niceeval 0.6.0 → 0.6.2

package/dist/report/index.d.ts

@@ -1,5 +1,5 @@
-export { defineMetric, passRate, examScore, durationMs, tokens, costUSD, turns } from "./metrics.ts";
-export { flag } from "./flag.ts";
+export { defineMetric, taskPassRate, executionReliability, endToEndPassRate, examScore, durationMs, tokens, costUSD, turns, } from "./metrics.ts";
+export { flag, config } from "./flag.ts";
 export { defineReport, isReportDefinition, renderReportToText } from "./report.ts";
 export type { ReportContext, ReportDefinition } from "./report.ts";
 export { defineComponent } from "./tree.ts";
@@ -8,12 +8,13 @@ export { Row, Col, Section, Text, Style, Table } from "./primitives.tsx";
 export type { LayoutProps, SectionProps, StyleProps, TableColumn, TableProps, TableRow } from "./primitives.tsx";
 export { stringWidth, padDisplay as padEnd, padStartDisplay as padStart, wrapDisplay as wrapText, indentBlock as indent, textBar as bar, joinColumns as columns, } from "./text/layout.ts";
 export type { ColumnAlign } from "./text/layout.ts";
-export { CostPassRateComparison } from "./built-ins/index.ts";
+export { ExperimentComparison } from "./built-ins/index.ts";
+export type { ExperimentComparisonData, ExperimentComparisonGroupData, ExperimentComparisonProps, } from "./built-ins/index.ts";
 export { DEFAULT_REPORT_LOCALE, resolveMetricLabel } from "./locale.ts";
 export type { LocalizedLabel, ReportLocale } from "./locale.ts";
 export { AttemptList, DeltaTable, EvalList, ExperimentList, GroupSummary, MetricBars, MetricLine, MetricMatrix, MetricScatter, MetricTable, RunOverview, Scoreboard, } from "./components.tsx";
 export type { AttemptListProps, DeltaTableProps, EvalListProps, ExperimentListProps, GroupSummaryProps, MetricLineProps, MetricMatrixProps, MetricScatterProps, MetricTableProps, RunOverviewProps, ScoreboardProps, } from "./components.tsx";
 export type { AttemptListDataOptions, DeltaDataOptions, DeltaPair, LineDataOptions, MatrixDataOptions, ScatterDataOptions, ScoreboardDataOptions, TableDataOptions, } from "./compute.ts";
-export type { Aggregator, AttemptEvidenceCapabilities, AttemptListItem, AttemptLocator, DeltaData, Dimension, DimensionInput, EvalListItem, ExperimentListEvalRow, ExperimentListItem, GroupSummaryData, LineAxis, LineData, MatrixData, Metric, MetricAggregate, MetricCell, MetricColumn, OverviewData, FlagRef, ScatterData, ScoreboardData, SelectionWarning, TableData, TableRowMeta, } from "./types.ts";
+export type { Aggregator, AttemptListItem, AttemptLocator, AxisInput, ConfigRef, DeltaData, Dimension, DimensionInput, EvalListItem, ExperimentListEvalRow, ExperimentListItem, GroupSummaryData, LineAxis, LineData, MatrixData, Metric, MetricAggregate, MetricCell, MetricColumn, OverviewData, FlagRef, ScatterData, ScoreboardData, SelectionWarning, TableData, TableRowMeta, } from "./types.ts";
 export type { AttemptHandle, Results, Selection, Snapshot } from "../results/types.ts";
 export { experimentGroupOf } from "../shared/aggregate.ts";

package/dist/report/index.js

@@ -7,9 +7,9 @@
 // 不需要 react-dom(renderReportToText);web 宿主的 renderReportToStaticHtml 在
 // ./web.ts,只有那一侧 import react-dom。写报告文件的项目要装 react(.tsx 编译产物
 // import react/jsx-runtime)。
-// 指标与 flag
-export { defineMetric, passRate, examScore, durationMs, tokens, costUSD, turns } from "./metrics.js";
-export { flag } from "./flag.js";
+// 指标与维度读取器(flag / config)
+export { defineMetric, taskPassRate, executionReliability, endToEndPassRate, examScore, durationMs, tokens, costUSD, turns, } from "./metrics.js";
+export { flag, config } from "./flag.js";
 // 报告基座与双面组件基座
 export { defineReport, isReportDefinition, renderReportToText } from "./report.js";
 export { defineComponent } from "./tree.js";
@@ -19,8 +19,9 @@ export { Row, Col, Section, Text, Style, Table } from "./primitives.js";
 // 表格有 <Table> 承担,这里只给表以外的形态用 —— 尤其别拿 String.prototype.padEnd 对齐:
 // 它数 UTF-16 码元不数显示列宽,一带中文列就撕歪。renderAlignedRows 刻意不导出。
 export { stringWidth, padDisplay as padEnd, padStartDisplay as padStart, wrapDisplay as wrapText, indentBlock as indent, textBar as bar, joinColumns as columns, } from "./text/layout.js";
-// 内置报告(show / view 裸跑时报告槽的出厂填充,一份普通 ReportDefinition,无 renderer 特权)
-export { CostPassRateComparison } from "./built-ins/index.js";
+// 内置报告兼组合件(show / view 裸跑时报告槽的出厂填充;也可作组件整体引用,
+// `<ExperimentComparison data={await ExperimentComparison.data(selection)} />`),无 renderer 特权
+export { ExperimentComparison } from "./built-ins/index.js";
 // locale:官方组件 chrome 文案的语言(en / zh-CN);指标 label 可按 locale 给字典
 export { DEFAULT_REPORT_LOCALE, resolveMetricLabel } from "./locale.js";
 // 官方双面组件(各自带 data 计算函数;MetricBars.data = MetricMatrix.data)

package/dist/report/locale.d.ts

@@ -22,13 +22,27 @@ declare const en: {
     readonly "table.viewBreakdown": "Per-eval breakdown";
     /** <Table> 的 locator 列表头(行带 locator 时自动追加)。 */
     readonly "table.attempt": "attempt";
+    readonly "experimentList.experiment": "Experiment";
+    readonly "experimentList.avgDuration": "Avg duration";
+    readonly "experimentList.passRate": "End-to-end pass rate";
+    readonly "experimentList.tokens": "Tokens";
+    readonly "experimentList.estimatedCost": "Est. cost";
+    readonly "experimentList.result": "Result";
+    readonly "experimentList.status": "Status";
+    readonly "experimentList.evalAttempt": "Eval / Attempt";
+    readonly "experimentList.duration": "Duration";
+    readonly "experimentList.cost": "Cost";
+    readonly "experimentList.filterPlaceholder": "Filter experiments…";
+    readonly "experimentList.defaultModel": "default";
+    readonly "experimentList.flags": "Flags";
+    readonly "entityList.average": "{value} avg";
     /** <Table> 压到下限仍放不下时,从右侧丢列并如实报数。 */
     readonly "table.columnsHidden.one": "({n} more column not shown)";
     readonly "table.columnsHidden.other": "({n} more columns not shown)";
     readonly "overview.snapshots": "Snapshots";
     readonly "overview.evals": "Evals";
     readonly "overview.attempts": "attempts";
-    readonly "overview.passRate": "Pass rate";
+    readonly "overview.passRate": "End-to-end pass rate";
     readonly "overview.totalCost": "Total cost";
     readonly "overview.totalDuration": "Total duration";
     readonly "overview.source": "Source: {n} snapshots";
@@ -38,6 +52,13 @@ declare const en: {
     readonly "overview.attemptsCount": "{n} attempts";
     /** GroupSummary 的 KPI dt 标签(其余字段复用 verdict.* / overview.* 已有 key)。 */
     readonly "groupSummary.experiments": "Experiments";
+    readonly "experimentComparison.groups": "Experiment groups";
+    readonly "experimentComparison.group": "Experiment group";
+    readonly "experimentComparison.results": "Results";
+    readonly "experimentComparison.lastRun": "Last run";
+    readonly "experimentComparison.selectGroup": "Select experiment group {group}";
+    readonly "experimentComparison.command": "Details: {command}";
+    readonly "experimentComparison.empty": "No experiment groups";
     readonly "composedFrom.one": "composed from {n} run";
     readonly "composedFrom.other": "composed from {n} runs";
     readonly latestRun: "latest {run}";
@@ -46,14 +67,13 @@ declare const en: {
     readonly "scatter.axisReversed": "(axis reversed: right = better)";
     /** 0 个可画点:x/y 指标没有可用数据。 */
     readonly "scatter.noData": "No data to plot {x} × {y}";
-    /** 恰好 1 个可画点:比较至少要两个实验。 */
-    readonly "scatter.needTwo": "At least 2 experiments needed to compare {x} × {y}";
     readonly "pointsMissing.one": "{n} point missing data";
     readonly "pointsMissing.other": "{n} points missing data";
     readonly "attemptList.empty": "No attempts";
     readonly "attemptList.truncated": "and {n} more not shown";
     readonly "attemptList.truncatedText": "({n} more not shown)";
     readonly "attemptList.score": "score {score}";
+    readonly "attemptList.unavailable": "unavailable";
     readonly "attemptList.details": "details";
     readonly "scoreboard.total": "Total";
     readonly "scoreboard.totalText": "total";

package/dist/report/locale.js

@@ -2,7 +2,7 @@
 // 只覆盖组件自带的固定文案(verdict 词、缺数据、覆盖率角标、注脚、占位符等);
 // 数据本身(display、维度键、warnings 的 message)不经这里。
 // 刻意不 import src/i18n/(CLI 专用字典,locale 来源与 key 面完全不同);
-// zh-CN 译法与 src/view/app/i18n.ts 现有词保持一致(成功率/平均耗时/预估成本…)。
+// zh-CN 译法与默认 Reports 契约保持一致(端到端成功率/平均耗时/预估成本…)。
 export const DEFAULT_REPORT_LOCALE = "en";
 const en = {
     "verdict.passed": "passed",
@@ -25,13 +25,27 @@ const en = {
     "table.viewBreakdown": "Per-eval breakdown",
     /** <Table> 的 locator 列表头(行带 locator 时自动追加)。 */
     "table.attempt": "attempt",
+    "experimentList.experiment": "Experiment",
+    "experimentList.avgDuration": "Avg duration",
+    "experimentList.passRate": "End-to-end pass rate",
+    "experimentList.tokens": "Tokens",
+    "experimentList.estimatedCost": "Est. cost",
+    "experimentList.result": "Result",
+    "experimentList.status": "Status",
+    "experimentList.evalAttempt": "Eval / Attempt",
+    "experimentList.duration": "Duration",
+    "experimentList.cost": "Cost",
+    "experimentList.filterPlaceholder": "Filter experiments…",
+    "experimentList.defaultModel": "default",
+    "experimentList.flags": "Flags",
+    "entityList.average": "{value} avg",
     /** <Table> 压到下限仍放不下时,从右侧丢列并如实报数。 */
     "table.columnsHidden.one": "({n} more column not shown)",
     "table.columnsHidden.other": "({n} more columns not shown)",
     "overview.snapshots": "Snapshots",
     "overview.evals": "Evals",
     "overview.attempts": "attempts",
-    "overview.passRate": "Pass rate",
+    "overview.passRate": "End-to-end pass rate",
     "overview.totalCost": "Total cost",
     "overview.totalDuration": "Total duration",
     "overview.source": "Source: {n} snapshots",
@@ -41,6 +55,13 @@ const en = {
     "overview.attemptsCount": "{n} attempts",
     /** GroupSummary 的 KPI dt 标签(其余字段复用 verdict.* / overview.* 已有 key)。 */
     "groupSummary.experiments": "Experiments",
+    "experimentComparison.groups": "Experiment groups",
+    "experimentComparison.group": "Experiment group",
+    "experimentComparison.results": "Results",
+    "experimentComparison.lastRun": "Last run",
+    "experimentComparison.selectGroup": "Select experiment group {group}",
+    "experimentComparison.command": "Details: {command}",
+    "experimentComparison.empty": "No experiment groups",
     "composedFrom.one": "composed from {n} run",
     "composedFrom.other": "composed from {n} runs",
     "latestRun": "latest {run}",
@@ -49,14 +70,13 @@ const en = {
     "scatter.axisReversed": "(axis reversed: right = better)",
     /** 0 个可画点:x/y 指标没有可用数据。 */
     "scatter.noData": "No data to plot {x} × {y}",
-    /** 恰好 1 个可画点:比较至少要两个实验。 */
-    "scatter.needTwo": "At least 2 experiments needed to compare {x} × {y}",
     "pointsMissing.one": "{n} point missing data",
     "pointsMissing.other": "{n} points missing data",
     "attemptList.empty": "No attempts",
     "attemptList.truncated": "and {n} more not shown",
     "attemptList.truncatedText": "({n} more not shown)",
     "attemptList.score": "score {score}",
+    "attemptList.unavailable": "unavailable",
     "attemptList.details": "details",
     "scoreboard.total": "Total",
     "scoreboard.totalText": "total",
@@ -90,12 +110,26 @@ const zhCN = {
     "table.reason": "原因",
     "table.viewBreakdown": "逐题明细",
     "table.attempt": "Attempt",
+    "experimentList.experiment": "实验",
+    "experimentList.avgDuration": "平均耗时",
+    "experimentList.passRate": "端到端成功率",
+    "experimentList.tokens": "Tokens",
+    "experimentList.estimatedCost": "预估成本",
+    "experimentList.result": "结果",
+    "experimentList.status": "状态",
+    "experimentList.evalAttempt": "题目 / Attempt",
+    "experimentList.duration": "耗时",
+    "experimentList.cost": "成本",
+    "experimentList.filterPlaceholder": "筛选实验…",
+    "experimentList.defaultModel": "默认",
+    "experimentList.flags": "Flags",
+    "entityList.average": "平均 {value}",
     "table.columnsHidden.one": "(还有 {n} 列未列出)",
     "table.columnsHidden.other": "(还有 {n} 列未列出)",
     "overview.snapshots": "快照",
     "overview.evals": "题目",
     "overview.attempts": "尝试",
-    "overview.passRate": "通过率",
+    "overview.passRate": "端到端成功率",
     "overview.totalCost": "总成本",
     "overview.totalDuration": "总耗时",
     "overview.source": "数据来源:{n} 个快照",
@@ -104,6 +138,13 @@ const zhCN = {
     "overview.evalsCount": "{n} 道题",
     "overview.attemptsCount": "{n} 次 attempt",
     "groupSummary.experiments": "实验数",
+    "experimentComparison.groups": "实验组",
+    "experimentComparison.group": "实验组",
+    "experimentComparison.results": "结果",
+    "experimentComparison.lastRun": "最后运行",
+    "experimentComparison.selectGroup": "选择实验组 {group}",
+    "experimentComparison.command": "查看：{command}",
+    "experimentComparison.empty": "没有实验组",
     "composedFrom.one": "合成自 {n} 个 run",
     "composedFrom.other": "合成自 {n} 个 run",
     "latestRun": "最新 {run}",
@@ -111,13 +152,13 @@ const zhCN = {
     "scatter.betterUpperRight": "更好 → 右上",
     "scatter.axisReversed": "(轴反向:右 = 更好)",
     "scatter.noData": "{x} × {y} 没有可绘制的数据",
-    "scatter.needTwo": "至少需要 2 个实验才能比较 {x} × {y}",
     "pointsMissing.one": "{n} 个点缺数据",
     "pointsMissing.other": "{n} 个点缺数据",
     "attemptList.empty": "没有 attempt",
     "attemptList.truncated": "还有 {n} 条未列出",
     "attemptList.truncatedText": "(还有 {n} 条未列出)",
     "attemptList.score": "得分 {score}",
+    "attemptList.unavailable": "评不了",
     "attemptList.details": "详情",
     "scoreboard.total": "总分",
     "scoreboard.totalText": "总分",

package/dist/report/metrics.d.ts

@@ -8,7 +8,19 @@ import type { Metric } from "./types.ts";
 export declare function defineMetric<const Name extends string>(def: Metric<Name>): Metric<Name>;
 /** 单 attempt 成本:网关实测(usage.costUSD)优先于价格表估算(estimatedCostUSD);都缺 → null,不编 0。 */
 export declare function attemptCostUSD(result: EvalResult): number | null;
-export declare const passRate: Metric<"pass-rate">;
+/**
+ * 条件答题质量:passed = 1,failed = 0,errored 记 null 不进分母。
+ * 这是「已形成可信判定」条件下的诊断指标,不能简称默认成功率
+ * (docs/feature/reports/library.md「内置指标」)。
+ */
+export declare const taskPassRate: Metric<"task-pass-rate">;
+/** 执行可靠性:跑到可判定(passed / failed)= 1,errored = 0;skipped → null。 */
+export declare const executionReliability: Metric<"execution-reliability">;
+/**
+ * 端到端合成:passed = 1,failed / errored = 0;哪边拖累用
+ * taskPassRate / executionReliability 拆开看。
+ */
+export declare const endToEndPassRate: Metric<"end-to-end-pass-rate">;
 export declare const examScore: Metric<"exam-score">;
 export declare const durationMs: Metric<"duration">;
 export declare const tokens: Metric<"tokens">;

package/dist/report/metrics.js

@@ -1,19 +1,22 @@
 // defineMetric 与内置指标。
 //
 // null ≠ 0:null = 此 attempt 测不了这个指标(不进聚合);0 = 测了,结果是零(照常进)。
-// 哪个 verdict 落哪边必须显式表态,内置指标按 docs/feature/reports/library.md「内置指标」的表格:
+// 哪个 verdict 落哪边必须显式表态,内置指标按 docs/feature/reports/library.md「内置指标」的表格。
+// 三个通过率指标把「Agent 答错」与「基建没跑起来」拆开,不互相伪装:
 //
-//   指标(name)             skipped  errored          failed  passed        better
-//   passRate(pass-rate)     null     0                0       1             higher
-//   examScore(exam-score)   null     0                0       soft 均分      higher
-//   durationMs(duration)    null     实测             实测     实测          lower
-//   tokens(tokens)          null     实测;无 usage→null 同左   同左          lower
-//   costUSD(cost)           null     同上             同左     同左          lower
-//   turns(turns)             null     实测;o11y 缺失→null 同左  同左          lower
+//   指标(name)                                    skipped  errored          failed  passed        better
+//   taskPassRate(task-pass-rate)                   null     null             0       1             higher
+//   executionReliability(execution-reliability)    null     0                1       1             higher
+//   endToEndPassRate(end-to-end-pass-rate)         null     0                0       1             higher
+//   examScore(exam-score)                          null     0                0       soft 均分      higher
+//   durationMs(duration)                           null     实测             实测     实测          lower
+//   tokens(tokens)                                 null     实测;无 usage→null 同左   同左          lower
+//   costUSD(cost)                                  null     同上             同左     同左          lower
+//   turns(turns)                                   null     实测;o11y 缺失→null 同左  同左          lower
 //
 // 两档指标(docs/feature/reports/library.md「内置指标」):以上除 turns 外全部只读 attempt.result
 // 的瘦身字段——任何 producer、任何 copySnapshots artifacts 选择都算得出,内置报告
-// CostPassRateComparison 只用这一档。turns 读 attempt.o11y()(懒加载 artifact),发布时若
+// ExperimentComparison 只用这一档。turns 读 attempt.o11y()(懒加载 artifact),发布时若
 // o11y 没随行就诚实渲染缺数据「—」,不算 0——报告作者自己摆时心里要有这根弦,内置报告不用它。
 /**
  * 定义一个指标。内置指标与自定义指标是同一个类型,没有特权;
@@ -33,10 +36,56 @@ export function defineMetric(def) {
 export function attemptCostUSD(result) {
     return result.usage?.costUSD ?? result.estimatedCostUSD ?? null;
 }
-export const passRate = defineMetric({
-    name: "pass-rate",
-    label: { en: "Pass rate", "zh-CN": "成功率" },
-    description: "Share of evals that passed (skipped attempts excluded).",
+/**
+ * 条件答题质量:passed = 1,failed = 0,errored 记 null 不进分母。
+ * 这是「已形成可信判定」条件下的诊断指标,不能简称默认成功率
+ * (docs/feature/reports/library.md「内置指标」)。
+ */
+export const taskPassRate = defineMetric({
+    name: "task-pass-rate",
+    label: { en: "Task pass rate", "zh-CN": "可判定任务通过率" },
+    description: "Conditional task quality among attempts that formed a trustworthy verdict: passed = 1, failed = 0; errored is null.",
+    better: "higher",
+    unit: "%",
+    value(a) {
+        switch (a.result.verdict) {
+            case "passed":
+                return 1;
+            case "failed":
+                return 0;
+            default:
+                // errored = 没形成可信判定 → null 不进这个条件指标;skipped 同为 null。
+                return null;
+        }
+    },
+});
+/** 执行可靠性:跑到可判定(passed / failed)= 1,errored = 0;skipped → null。 */
+export const executionReliability = defineMetric({
+    name: "execution-reliability",
+    label: { en: "Execution reliability", "zh-CN": "执行可靠性" },
+    description: "Execution reliability: reached a trustworthy verdict (passed / failed) = 1, errored = 0.",
+    better: "higher",
+    unit: "%",
+    value(a) {
+        switch (a.result.verdict) {
+            case "passed":
+            case "failed":
+                return 1;
+            case "errored":
+                return 0;
+            default:
+                return null; // skipped
+        }
+    },
+});
+/**
+ * 端到端合成:passed = 1,failed / errored = 0;哪边拖累用
+ * taskPassRate / executionReliability 拆开看。
+ */
+export const endToEndPassRate = defineMetric({
+    name: "end-to-end-pass-rate",
+    label: { en: "End-to-end pass rate", "zh-CN": "端到端成功率" },
+    description: "End-to-end composite: passed = 1, failed / errored = 0. Split blame with taskPassRate and executionReliability.",
     better: "higher",
     unit: "%",
     value: (a) => a.result.verdict === "skipped" ? null : a.result.verdict === "passed" ? 1 : 0,
@@ -56,10 +105,12 @@ export const examScore = defineMetric({
         // failed 同理得 0:--strict 下被翻成 failed 的哪怕 soft 分不低也是 0(报告不重新判卷)。
         if (verdict !== "passed")
             return 0;
-        const soft = assertions.filter((x) => x.severity === "soft");
+        // unavailable 没有分数:不计入均分分母(评不了 ≠ 0 分;非 optional 的 unavailable
+        // 早已把 verdict 拖成 errored,走不到这个分支)。
+        const soft = assertions.filter((x) => x.severity === "soft" && x.outcome !== "unavailable");
         if (soft.length === 0)
             return 1;
-        return soft.reduce((sum, x) => sum + x.score, 0) / soft.length;
+        return soft.reduce((sum, x) => sum + (x.outcome === "unavailable" ? 0 : x.score), 0) / soft.length;
     },
 });
 export const durationMs = defineMetric({

package/dist/report/primitives.d.ts

@@ -30,6 +30,12 @@ export interface TableColumn {
     header: string;
     /** 对齐方向,默认 `"left"`;`"right"` 按显示宽度右对齐,数字列用。 */
     align?: ColumnAlign;
+    /**
+     * text 面:单元格折行后的最大物理行数,放不下的部分以 `…` 收口;省略则不限行数。
+     * 摘要类列(如比较列表的 Result)用它保证「格子是可扫读的预览」;完整值在下钻面。
+     * web 面不消费——网页的高度约束是组件自己的 CSS 决定。
+     */
+    maxLines?: number;
 }
 /** 一行的数据:身份键、已格式化的格子、可选的 attempt locator。 */
 export interface TableRow {

package/dist/report/react/AttemptList.d.ts

@@ -2,13 +2,13 @@ import type { ReactElement } from "react";
 import type { AttemptListItem } from "../types.ts";
 import type { AttemptLocator } from "../../results/locator.ts";
 import { type ReportLocale } from "../locale.ts";
-/** locator + 判定符 + 证据能力标记的普通 <a>,AttemptList/EvalList/ExperimentList 共用。 */
+/** locator + 判定符的普通 <a>,AttemptList/EvalList/ExperimentList 共用。 */
 export declare function AttemptLocatorBadge({ item, attemptHref, }: {
-    item: Pick<AttemptListItem, "locator" | "verdict" | "capabilities">;
+    item: Pick<AttemptListItem, "locator" | "verdict">;
     attemptHref: (locator: AttemptLocator) => string;
 }): ReactElement;
-/** 一条 Attempt 的完整卡片:AttemptList 自己的一项,也是 EvalList 展开区里的一行。 */
-export declare function AttemptRow({ item, attemptHref, locale, }: {
+/** 一条 Attempt 的比较卡片；完整 assertions 通过 locator 下钻，不在列表内展开。 */
+export declare function AttemptRow({ item, attemptHref, }: {
     item: AttemptListItem;
     attemptHref: (locator: AttemptLocator) => string;
     locale?: ReportLocale;

package/dist/report/react/AttemptList.js

@@ -1,18 +1,16 @@
 import { jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
 import { DEFAULT_REPORT_LOCALE, localeText } from "../locale.js";
+import { attemptItemReason } from "../format.js";
 import { colorClassForKey } from "./colors.js";
-import { capabilityBadge, cx, formatDurationMs, formatUSD, verdictMark } from "./format.js";
-/** locator + 判定符 + 证据能力标记的普通 <a>,AttemptList/EvalList/ExperimentList 共用。 */
+import { cx, formatDurationMs, formatUSD, verdictMark } from "./format.js";
+/** locator + 判定符的普通 <a>,AttemptList/EvalList/ExperimentList 共用。 */
 export function AttemptLocatorBadge({ item, attemptHref, }) {
-    const caps = capabilityBadge(item.capabilities);
-    return (_jsxs("a", { className: cx("nre-locator", `nre-verdict-${item.verdict}`), href: attemptHref(item.locator), children: [item.locator, _jsx("span", { className: "nre-locator-mark", children: verdictMark(item.verdict) }), caps && _jsx("span", { className: "nre-locator-caps", children: caps })] }));
+    return (_jsxs("a", { className: cx("nre-locator", `nre-verdict-${item.verdict}`), href: attemptHref(item.locator), children: [item.locator, _jsx("span", { className: "nre-locator-mark", children: verdictMark(item.verdict) })] }));
 }
-function AssertionRow({ assertion, locale }) {
-    return (_jsxs("li", { className: cx("nre-assertion", `nre-assertion-${assertion.severity}`, !assertion.passed && "nre-assertion-failed"), children: [_jsx("span", { className: "nre-assertion-severity", children: assertion.severity }), _jsx("span", { className: "nre-assertion-name", children: assertion.name }), _jsx("span", { className: "nre-assertion-score", children: localeText(locale, "attemptList.score", { score: assertion.score }) }), (assertion.detail || assertion.evidence) && (_jsxs("details", { className: "nre-assertion-more", children: [_jsx("summary", { children: localeText(locale, "attemptList.details") }), assertion.detail && _jsx("p", { className: "nre-assertion-detail", children: assertion.detail }), assertion.evidence && _jsx("blockquote", { className: "nre-assertion-evidence", children: assertion.evidence })] }))] }));
-}
-/** 一条 Attempt 的完整卡片:AttemptList 自己的一项,也是 EvalList 展开区里的一行。 */
-export function AttemptRow({ item, attemptHref, locale = DEFAULT_REPORT_LOCALE, }) {
-    return (_jsxs("li", { className: cx("nre-attempt", `nre-attempt-${item.verdict}`), children: [_jsxs("div", { className: "nre-attempt-head", children: [_jsx(AttemptLocatorBadge, { item: item, attemptHref: attemptHref }), _jsx("span", { className: "nre-attempt-eval", children: item.evalId }), _jsx("span", { className: "nre-attempt-experiment", children: item.experimentId }), _jsx("span", { className: cx("nre-attempt-agent", "nre-key", colorClassForKey(item.agent)), children: item.agent }), _jsx("span", { className: "nre-attempt-duration", children: formatDurationMs(item.durationMs) }), item.costUSD !== undefined && _jsx("span", { className: "nre-attempt-cost", children: formatUSD(item.costUSD) })] }), item.error && _jsx("p", { className: "nre-attempt-error", children: item.error }), item.assertions.length > 0 && (_jsx("ul", { className: "nre-assertions", children: item.assertions.map((a, i) => (_jsx(AssertionRow, { assertion: a, locale: locale }, i))) }))] }));
+/** 一条 Attempt 的比较卡片；完整 assertions 通过 locator 下钻，不在列表内展开。 */
+export function AttemptRow({ item, attemptHref, }) {
+    const reason = attemptItemReason(item);
+    return (_jsxs("li", { className: cx("nre-attempt", `nre-attempt-${item.verdict}`), children: [_jsxs("div", { className: "nre-attempt-head", children: [_jsx(AttemptLocatorBadge, { item: item, attemptHref: attemptHref }), _jsx("span", { className: "nre-attempt-eval", children: item.evalId }), _jsx("span", { className: "nre-attempt-experiment", children: item.experimentId }), _jsx("span", { className: cx("nre-attempt-agent", "nre-key", colorClassForKey(item.agent)), children: item.agent }), _jsx("span", { className: "nre-attempt-duration", children: formatDurationMs(item.durationMs) }), item.costUSD !== undefined && _jsx("span", { className: "nre-attempt-cost", children: formatUSD(item.costUSD) })] }), reason && _jsx("p", { className: "nre-attempt-result", children: reason })] }));
 }
 export function AttemptList({ items, total, attemptHref, className, locale = DEFAULT_REPORT_LOCALE, }) {
     const remaining = (total ?? items.length) - items.length;

package/dist/report/react/EvalList.d.ts

@@ -1,6 +1,6 @@
 import type { ReactElement } from "react";
-import type { EvalListItem } from "../types.ts";
 import type { AttemptLocator } from "../../results/locator.ts";
+import type { EvalListItem } from "../types.ts";
 import { type ReportLocale } from "../locale.ts";
 export declare function EvalList({ items, attemptHref, className, locale, }: {
     items: EvalListItem[];

package/dist/report/react/ExperimentComparison.d.ts

@@ -0,0 +1,8 @@
+import type { ReactElement } from "react";
+import type { ExperimentComparisonData } from "../built-ins/experiment-comparison.tsx";
+import { type ReportLocale } from "../locale.ts";
+export declare function ExperimentComparisonView({ data, className, locale, }: {
+    data: ExperimentComparisonData;
+    className?: string;
+    locale?: ReportLocale;
+}): ReactElement;

package/dist/report/react/ExperimentComparison.js

@@ -0,0 +1,11 @@
+import { jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
+import { DEFAULT_REPORT_LOCALE, localeText } from "../locale.js";
+import { ExperimentList, MetricScatter } from "../components.js";
+import { GroupSummary } from "./GroupSummary.js";
+import { cx } from "./format.js";
+export function ExperimentComparisonView({ data, className, locale = DEFAULT_REPORT_LOCALE, }) {
+    if (data.groups.length === 0) {
+        return (_jsx("div", { className: cx("nre", "nre-experiment-comparison", className), children: _jsx("p", { className: "nre-experiment-groups-empty", children: localeText(locale, "experimentComparison.empty") }) }));
+    }
+    return (_jsxs("div", { className: cx("nre", "nre-experiment-comparison", className), "data-nre-experiment-groups": true, children: [_jsx("div", { className: "nre-experiment-group-tabs", role: "tablist", "aria-label": localeText(locale, "experimentComparison.groups"), children: data.groups.map((group, index) => (_jsxs("div", { className: "nre-experiment-group-tab", role: "tab", tabIndex: index === 0 ? 0 : -1, "aria-selected": index === 0 ? "true" : "false", "aria-label": localeText(locale, "experimentComparison.selectGroup", { group: group.key }), "data-nre-experiment-group-select": index, children: [_jsx("strong", { className: "nre-experiment-group-name", children: group.key }), _jsx(GroupSummary, { data: group.summary, locale: locale })] }, group.key))) }), _jsx("div", { className: "nre-experiment-group-panels", children: data.groups.map((group, index) => (_jsxs("details", { className: "nre-experiment-group-panel", "data-nre-experiment-group-panel": index, role: "tabpanel", open: index === 0, children: [_jsx("summary", { children: group.key }), _jsx(MetricScatter, { data: group.scatter, locale: locale }), _jsx(ExperimentList, { items: group.experiments, filter: true, locale: locale, relativeTo: group.key })] }, group.key))) })] }));
+}

package/dist/report/react/ExperimentList.d.ts

@@ -1,10 +1,12 @@
 import type { ReactElement } from "react";
-import type { ExperimentListItem } from "../types.ts";
 import type { AttemptLocator } from "../../results/locator.ts";
+import type { ExperimentListItem } from "../types.ts";
 import { type ReportLocale } from "../locale.ts";
-export declare function ExperimentList({ items, attemptHref, className, locale, }: {
+export declare function ExperimentList({ items, attemptHref, filter, className, locale, relativeTo, }: {
     items: ExperimentListItem[];
     attemptHref: (locator: AttemptLocator) => string;
+    filter?: boolean;
     className?: string;
     locale?: ReportLocale;
+    relativeTo?: string;
 }): ReactElement;

package/dist/report/react/ExperimentList.js

@@ -1,15 +1,65 @@
 import { Fragment as _Fragment, jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
+import { attemptItemReason, experimentDisplayName } from "../format.js";
 import { DEFAULT_REPORT_LOCALE, localeText } from "../locale.js";
-import { MetricCellView } from "./cell.js";
 import { AttemptLocatorBadge } from "./AttemptList.js";
+import { MetricCellView } from "./cell.js";
 import { colorClassForKey } from "./colors.js";
 import { cx, formatDurationMs, formatUSD, verdictMark } from "./format.js";
-function EvalRow({ row, attemptHref, }) {
-    return (_jsxs("li", { className: cx("nre-experiment-eval-row", `nre-eval-${row.verdict}`), children: [_jsx("span", { className: cx("nre-eval-verdict", `nre-verdict-${row.verdict}`), children: verdictMark(row.verdict) }), _jsx("span", { className: "nre-eval-id", children: row.evalId }), _jsx("span", { className: "nre-eval-attempt-badges", children: row.attempts.map((attempt) => (_jsx(AttemptLocatorBadge, { item: attempt, attemptHref: attemptHref }, attempt.locator))) }), row.verdict === "passed" ? (_jsxs("span", { className: "nre-eval-avg", children: [formatDurationMs(row.duration.value ?? 0), row.cost.value !== null && _jsxs(_Fragment, { children: [" \u00B7 ", formatUSD(row.cost.value)] })] })) : (_jsx("span", { className: "nre-eval-reason", children: row.reason }))] }));
+const verdictOrder = ["passed", "failed", "errored", "skipped"];
+function passRateTone(value) {
+    if (value === null)
+        return undefined;
+    if (value >= 0.8)
+        return "nre-good";
+    if (value >= 0.5)
+        return "nre-warn";
+    return "nre-bad";
+}
+function formatDate(value, locale) {
+    const date = new Date(value);
+    if (Number.isNaN(date.valueOf()))
+        return value;
+    return new Intl.DateTimeFormat(locale === "zh-CN" ? "zh-CN" : "en", {
+        month: "short",
+        day: "numeric",
+        hour: "2-digit",
+        minute: "2-digit",
+    }).format(date);
+}
+function VerdictSummary({ item, locale }) {
+    const parts = verdictOrder
+        .filter((verdict) => item.verdicts[verdict] > 0)
+        .map((verdict) => `${item.verdicts[verdict]} ${localeText(locale, `verdict.${verdict}`)}`);
+    return _jsx("span", { className: "nre-experiment-pill", children: parts.join(" / ") || "—" });
+}
+function ExperimentAttemptRow({ attempt, last, attemptHref, }) {
+    const reason = attemptItemReason(attempt);
+    return (_jsxs("li", { className: cx("nre-experiment-attempt-row", `nre-eval-${attempt.verdict}`), children: [_jsx("span", { className: "nre-attempt-branch", "aria-hidden": "true", children: last ? "└─" : "├─" }), _jsx("span", { className: "nre-eval-attempt-badges", children: _jsx(AttemptLocatorBadge, { item: attempt, attemptHref: attemptHref }) }), _jsxs("span", { className: "nre-eval-attempt-metrics", children: [formatDurationMs(attempt.durationMs), attempt.costUSD !== undefined && _jsxs(_Fragment, { children: [" \u00B7 ", formatUSD(attempt.costUSD)] })] }), _jsx("span", { className: "nre-eval-reason", children: reason ?? "—" })] }));
+}
+function EvalAttempts({ row, attemptHref, locale, }) {
+    const duration = row.duration.value === null ? localeText(locale, "cell.missing") : formatDurationMs(row.duration.value);
+    const cost = row.cost.value === null ? localeText(locale, "cell.missing") : formatUSD(row.cost.value);
+    return (_jsxs("li", { className: "nre-experiment-eval", children: [_jsxs("div", { className: cx("nre-experiment-eval-header", `nre-eval-${row.verdict}`), children: [_jsx("span", { className: cx("nre-eval-verdict", `nre-verdict-${row.verdict}`), children: verdictMark(row.verdict) }), _jsx("span", { className: "nre-eval-id", children: row.evalId }), _jsx("span", { className: "nre-eval-attempt-count", children: localeText(locale, "overview.attemptsCount", { n: row.attempts.length }) }), _jsxs("span", { className: "nre-eval-rollup", children: [localeText(locale, "entityList.average", { value: duration }), " · ", localeText(locale, "entityList.average", { value: cost })] })] }), _jsx("ul", { className: "nre-experiment-attempts", children: row.attempts.map((attempt, index) => (_jsx(ExperimentAttemptRow, { attempt: attempt, last: index === row.attempts.length - 1, attemptHref: attemptHref }, attempt.locator))) })] }));
+}
+function Flags({ flags, locale }) {
+    if (!flags || Object.keys(flags).length === 0)
+        return null;
+    return (_jsxs("div", { className: "nre-experiment-flags", children: [_jsx("span", { children: localeText(locale, "experimentList.flags") }), Object.entries(flags).map(([key, value]) => (_jsxs("b", { children: [key, "=", String(value)] }, key)))] }));
 }
-function ExperimentCard({ item, attemptHref, locale, }) {
-    return (_jsx("li", { className: "nre-experiment-entry", children: _jsxs("details", { className: "nre-experiment-details", children: [_jsxs("summary", { className: "nre-experiment-summary", children: [_jsx("span", { className: cx("nre-experiment-id", "nre-key", colorClassForKey(item.experimentId)), children: item.experimentId }), _jsx("span", { className: "nre-experiment-agent", children: item.agent }), item.model && _jsx("span", { className: "nre-experiment-model", children: item.model }), _jsx("span", { className: "nre-experiment-cell", children: _jsx(MetricCellView, { cell: item.passRate, attemptHref: attemptHref, locale: locale }) }), _jsx("span", { className: "nre-experiment-cell", children: _jsx(MetricCellView, { cell: item.cost, attemptHref: attemptHref, locale: locale }) }), _jsx("span", { className: "nre-experiment-cell", children: _jsx(MetricCellView, { cell: item.duration, attemptHref: attemptHref, locale: locale }) }), _jsxs("span", { className: "nre-experiment-meta-sub", children: [localeText(locale, "overview.evalsCount", { n: item.evals }), item.attempts > item.evals ? ` · ${localeText(locale, "overview.attemptsCount", { n: item.attempts })}` : "", ` · ${localeText(locale, "latestRun", { run: item.lastRunAt })}`] })] }), _jsx("ul", { className: "nre-experiment-evals", children: item.evalRows.map((row) => (_jsx(EvalRow, { row: row, attemptHref: attemptHref }, row.evalId))) })] }) }));
+function ExperimentRow({ item, attemptHref, locale, relativeTo, }) {
+    return (_jsxs("details", { className: "nre-experiment-entry", children: [_jsxs("summary", { className: "nre-experiment-summary", children: [_jsxs("span", { className: "nre-experiment-name", "data-sort-value": item.experimentId, children: [_jsx("b", { className: cx("nre-experiment-id", "nre-key", colorClassForKey(item.experimentId)), children: experimentDisplayName(item.experimentId, relativeTo) }), _jsxs("small", { children: [localeText(locale, "overview.evalsCount", { n: item.evals }), item.attempts > item.evals ? ` · ${localeText(locale, "overview.attemptsCount", { n: item.attempts })}` : "", ` · ${formatDate(item.lastRunAt, locale)}`] })] }), _jsx("span", { "data-sort-value": item.model ?? "", children: item.model ?? localeText(locale, "experimentList.defaultModel") }), _jsx("span", { "data-sort-value": item.agent, children: item.agent }), _jsx("span", { className: "nre-num", "data-sort-value": item.duration.value ?? "", children: _jsx(MetricCellView, { cell: item.duration, locale: locale }) }), _jsx("span", { className: cx("nre-num", passRateTone(item.passRate.value)), "data-sort-value": item.passRate.value ?? "", children: _jsx(MetricCellView, { cell: item.passRate, locale: locale }) }), _jsx("span", { className: "nre-num", "data-sort-value": item.tokens.value ?? "", children: _jsx(MetricCellView, { cell: item.tokens, locale: locale }) }), _jsx("span", { className: "nre-num", "data-sort-value": item.cost.value ?? "", children: _jsx(MetricCellView, { cell: item.cost, locale: locale }) }), _jsx("span", { "data-sort-value": item.verdicts.passed, children: _jsx(VerdictSummary, { item: item, locale: locale }) })] }), _jsxs("div", { className: "nre-experiment-detail", children: [_jsx(Flags, { flags: item.flags, locale: locale }), _jsx("ul", { className: "nre-experiment-evals", children: item.evalRows.map((row) => (_jsx(EvalAttempts, { row: row, attemptHref: attemptHref, locale: locale }, row.evalId))) })] })] }));
 }
-export function ExperimentList({ items, attemptHref, className, locale = DEFAULT_REPORT_LOCALE, }) {
-    return (_jsxs("div", { className: cx("nre", "nre-experiment-list", className), children: [items.length === 0 && _jsx("p", { className: "nre-experiment-list-empty", children: localeText(locale, "attemptList.empty") }), _jsx("ul", { className: "nre-experiments", children: items.map((item) => (_jsx(ExperimentCard, { item: item, attemptHref: attemptHref, locale: locale }, item.experimentId))) })] }));
+export function ExperimentList({ items, attemptHref, filter = false, className, locale = DEFAULT_REPORT_LOCALE, relativeTo, }) {
+    const labels = [
+        localeText(locale, "experimentList.experiment"),
+        localeText(locale, "table.model"),
+        localeText(locale, "table.agent"),
+        localeText(locale, "experimentList.avgDuration"),
+        localeText(locale, "experimentList.passRate"),
+        localeText(locale, "experimentList.tokens"),
+        localeText(locale, "experimentList.estimatedCost"),
+        localeText(locale, "experimentList.result"),
+    ];
+    const board = (_jsxs("div", { className: "nre-experiment-table", children: [_jsx("div", { className: "nre-experiment-head", children: labels.map((label, index) => (_jsx("button", { type: "button", "data-nre-experiment-sort": index, className: index === 4 ? "nre-sort-desc" : undefined, children: label }, label))) }), items.length === 0 && _jsx("p", { className: "nre-experiment-list-empty", children: localeText(locale, "attemptList.empty") }), items.map((item) => (_jsx(ExperimentRow, { item: item, attemptHref: attemptHref, locale: locale, relativeTo: relativeTo }, item.experimentId)))] }));
+    return (_jsxs("div", { className: cx("nre", "nre-experiment-list", filter && "nre-experiment-table-wrap", className), children: [filter && (_jsx("input", { className: "nre-filter", "data-nre-experiment-filter": "", type: "search", placeholder: localeText(locale, "experimentList.filterPlaceholder") })), board] }));
 }