niceeval 0.5.4 → 0.6.0

package/src/report/report.ts

@@ -1,26 +1,28 @@
 // defineReport:一份报告 = 一个报告文件(默认导出),两扇门(show / view)共用。
-// 宿主打开结果目录、按官方口径挑好选集、注入上下文;报告函数折数据、摆积木,
+// 宿主打开结果目录、按官方口径挑好 Selection、注入上下文;报告函数折数据、摆积木,
 // 返回一棵组件树。计算全部发生在报告函数体里(读句柄、await 折数据只在这里合法);
-// 渲染面是纯同步函数 —— 可达百 MB 的工件永远不进渲染路径。
+// 渲染面是纯同步函数 —— 可达百 MB 的 artifact 永远不进渲染路径。
 //
 // renderReportToText 是 text 宿主(show)的装载入口;web 宿主(view)的
 // renderReportToStaticHtml 在 ./web.ts(那一侧才 import react-dom)。宿主接线是下一波,
 // 这两个入口先以内部函数的身份可独立测试。
 
-import type { Results, Selection } from "../results/index.ts";
+import type { Results, Selection, SelectionWarning } from "../results/types.ts";
 import {
   createTextContext,
   renderNodeToText,
+  resolveReportTree,
   validateReportTree,
   type ReportNode,
   type TextRenderOptions,
 } from "./tree.ts";
-import { prepareDefaultReportData, runWithDefaultReportData } from "./default-report.tsx";
+import type { ReportLocale } from "./locale.ts";
 
 export interface ReportContext {
-  /** results.latest() 挑好的选集:现刻水位快照 + 结构化挑选警告,同默认报告口径。 */
+  /** 宿主按现刻水位规则挑好的 Selection:每个 experiment × eval 取跨快照合成的最新判定,外加结构化挑选
+   警告;show 的默认索引、view 默认报告与两者的 --report 使用同一选择口径。 */
   selection: Selection;
-  /** 默认挑法不合口径时,全量数据自己挑(见 docs/results-lib.md)。 */
+  /** 默认挑法不合口径时,全量数据自己挑(见 docs/feature/results/library.md)。 */
   results: Results;
 }
 
@@ -53,8 +55,18 @@ export function isReportDefinition(value: unknown): value is ReportDefinition {
 }
 
 /**
- * text 宿主的装载语义:build → 渲染前树校验 → 备好官方水位(DefaultReport 的数据)
- * → 遍历渲染 text 面。不需要 react-dom。
+ * 挑选警告的 text 形态:每条渲染好的 message 前缀 "! ",一行一条(与 RunOverview /
+ * overviewText 里 warnings 的 "! <message>" 约定一致)。宿主级前置块——不依赖报告是否
+ * 摆了 RunOverview,裸跑 / --report 都在报告顶上如实报残缺,不静默。
+ */
+function renderSelectionWarningsText(warnings: SelectionWarning[], _locale: ReportLocale): string {
+  return warnings.map((w) => `! ${w.message}`).join("\n");
+}
+
+/**
+ * text 宿主的装载语义:build → 渲染前解析数据组件(唯一的 await 边界)→ 树校验 → 遍历渲染
+ * text 面;Selection 有挑选警告时在报告顶部前置一块 "! <message>"；报告树里的 RunOverview
+ * 已经渲染同一条时不重复。不需要 react-dom。
  */
 export async function renderReportToText(
   definition: ReportDefinition,
@@ -62,8 +74,14 @@ export async function renderReportToText(
   options?: TextRenderOptions,
 ): Promise<string> {
   const node = await definition.build(ctx);
-  validateReportTree(node);
-  const defaultData = await prepareDefaultReportData(ctx.selection);
+  const resolved = await resolveReportTree(node);
+  validateReportTree(resolved);
   const textCtx = createTextContext(options);
-  return runWithDefaultReportData(defaultData, () => renderNodeToText(node, textCtx));
+  const body = renderNodeToText(resolved, textCtx);
+  // RunOverview can render the same Selection warnings as part of the user tree. Keep the
+  // host guarantee for reports that omit it, while never printing an identical warning twice.
+  const missingWarnings = ctx.selection.warnings.filter((warning) => !body.includes(`! ${warning.message}`));
+  return missingWarnings.length > 0
+    ? [renderSelectionWarningsText(missingWarnings, textCtx.locale), body].join("\n\n")
+    : body;
 }

package/src/report/text/faces.ts

@@ -2,10 +2,16 @@
 // 输出形态照 docs-site/zh/guides/report-components.mdx 的示例块;与 web 面共守
 // 诚实契约:排序随 better、samples < total 角标、缺数据 — 不补 0、截断报剩余。
 // 零 react、零 IO、纯同步 —— 这是 text 宿主不需要 react-dom 的那一半。
+// chrome 文案(注脚、verdict 词、截断提示)经 ctx.locale 查 locale 字典,
+// 默认 en 与历史输出逐字一致;数据(display、键、warnings message)不本地化。
 
 import type {
-  CaseListData,
+  AttemptListItem,
   DeltaData,
+  EvalListItem,
+  ExperimentListEvalRow,
+  ExperimentListItem,
+  GroupSummaryData,
   LineData,
   MatrixData,
   MetricColumn,
@@ -15,12 +21,28 @@ import type {
   TableData,
 } from "../types.ts";
 import type { TextContext } from "../tree.ts";
-import { MISSING_TEXT, formatDurationMs, formatMetricValue, formatPlainNumber, formatUSD } from "../format.ts";
-import { indentBlock, padDisplay, renderAlignedRows, textBar, wrapDisplay } from "./layout.ts";
+import type { TableColumn, TableRow } from "../primitives.tsx";
+import {
+  attemptItemReason,
+  capabilityBadge,
+  formatDurationMs,
+  formatMetricValue,
+  formatPlainNumber,
+  formatUSD,
+  verdictMark,
+} from "../format.ts";
+import { countText, localeText, resolveMetricLabel, type ReportLocale } from "../locale.ts";
+import { indentBlock, padDisplay, textBar, wrapDisplay } from "./layout.ts";
+import { renderTableText } from "./table.ts";
 import { renderCharPlot, renderCoordinateTable, type PlotPoint } from "./plot.ts";
 
 const MISSING_MARK = "—";
 
+/** 缺数据文案随 locale(en = "no data",与 MISSING_TEXT 一致)。 */
+function missingText(locale: ReportLocale): string {
+  return localeText(locale, "cell.missing");
+}
+
 /** 格子的文本形态:缺数据 —,覆盖不全带 samples/total 角标。 */
 export function cellText(cell: { value: number | null; display: string; samples: number; total: number }): string {
   if (cell.value === null) return MISSING_MARK;
@@ -29,23 +51,27 @@ export function cellText(cell: { value: number | null; display: string; samples:
 
 // ───────────────────────── RunOverview ─────────────────────────
 
-export function overviewText(data: OverviewData): string {
+export function overviewText(data: OverviewData, ctx: TextContext): string {
   const { totals, snapshots } = data;
+  const locale = ctx.locale;
   const runs = new Set(snapshots.map((s) => s.startedAt)).size;
   const latest = snapshots.map((s) => s.startedAt).sort().at(-1);
   const head = [
-    `${snapshots.length} ${snapshots.length === 1 ? "experiment" : "experiments"}`,
-    `${totals.evals} evals`,
-    `${totals.attempts} attempts`,
-    `composed from ${runs} ${runs === 1 ? "run" : "runs"}`,
-    ...(latest ? [`latest ${latest}`] : []),
+    countText(locale, "overview.experiments", snapshots.length),
+    localeText(locale, "overview.evalsCount", { n: totals.evals }),
+    localeText(locale, "overview.attemptsCount", { n: totals.attempts }),
+    // 通过率只渲染 computeCell 算好的同一个 MetricCell(cellText 复用缺数据/coverage 角标语义),
+    // 不从 passed/failed/errored 现场重算——与 web 面 RunOverview 同一份数据同一种读法。
+    `${localeText(locale, "overview.passRate")} ${cellText(totals.passRate)}`,
+    countText(locale, "composedFrom", runs),
+    ...(latest ? [localeText(locale, "latestRun", { run: latest })] : []),
   ].join(" · ");
   const tallies = [
-    `passed ${totals.passed}`,
-    `failed ${totals.failed}`,
-    `errored ${totals.errored}`,
-    `skipped ${totals.skipped}`,
-    totals.costUSD === null ? MISSING_TEXT : formatUSD(totals.costUSD),
+    `${localeText(locale, "verdict.passed")} ${totals.passed}`,
+    `${localeText(locale, "verdict.failed")} ${totals.failed}`,
+    `${localeText(locale, "verdict.errored")} ${totals.errored}`,
+    `${localeText(locale, "verdict.skipped")} ${totals.skipped}`,
+    totals.costUSD === null ? missingText(locale) : formatUSD(totals.costUSD),
     formatDurationMs(totals.durationMs),
   ].join(" · ");
   const lines = [head, tallies];
@@ -53,23 +79,93 @@ export function overviewText(data: OverviewData): string {
   return lines.join("\n");
 }
 
+// ───────────────────────── GroupSummary ─────────────────────────
+
+/**
+ * 一至两行:头行是通过率(GroupSummaryData.passRate.display,不重算比例)+ experiment/eval 数 +
+ * failed(+ errored,非零才列,与旧 GroupSelector 卡片一致)+ 总成本;第二行(有则加)是最后运行时间。
+ * 不依赖固定网格宽度,窄终端自然换行。
+ */
+export function groupSummaryText(data: GroupSummaryData, ctx: TextContext): string {
+  const locale = ctx.locale;
+  const head = [
+    `${localeText(locale, "overview.passRate")} ${cellText(data.passRate)}`,
+    countText(locale, "overview.experiments", data.experiments),
+    localeText(locale, "overview.evalsCount", { n: data.evals }),
+    `${localeText(locale, "verdict.failed")} ${data.verdicts.failed}`,
+    ...(data.verdicts.errored > 0 ? [`${localeText(locale, "verdict.errored")} ${data.verdicts.errored}`] : []),
+    data.totalCostUSD === null ? missingText(locale) : formatUSD(data.totalCostUSD),
+  ].join(" · ");
+  const lines = [head];
+  if (data.lastRunAt) lines.push(localeText(locale, "latestRun", { run: data.lastRunAt }));
+  return lines.join("\n");
+}
+
 // ───────────────────────── MetricTable ─────────────────────────
 
-export function tableText(data: TableData): string {
-  const header = [data.dimension, ...data.columns.map((c) => c.label)];
-  const rows = data.rows.map((row) => [
-    row.key,
-    ...data.columns.map((col) => {
+/** verdict 计票的紧凑文案("3 passed / 1 failed"):非零判定逐个列,全部为零如实 —。 */
+export function verdictTallyText(
+  verdicts: NonNullable<NonNullable<TableData["rows"][number]["meta"]>["verdicts"]>,
+  locale: ReportLocale,
+): string {
+  const parts: string[] = [];
+  for (const kind of ["passed", "failed", "errored", "skipped"] as const) {
+    if (verdicts[kind] > 0) parts.push(`${verdicts[kind]} ${localeText(locale, `verdict.${kind}`)}`);
+  }
+  return parts.length > 0 ? parts.join(" / ") : MISSING_MARK;
+}
+
+export function tableText(data: TableData, ctx: TextContext): string {
+  const locale = ctx.locale;
+  // meta 在场时补 Model / Agent / Verdicts 列(rows: "experiment" 的榜单 parity);
+  // 列序对齐 view 原生榜单:experiment、model、agent、指标列…、verdicts
+  const hasMeta = data.rows.some((row) => row.meta !== undefined);
+  const hasModel = data.rows.some((row) => row.meta?.model !== undefined);
+  const hasVerdicts = data.rows.some((row) => row.meta?.verdicts !== undefined);
+  const columns: TableColumn[] = [
+    { key: "dimension", header: data.dimension },
+    ...(hasMeta && hasModel ? [{ key: "model", header: localeText(locale, "table.model") }] : []),
+    ...(hasMeta ? [{ key: "agent", header: localeText(locale, "table.agent") }] : []),
+    // 指标列的键用序号,不用 metric name —— 维度名、"model" 这些键都在同一个命名空间里,
+    // 指标恰好叫 "agent" 时不能把 meta 列顶掉。
+    ...data.columns.map((c, i) => ({ key: `metric${i}`, header: resolveMetricLabel(c.label, locale, c.key) })),
+    ...(hasVerdicts ? [{ key: "verdicts", header: localeText(locale, "table.verdicts") }] : []),
+  ];
+  const rows: TableRow[] = data.rows.map((row) => {
+    const cells: Record<string, string | null> = { dimension: row.key };
+    if (hasMeta && hasModel) cells.model = row.meta?.model ?? null;
+    if (hasMeta) cells.agent = row.meta?.agent ?? null;
+    data.columns.forEach((col, i) => {
       const cell = (row.cells as Record<string, TableData["rows"][number]["cells"][string]>)[col.key];
-      return cell ? cellText(cell) : MISSING_MARK;
-    }),
-  ]);
-  return renderAlignedRows([header, ...rows]);
+      cells[`metric${i}`] = cell ? cellText(cell) : null;
+    });
+    if (hasVerdicts) cells.verdicts = row.meta?.verdicts ? verdictTallyText(row.meta.verdicts, locale) : null;
+    return { key: row.key, cells };
+  });
+  const table = renderTableText({ columns, rows, locale }, ctx);
+
+  // rows: "experiment" 专属的行摘要(eval/attempt 数 + 最后运行时间):表格列已经挤满
+  // dimension/model/agent/指标/verdicts,这几个数字挤进行键下面单独一行,与 web 面
+  // 「行键下的小 sub-line」同一份信息、同一种「不进列」的取舍。
+  const metaLines: string[] = [];
+  for (const row of data.rows) {
+    if (row.meta?.evals === undefined) continue;
+    const parts = [localeText(locale, "overview.evalsCount", { n: row.meta.evals })];
+    if (row.meta.attempts !== undefined && row.meta.attempts > row.meta.evals) {
+      parts.push(localeText(locale, "overview.attemptsCount", { n: row.meta.attempts }));
+    }
+    if (row.meta.lastRunAt) parts.push(localeText(locale, "latestRun", { run: row.meta.lastRunAt }));
+    metaLines.push(`  ${row.key}: ${parts.join(" · ")}`);
+  }
+
+  const blocks = [table, ...(metaLines.length > 0 ? [metaLines.join("\n")] : [])];
+  return blocks.join("\n\n");
 }
 
 // ───────────────────────── MetricMatrix ─────────────────────────
 
-export function matrixText(data: MatrixData): string {
+export function matrixText(data: MatrixData, ctx: TextContext): string {
+  // 表体全是维度键与 display,没有 chrome 文案;"next:" 是命令提示,不本地化。
   const rowKeys: string[] = [];
   const columnKeys: string[] = [];
   const byPosition = new Map<string, MatrixData["cells"][number]["cell"]>();
@@ -78,15 +174,19 @@ export function matrixText(data: MatrixData): string {
     if (!columnKeys.includes(entry.column)) columnKeys.push(entry.column);
     byPosition.set(JSON.stringify([entry.row, entry.column]), entry.cell);
   }
-  const header = [data.rows, ...columnKeys];
-  const rows = rowKeys.map((row) => [
-    row,
-    ...columnKeys.map((column) => {
+  const columns: TableColumn[] = [
+    { key: "dimension", header: data.rows },
+    ...columnKeys.map((column, i) => ({ key: `column${i}`, header: column })),
+  ];
+  const rows: TableRow[] = rowKeys.map((row) => {
+    const cells: Record<string, string | null> = { dimension: row };
+    columnKeys.forEach((column, i) => {
       const cell = byPosition.get(JSON.stringify([row, column]));
-      return cell ? cellText(cell) : MISSING_MARK; // 稀疏格子在文本里以 — 呈现,不编数
-    }),
-  ]);
-  const table = renderAlignedRows([header, ...rows]);
+      cells[`column${i}`] = cell ? cellText(cell) : null; // 稀疏格子在文本里以 — 呈现,不编数
+    });
+    return { key: row, cells };
+  });
+  const table = renderTableText({ columns, rows, locale: ctx.locale }, ctx);
 
   // 下钻命令:行维度是 eval 时,指向最值得看的一行(先挑有缺格的,再挑按 better 最差的)
   if (data.rows !== "eval" || rowKeys.length === 0) return table;
@@ -166,54 +266,75 @@ export function barsText(data: MatrixData): string {
 
 // ───────────────────────── Scoreboard ─────────────────────────
 
-export function scoreboardText(data: ScoreboardData): string {
+export function scoreboardText(data: ScoreboardData, ctx: TextContext): string {
+  const locale = ctx.locale;
   const subjectKeys: string[] = [];
   for (const row of data.rows) {
     for (const subject of row.subjects) {
       if (!subjectKeys.includes(subject.key)) subjectKeys.push(subject.key);
     }
   }
-  const header = [data.dimension, "total", ...subjectKeys];
-  const rows = data.rows.map((row) => [
-    row.key,
-    `${row.total.display}/${data.fullMarks}`,
-    ...subjectKeys.map((key) => {
+  const columns: TableColumn[] = [
+    { key: "dimension", header: data.dimension },
+    { key: "total", header: localeText(locale, "scoreboard.totalText") },
+    ...subjectKeys.map((key, i) => ({ key: `subject${i}`, header: key })),
+  ];
+  const rows: TableRow[] = data.rows.map((row) => {
+    const cells: Record<string, string | null> = {
+      dimension: row.key,
+      total: `${row.total.display}/${data.fullMarks}`,
+    };
+    subjectKeys.forEach((key, i) => {
       const subject = row.subjects.find((s) => s.key === key);
-      if (!subject) return MISSING_MARK;
+      if (!subject) {
+        cells[`subject${i}`] = null;
+        return;
+      }
       const score = `${formatPlainNumber(subject.earned)}/${formatPlainNumber(subject.possible)}`;
-      return subject.missing > 0 ? `${score} (${subject.missing} missing)` : score;
-    }),
-  ]);
-  const table = renderAlignedRows([header, ...rows]);
+      cells[`subject${i}`] =
+        subject.missing > 0
+          ? `${score} ${localeText(locale, "scoreboard.missingText", { n: subject.missing })}`
+          : score;
+    });
+    return { key: row.key, cells };
+  });
+  const table = renderTableText({ columns, rows, locale }, ctx);
   if (data.weights.length === 0) return table;
   // 实际生效的权重表 —— 成绩单可审计
   const weights = data.weights.map((w) => `${w.prefix} ×${w.weight}`).join(" · ");
-  return `${table}\nweights: ${weights} · others ×1`;
+  return `${table}\n${localeText(locale, "scoreboard.weights")} ${weights} · ${localeText(locale, "scoreboard.othersWeight")}`;
 }
 
 // ───────────────────────── MetricScatter ─────────────────────────
 
 const POINT_MARKS = "ABCDEFGHIJKLMNOPQRSTUVWXYZ";
 
-function axisLabel(col: MetricColumn): string {
-  return col.label;
+function axisLabel(col: MetricColumn, locale: ReportLocale): string {
+  return resolveMetricLabel(col.label, locale, col.key);
 }
 
 export function scatterText(data: ScatterData, ctx: TextContext): string {
+  const locale = ctx.locale;
   const drawable = data.rows.filter((r) => r.x.value !== null && r.y.value !== null);
   const missing = data.rows.length - drawable.length;
   const footnotes: string[] = [];
-  if (missing > 0) footnotes.push(`${missing} ${missing === 1 ? "point" : "points"} missing data`);
+  if (missing > 0) footnotes.push(countText(locale, "pointsMissing", missing));
 
+  const axes = { x: axisLabel(data.x, locale), y: axisLabel(data.y, locale) };
+  // 0 个可画点:x/y 指标没有可用数据(与 web 面同一事实)。
   if (drawable.length === 0) {
-    return [MISSING_TEXT, ...footnotes].join("\n");
+    return [localeText(locale, "scatter.noData", axes), ...footnotes].join("\n");
+  }
+  // 恰好 1 个可画点:成本 × 通过率的比较至少要两个实验,单点不成图。
+  if (drawable.length === 1) {
+    return [localeText(locale, "scatter.needTwo", axes), ...footnotes].join("\n");
   }
 
   // 点太密排不下时降级为坐标表,不硬挤
   if (drawable.length > POINT_MARKS.length || ctx.width < 44) {
     const table = renderCoordinateTable(
       drawable.map((r) => ({ key: r.key, x: r.x.display, y: r.y.display })),
-      { key: data.points, x: axisLabel(data.x), y: axisLabel(data.y) },
+      { key: data.points, x: axisLabel(data.x, locale), y: axisLabel(data.y, locale) },
     );
     return [table, ...footnotes].join("\n");
   }
@@ -238,26 +359,27 @@ export function scatterText(data: ScatterData, ctx: TextContext): string {
     width: ctx.width,
     points,
     lines: [...bySeries.values()].filter((l) => l.length > 1),
-    xLabel: `${axisLabel(data.x)}${invertX ? " (axis reversed: right = better)" : ""}`,
-    yLabel: axisLabel(data.y),
+    xLabel: `${axisLabel(data.x, locale)}${invertX ? ` ${localeText(locale, "scatter.axisReversed")}` : ""}`,
+    yLabel: axisLabel(data.y, locale),
     formatX: (v) => formatMetricValue(v, data.x.unit),
     formatY: (v) => formatMetricValue(v, data.y.unit),
     invertX,
     invertY: data.y.better === "lower",
   });
   const legend = drawable.map((r, i) => `${POINT_MARKS[i]} ${r.key}`).join("   ");
-  return [plot, "", `better → upper right`, legend, ...footnotes].join("\n");
+  return [plot, "", localeText(locale, "scatter.betterUpperRight"), legend, ...footnotes].join("\n");
 }
 
 // ───────────────────────── MetricLine ─────────────────────────
 
 export function lineText(data: LineData, ctx: TextContext): string {
+  const locale = ctx.locale;
   const drawable = data.rows.filter((r) => r.x !== null && r.y.value !== null);
   const missing = data.rows.length - drawable.length;
   const footnotes: string[] = [];
-  if (missing > 0) footnotes.push(`${missing} ${missing === 1 ? "point" : "points"} missing data`);
+  if (missing > 0) footnotes.push(countText(locale, "pointsMissing", missing));
 
-  if (drawable.length === 0) return [MISSING_TEXT, ...footnotes].join("\n");
+  if (drawable.length === 0) return [missingText(locale), ...footnotes].join("\n");
 
   // 系列 → 字母;无系列 = 单系列
   const seriesKeys: string[] = [];
@@ -269,7 +391,7 @@ export function lineText(data: LineData, ctx: TextContext): string {
   if (seriesKeys.length > POINT_MARKS.length || ctx.width < 44) {
     const table = renderCoordinateTable(
       drawable.map((r) => ({ key: r.series ? `${r.key} (${r.series})` : r.key, x: r.xDisplay, y: r.y.display })),
-      { key: "experiment", x: data.x.label, y: axisLabel(data.y) },
+      { key: "experiment", x: data.x.label, y: axisLabel(data.y, locale) },
     );
     return [table, ...footnotes].join("\n");
   }
@@ -292,65 +414,151 @@ export function lineText(data: LineData, ctx: TextContext): string {
     points,
     lines: lines.filter((l) => l.length > 1),
     xLabel: data.x.label,
-    yLabel: axisLabel(data.y),
+    yLabel: axisLabel(data.y, locale),
     formatX: (v) => formatMetricValue(v, data.x.unit),
     formatY: (v) => formatMetricValue(v, data.y.unit),
     invertY: data.y.better === "lower",
   });
   const legend = seriesKeys
-    .map((key, i) => `${POINT_MARKS[i]} ${key === "" ? data.y.label : key}`)
+    .map((key, i) => `${POINT_MARKS[i]} ${key === "" ? axisLabel(data.y, locale) : key}`)
     .join("   ");
   return [plot, "", legend, ...footnotes].join("\n");
 }
 
 // ───────────────────────── DeltaTable ─────────────────────────
 
-export function deltaText(data: DeltaData): string {
-  const header = ["pair", ...data.columns.map((c) => c.label)];
-  const rows = data.rows.map((row) => [
-    row.key,
-    ...data.columns.map((col) => {
+export function deltaText(data: DeltaData, ctx: TextContext): string {
+  const columns: TableColumn[] = [
+    { key: "pair", header: "pair" },
+    ...data.columns.map((c, i) => ({ key: `metric${i}`, header: resolveMetricLabel(c.label, ctx.locale, c.key) })),
+  ];
+  const rows: TableRow[] = data.rows.map((row) => {
+    const cells: Record<string, string | null> = { pair: row.key };
+    data.columns.forEach((col, i) => {
       const cell = (row.cells as Record<string, DeltaData["rows"][number]["cells"][string]>)[col.key];
-      if (!cell) return MISSING_MARK;
+      if (!cell) {
+        cells[`metric${i}`] = null;
+        return;
+      }
       const a = cell.a.value === null ? MISSING_MARK : cell.a.display;
       const b = cell.b.value === null ? MISSING_MARK : cell.b.display;
-      return `${a} → ${b}   ${cell.display}`;
-    }),
-  ]);
-  return renderAlignedRows([header, ...rows]);
+      cells[`metric${i}`] = `${a} → ${b}   ${cell.display}`;
+    });
+    return { key: row.key, cells };
+  });
+  return renderTableText({ columns, rows, locale: ctx.locale }, ctx);
 }
 
-// ───────────────────────── CaseList ─────────────────────────
+// ───────────────────────── 实体列表(ExperimentList / EvalList / AttemptList)─────────────────────────
+//
+// 三面共用的紧凑标记:`locator✓[E,X,⏱]`(判定符紧跟 locator,证据能力方括号紧跟判定符,
+// 中间不留空格)——docs-site/zh/guides/report-components.mdx「终端输出形成反馈闭环」定的形态。
+// ExperimentList / EvalList 逐 attempt 只列这一个标记 + 各自的原因/耗时摘要,不重复整段
+// niceeval show 命令;要看某个 attempt 的完整证据,agent 自己拼 `niceeval show <locator>`——
+// 命令模板只在 AttemptList(叶子层)展示完整断言明细时才值得,不在中间层重复。
 
-export function caseListText(data: CaseListData, ctx: TextContext): string {
-  if (data.rows.length === 0) return "No failed or errored attempts";
-  const lines: string[] = [];
-  for (const row of data.rows) {
-    const head = [
-      `✗ ${row.eval}`,
-      row.experimentId,
-      row.outcome,
-      formatDurationMs(row.durationMs),
-      ...(row.costUSD !== undefined ? [formatUSD(row.costUSD)] : []),
+function locatorBadge(item: { locator: string; verdict: AttemptListItem["verdict"]; capabilities: AttemptListItem["capabilities"] }): string {
+  return `${item.locator}${verdictMark(item.verdict)}${capabilityBadge(item.capabilities)}`;
+}
+
+// ── ExperimentList ──
+
+function experimentListEvalLine(row: ExperimentListEvalRow): string {
+  const badges = row.attempts.map(locatorBadge).join(" ");
+  const trailer =
+    row.verdict === "passed"
+      ? [formatDurationMs(row.duration.value ?? 0), row.cost.value === null ? undefined : formatUSD(row.cost.value)]
+          .filter((s): s is string => s !== undefined)
+          .join(" · ")
+      : (row.reason ?? "");
+  return `  ${verdictMark(row.verdict)} ${row.evalId}   ${badges}   ${trailer}`;
+}
+
+export function experimentListText(items: ExperimentListItem[], ctx: TextContext): string {
+  const locale = ctx.locale;
+  if (items.length === 0) return localeText(locale, "attemptList.empty");
+  const blocks = items.map((item) => {
+    const identity = item.model ? `${item.experimentId} · ${item.agent} · ${item.model}` : `${item.experimentId} · ${item.agent}`;
+    const summary = [
+      `${localeText(locale, "overview.passRate")} ${cellText(item.passRate)}`,
+      verdictTallyText(item.verdicts, locale),
+      localeText(locale, "overview.attemptsCount", { n: item.attempts }),
+      formatDurationMs(item.duration.value ?? 0),
+      item.cost.value === null ? missingText(locale) : formatUSD(item.cost.value),
     ].join(" · ");
-    lines.push(head);
-    if (row.error) {
-      lines.push(indentBlock(wrapDisplay(row.error, ctx.width - 4).join("\n"), "    "));
-    }
-    for (const assertion of row.failedAssertions) {
-      const summary = assertion.detail
-        ? `${assertion.name} — ${assertion.detail}`
-        : `${assertion.name} — score ${assertion.score}`;
-      lines.push(indentBlock(wrapDisplay(summary, ctx.width - 4).join("\n"), "    "));
-      if (assertion.evidence) {
-        lines.push(indentBlock(wrapDisplay(assertion.evidence, ctx.width - 6).join("\n"), "      "));
-      }
-    }
-    lines.push(`    → niceeval show ${row.eval}`);
+    const evalLines = item.evalRows.map(experimentListEvalLine);
+    return [identity, `  ${summary}`, ...evalLines].join("\n");
+  });
+  return blocks.join("\n\n");
+}
+
+// ── EvalList ──
+
+function evalListAttemptLine(item: AttemptListItem): string {
+  const reason = attemptItemReason(item);
+  return `  ${locatorBadge(item)}${reason ? ` · ${reason}` : ""}`;
+}
+
+export function evalListText(items: EvalListItem[], ctx: TextContext): string {
+  const locale = ctx.locale;
+  if (items.length === 0) return localeText(locale, "attemptList.empty");
+  const blocks = items.map((item) => {
+    const identity = `${item.evalId} · ${item.experimentId} · ${localeText(locale, `verdict.${item.verdict}`)}`;
+    const summary = [
+      localeText(locale, "attemptList.score", { score: cellText(item.score) }),
+      localeText(locale, "overview.attemptsCount", { n: item.attempts.length }),
+      `${formatDurationMs(item.duration.value ?? 0)} avg`,
+      item.cost.value === null ? `${missingText(locale)} avg` : `${formatUSD(item.cost.value)} avg`,
+    ].join(" · ");
+    const attemptLines = item.attempts.map(evalListAttemptLine);
+    return [identity, `  ${summary}`, ...attemptLines].join("\n");
+  });
+  return blocks.join("\n\n");
+}
+
+// ── AttemptList ──
+
+/** 一个 AttemptListItem 的完整 text 卡片:判定符 + locator + 身份 + 耗时/成本 + 证据能力,
+ * 然后逐条断言(gate 与 soft 都列,与 web 面的 AttemptRow 同一份材料)。 */
+function attemptListItemText(item: AttemptListItem, ctx: TextContext, locale: ReportLocale): string {
+  const head = [
+    `${verdictMark(item.verdict)} ${item.locator}`,
+    item.evalId,
+    item.experimentId,
+    formatDurationMs(item.durationMs),
+    ...(item.costUSD !== undefined ? [formatUSD(item.costUSD)] : []),
+    ...(capabilityBadge(item.capabilities) ? [capabilityBadge(item.capabilities)] : []),
+  ].join(" · ");
+  const lines = [head];
+  if (item.error) {
+    lines.push(indentBlock(wrapDisplay(item.error, ctx.width - 4).join("\n"), "    "));
   }
-  if (data.truncated > 0) {
-    lines.push("");
-    lines.push(`(${data.truncated} more not shown)`);
+  for (const assertion of item.assertions) {
+    const scoreText =
+      assertion.threshold !== undefined
+        ? `${formatPlainNumber(assertion.score)}/${formatPlainNumber(assertion.threshold)}`
+        : formatPlainNumber(assertion.score);
+    lines.push(
+      `  ${assertion.severity} ${assertion.name} · ${localeText(locale, `verdict.${assertion.passed ? "passed" : "failed"}`)}${assertion.severity === "soft" ? ` ${scoreText}` : ""}`,
+    );
+    if (assertion.detail) lines.push(indentBlock(wrapDisplay(assertion.detail, ctx.width - 4).join("\n"), "    "));
+    if (assertion.evidence) {
+      const limit = Math.max(240, ctx.width * 5);
+      const evidence =
+        assertion.evidence.length <= limit
+          ? assertion.evidence
+          : `${assertion.evidence.slice(0, limit)}… (${assertion.evidence.length - limit} more chars; open ${item.locator} for full evidence)`;
+      lines.push(indentBlock(wrapDisplay(evidence, ctx.width - 6).join("\n"), "      "));
+    }
   }
   return lines.join("\n");
 }
+
+export function attemptListText(items: AttemptListItem[], total: number | undefined, ctx: TextContext): string {
+  const locale = ctx.locale;
+  if (items.length === 0) return localeText(locale, "attemptList.empty");
+  const blocks = items.map((item) => attemptListItemText(item, ctx, locale));
+  const remaining = (total ?? items.length) - items.length;
+  if (remaining > 0) blocks.push(localeText(locale, "attemptList.truncatedText", { n: remaining }));
+  return blocks.join("\n\n");
+}