailable", async () => {
     const root = await makeRoot();
-    const writer = await createRunWriter(root, { producer: { name: "niceeval", version: "0.0.0" } });
-    const snap = writer.snapshot({
-      experiment: "compare/codex",
-      agent: "codex",
-      startedAt: "2026-07-09T10:00:00.000Z",
-    });
-    await snap.writeAttempt(
-      {
-        id: "weather/brooklyn",
-        outcome: "failed",
-        attempt: 2,
-        durationMs: 41_000,
-        assertions: [],
-      },
-      {
-        events: [
-          { type: "message", role: "user", text: "布鲁克林今天天气怎么样?" },
-          { type: "message", role: "assistant", text: "布鲁克林今天大约 24°C,晴。" },
-        ],
-      },
+    const dir = await writeSnapshot(
+      root,
+      "2026-07-09T10-00-00-000Z",
+      { experimentId: "compare/codex", agent: "codex", startedAt: "2026-07-09T10:00:00.000Z" },
+      [res("weather/brooklyn", "failed", { attempt: 2, durationMs: 41_000, hasEvents: true })],
     );
-    await writer.finish();
+    await writeFile(
+      join(dir, "weather/brooklyn/a2/events.json"),
+      JSON.stringify([
+        { type: "message", role: "user", text: "布鲁克林今天天气怎么样?" },
+        { type: "message", role: "assistant", text: "布鲁克林今天大约 24°C,晴。" },
+      ]),
+      "utf-8",
+    );
+
+    const { out, code } = await show(root, ["weather/brooklyn"], { execution: true });
+    expect(code).toBe(0);
+    expect(out).toMatch(/^@1[0-9a-z]{7} · weather\/brooklyn · compare\/codex · failed/m);
+    expect(out).toMatch(/USER\s+布鲁克林今天天气怎么样?/);
+    expect(out).toMatch(/ASSISTANT\s+布鲁克林今天大约 24°C,晴。/);
+    expect(out).toContain("timing unavailable · OTel trace was not collected");
+    expect(out).toContain("full events: ");
+    expect(out).toContain("events.json");
+  });
 
-    const { out, code } = await show(root, ["weather/brooklyn"], { transcript: true });
+  it("execution === null 时如实说没有事件记录", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(
+      root,
+      "2026-07-08T10-00-00-000Z",
+      { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" },
+      [res("weather/brooklyn", "passed")],
+    );
+    const { out, code } = await show(root, ["weather/brooklyn"], { execution: true });
     expect(code).toBe(0);
-    expect(out).toContain("attempt 3 · compare/codex · failed");
-    expect(out).toContain("[user]");
-    expect(out).toContain("布鲁克林今天天气怎么样?");
-    expect(out).toContain("[assistant]");
-    expect(out).toContain("(2 events · no tool calls · full stream: ");
-    expect(out).toContain("events.json)");
+    expect(out).toContain("no events recorded for this attempt");
   });
 
-  it("多个 eval 匹配时证据切面报错:先收窄到单个 eval", async () => {
+  it("不把未关联的 telemetry spans 混进 Agent 执行记录", async () => {
+    const root = await makeRoot();
+    const dir = await writeSnapshot(root, "2026-07-09T10-00-00-000Z", { experimentId: "compare/codex", startedAt: "2026-07-09T10:00:00.000Z" }, [
+      res("weather/brooklyn", "failed", { hasEvents: true, hasTrace: true }),
+    ]);
+    await writeFile(join(dir, "weather/brooklyn/a0/events.json"), JSON.stringify([{ type: "message", role: "user", text: "hello" }]), "utf-8");
+    await writeFile(join(dir, "weather/brooklyn/a0/trace.json"), JSON.stringify([{ traceId: "t", spanId: "s", name: "sdk.internal", startMs: 1, endMs: 2 }]), "utf-8");
+    const { out } = await show(root, ["weather/brooklyn"], { execution: true });
+    expect(out).toContain("USER");
+    expect(out).not.toMatch(/telemetry\s+sdk\.internal/);
+    expect(out).toContain("unlinked telemetry spans omitted");
+  });
+
+  it("多个 eval 匹配时证据切面报错:给紧凑索引(locator + 失败原因)而不是只报个数", async () => {
     const root = await seedComposedRoot();
-    const { err, code } = await show(root, [], { transcript: true });
+    const { err, code } = await show(root, [], { execution: true });
+    expect(code).toBe(1);
+    expect(err).toContain("matched 2 evals");
+    expect(err).toMatch(/✗ fixtures\/button\s+@1[0-9a-z]{7}/);
+    expect(err).toContain('gate fileChanged("src/components/Button.tsx")');
+  });
+});
+
+// ───────────────────────── 证据切面:--eval ─────────────────────────
+
+describe("--eval", () => {
+  it("evalSource === null 时如实说源码未捕获", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(
+      root,
+      "2026-07-08T10-00-00-000Z",
+      { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" },
+      [res("weather/brooklyn", "failed", { assertions: [{ name: "succeeded()", severity: "gate", score: 0, passed: false }] })],
+    );
+    const { out, code } = await show(root, ["weather/brooklyn"], { eval: true });
+    expect(code).toBe(0);
+    expect(out).toContain("eval source unavailable for this attempt");
+  });
+
+  it("有捕获的源码时:源码行标回断言,unmapped 断言单独成段,断言计票摘要", 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", {
+          assertions: [
+            {
+              name: 'calledTool("get_weather")',
+              severity: "gate",
+              score: 0,
+              passed: false,
+              detail: "tool was never called",
+              loc: { file: "evals/weather/brooklyn.eval.ts", line: 2 },
+            },
+            { name: "unlocated()", severity: "soft", score: 0.5, passed: false, detail: "no loc on this one" },
+          ],
+        }),
+      ],
+    );
+    const attemptDir = join(dir, "weather/brooklyn/a0");
+    const content = 'defineEval({\n  turn.calledTool("get_weather");\n});\n';
+    const sha = createHash("sha256").update(content).digest("hex");
+    await writeFile(join(attemptDir, "sources.json"), JSON.stringify([{ path: "evals/weather/brooklyn.eval.ts", sha256: sha }]), "utf-8");
+    await mkdir(join(dir, "sources"), { recursive: true });
+    await writeFile(join(dir, "sources", `${sha}.json`), JSON.stringify({ content }), "utf-8");
+
+    const { out, code } = await show(root, ["weather/brooklyn"], { eval: true });
+    expect(code).toBe(0);
+    expect(out).toContain(`eval source: evals/weather/brooklyn.eval.ts · sha256:${sha.slice(0, 8)}`);
+    // 源码行的缩进是语义的一部分,渲染必须原样保留(不按词重排折叠掉这两个空格)。
+    expect(out).toMatch(/2✗ {3}turn\.calledTool\("get_weather"\);/);
+    expect(out).toContain('gate · calledTool("get_weather") · tool was never called');
+    expect(out).toContain("unmapped assertions (1, no source location):");
+    expect(out).toContain("unlocated()");
+    expect(out).toContain("assertions: 1 gate failed · 1 soft below target");
+  });
+
+  it("源码行标注保留 group 与值断言的 expected/received", 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("manager", "failed", { assertions: [{ name: "equals(4)", severity: "gate", score: 0, passed: false, evidence: "1", group: "Issue 15193", loc: { file: "evals/manager.eval.ts", line: 1 } }] }),
+    ]);
+    const attemptDir = join(dir, "manager/a0");
+    const content = "t.check(actual, equals(expected));\n";
+    const sha = createHash("sha256").update(content).digest("hex");
+    await writeFile(join(attemptDir, "sources.json"), JSON.stringify([{ path: "evals/manager.eval.ts", sha256: sha }]), "utf-8");
+    await mkdir(join(dir, "sources"), { recursive: true });
+    await writeFile(join(dir, "sources", `${sha}.json`), JSON.stringify({ content }), "utf-8");
+    const { out } = await show(root, ["manager"], { eval: true });
+    expect(out).toContain("gate · Issue 15193 · equals(4) · expected 4 · received 1");
+  });
+});
+
+// ───────────────────────── @<locator> ─────────────────────────
+
+describe("show @<locator>", () => {
+  it("默认页直接解释失败断言的 group、期望值、实际值和源码位置", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("manager", "failed", { assertions: [{ name: "equals(4)", severity: "gate", score: 0, passed: false, evidence: "1", group: "Issue 15193", loc: { file: "evals/manager.eval.ts", line: 40, column: 11 } }] }),
+    ]);
+    const results = await openResults(root);
+    const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
+    const { out } = await show(root, [locator]);
+    expect(out).toContain("gate · Issue 15193");
+    expect(out).toContain("assertion: equals(4)");
+    expect(out).toContain("expected: 4");
+    expect(out).toContain("received: 1");
+    expect(out).toContain("source: evals/manager.eval.ts:40:11");
+  });
+  it("解析到对应 attempt,渲染紧凑全景(不当成 eval id 前缀匹配)", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("weather/brooklyn", "passed"),
+    ]);
+    const results = await openResults(root);
+    const attempt = results.experiments[0]!.latest.evals[0]!.attempts[0]!;
+    const locator = attempt.locator!;
+    expect(locator).toMatch(/^@1[0-9a-z]{7}$/);
+
+    const { out, code } = await show(root, [locator]);
+    expect(code).toBe(0);
+    expect(out).toContain(`${locator} · weather/brooklyn · compare/bub · passed`);
+    expect(out).toContain("attempt 1 · ");
+    expect(out).toContain("execution: unavailable (no events recorded for this attempt)");
+    expect(out).toContain("changes: diff unavailable");
+    expect(out).not.toContain("evidence:");
+    expect(out).not.toContain("available:");
+  });
+
+  it("available 只逐行列出实际存在的证据命令,不打印能力字母或合并式 next", async () => {
+    const root = await makeRoot();
+    const dir = await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("weather/brooklyn", "failed", { hasEvents: true }),
+    ]);
+    const attemptDir = join(dir, "weather/brooklyn/a0");
+    const content = "t.check(false, equals(true));\n";
+    const sha = createHash("sha256").update(content).digest("hex");
+    await writeFile(join(attemptDir, "sources.json"), JSON.stringify([{ path: "evals/weather.eval.ts", sha256: sha }]), "utf-8");
+    await mkdir(join(dir, "sources"), { recursive: true });
+    await writeFile(join(dir, "sources", `${sha}.json`), JSON.stringify({ content }), "utf-8");
+    await writeFile(join(attemptDir, "events.json"), JSON.stringify([{ type: "message", role: "user", text: "hello" }]), "utf-8");
+    const results = await openResults(root);
+    const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
+
+    const { out, code } = await show(root, [locator]);
+    expect(code).toBe(0);
+    expect(out).toContain(`available:\n  niceeval show ${locator} --eval\n  niceeval show ${locator} --execution`);
+    expect(out).not.toContain(`niceeval show ${locator} --diff`);
+    expect(out).not.toContain("evidence:");
+    expect(out).not.toContain("next:");
+  });
+
+  it("配 --execution 走证据切面,不是紧凑全景", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("weather/brooklyn", "passed"),
+    ]);
+    const results = await openResults(root);
+    const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
+
+    const { out, code } = await show(root, [locator], { execution: true });
+    expect(code).toBe(0);
+    expect(out).toContain(`${locator} · weather/brooklyn · compare/bub · passed`);
+    expect(out).toContain("no events recorded for this attempt");
+    // 紧凑全景独有的行不应该出现在证据切面输出里
+    expect(out).not.toContain("available:");
+  });
+
+  it("语法不对的 locator 报「not a valid attempt locator」,退出码 1,不崩", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("weather/brooklyn", "passed"),
+    ]);
+    const { err, code } = await show(root, ["@not-valid"]);
+    expect(code).toBe(1);
+    expect(err).toContain("not a valid attempt locator");
+  });
+
+  it("语法合法但索引里没有的 locator 报「No attempt found」,退出码 1,不崩", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("weather/brooklyn", "passed"),
+    ]);
+    const { err, code } = await show(root, ["@1nosuch1"]);
+    expect(code).toBe(1);
+    expect(err).toContain("No attempt found");
+  });
+
+  it("locator 与其它位置参数混用时报错,不静默只取第一个", async () => {
+    const root = await makeRoot();
+    await writeSnapshot(root, "2026-07-08T10-00-00-000Z", { experimentId: "compare/bub", startedAt: "2026-07-08T10:00:00.000Z" }, [
+      res("weather/brooklyn", "passed"),
+    ]);
+    const results = await openResults(root);
+    const locator = results.experiments[0]!.latest.evals[0]!.attempts[0]!.locator!;
+
+    const { err, code } = await show(root, [locator, "weather/brooklyn"]);
     expect(code).toBe(1);
-    expect(err).toContain("Narrow to a single eval id");
+    expect(err).toContain("must be the only positional argument");
   });
 });