niceeval 0.6.0 → 0.6.1

package/dist/agents/types.d.ts

@@ -224,7 +224,11 @@ export interface AgentContext {
      * 具体取值)是两个维度——这里只是「跑的是哪个实验」的稳定标识,不携带条件内容。
      */
     readonly experimentId?: string;
-    /** 仅沙箱型 agent 有(运行器按 --sandbox 备好)。 */
+    /**
+     * 所有 agent 都有:沙箱型是运行器按项目/experiment 配置备好的真实沙箱句柄,remote 型是
+     * `createRemoteSandbox()` 产出的 stub(仅含 `workdir`/`sandboxId`/`otlpHost`/`stop` 等
+     * 元信息,其余方法调用即抛错)。
+     */
     readonly sandbox: Sandbox;
     readonly session: AgentSession;
     /**

package/dist/report/built-ins/experiment-comparison.d.ts

@@ -0,0 +1 @@
+export declare const ExperimentComparison: import("../report.ts").ReportDefinition;

package/dist/report/built-ins/experiment-comparison.js

@@ -0,0 +1,13 @@
+import { jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
+// ExperimentComparison:niceeval view 裸跑时的内置 experiment 比较报告。
+//
+// 名字描述整份报告回答的问题,不绑定其中某一张图。报告由成本 × 通过率散点和固定列
+// ExperimentList 组成；它是一份普通 ReportDefinition,没有 renderer 私有通道。
+import { Col } from "../primitives.js";
+import { ExperimentList, MetricScatter } from "../components.js";
+import { costUSD, passRate } from "../metrics.js";
+import { defineReport } from "../report.js";
+export const ExperimentComparison = defineReport(async ({ selection }) => {
+    const experiments = await ExperimentList.data(selection);
+    return (_jsxs(Col, { children: [_jsx(MetricScatter, { selection: selection, points: "experiment", series: "agent", x: costUSD, y: passRate }), _jsx(ExperimentList, { items: experiments, filter: true })] }));
+});

package/dist/report/built-ins/index.d.ts

@@ -1 +1 @@
-export { CostPassRateComparison } from "./cost-pass-rate-comparison.tsx";
+export { ExperimentComparison } from "./experiment-comparison.tsx";

package/dist/report/built-ins/index.js

@@ -1,2 +1,2 @@
 // 内置报告目录:一个文件一个按展示内容命名的普通 ReportDefinition,显式导出值,不建字符串 registry。
-export { CostPassRateComparison } from "./cost-pass-rate-comparison.js";
+export { ExperimentComparison } from "./experiment-comparison.js";

package/dist/report/components.d.ts

@@ -24,6 +24,8 @@ export interface GroupSummaryProps {
  */
 export interface ExperimentListProps {
     items: ExperimentListItem[];
+    /** web 面在比较表前显示实验过滤框；text 面忽略。 */
+    filter?: boolean;
     /** chrome 文案 locale;省略时随宿主上下文(宿主外默认 "en")。 */
     locale?: ReportLocale;
     className?: string;
@@ -141,7 +143,7 @@ export declare const EvalList: ReportComponent<EvalListProps> & {
 };
 /**
  * Attempt 列表:实体列表的叶子层,每项一个 Attempt,固定展示判定、断言、error、Judge 评语
- * (assertions 的 detail/evidence)与证据引用(locator + 证据能力标记)。它不预设只看失败;
+ * (assertions 的 detail/evidence)与证据引用(locator)。它不预设只看失败;
  * 报告作者过滤 `AttemptListItem[]`、用 `.slice()` 限量,`total` 让渲染面如实报告剩余数量。
  */
 export declare const AttemptList: ReportComponent<AttemptListProps> & {

package/dist/report/components.js

@@ -71,7 +71,7 @@ export const EvalList = Object.assign(defineComponent({
 EvalList.displayName = "EvalList";
 /**
  * Attempt 列表:实体列表的叶子层,每项一个 Attempt,固定展示判定、断言、error、Judge 评语
- * (assertions 的 detail/evidence)与证据引用(locator + 证据能力标记)。它不预设只看失败;
+ * (assertions 的 detail/evidence)与证据引用(locator)。它不预设只看失败;
  * 报告作者过滤 `AttemptListItem[]`、用 `.slice()` 限量,`total` 让渲染面如实报告剩余数量。
  */
 export const AttemptList = Object.assign(defineComponent({

package/dist/report/compute.js

@@ -174,33 +174,6 @@ export async function tableData(input, opts) {
 // 没有列配置;过滤是报告作者对返回数组调用 .filter()/.slice() 的事,不进这里
 // (docs/feature/reports/library.md「实体列表」)。AttemptListItem 是三者共用的叶子形状——
 // ExperimentList / EvalList 的下钻数组直接复用它,不各自精简一份。
-/**
- * 一个 attempt 的证据能力标记,只读已有的瘦身摘要位(`hasSources` / `hasEvents` / `hasTrace`)
- * 加一次 `diff()` 懒加载——不调用完整的 `loadAttemptEvidence`(那还会额外装配 Eval 源码标注
- * 与 ExecutionTree,这里只要四个布尔位)。三位定义与 `AttemptEvidence.capabilities` 完全一致:
- *
- * - `eval`:`hasSources` 为真。真正的门槛是 `AnnotatedEvalSource` 非空且有内容(源码行或断言
- *   非空),但 sources 存在时 `buildAnnotatedEvalSource` 恒产出至少一行源码,断言又永远来自
- *   `result.assertions`(不需要 IO)——`hasSources` 是这个条件的精确瘦身版,不是近似。
- * - `execution`:`hasEvents` 为真,与 `loadAttemptEvidence` 的 `events !== null && events.length > 0`
- *   同一个判定,写入面按同一条规则算的 `hasEvents` 就是它的瘦身镜像。
- * - `timing`:`hasEvents && hasTrace` 为真——`execution` 成立且这次运行接入过 OTel
- *   (`ExecutionTree.timingAvailable` = `spans.length > 0`,`hasTrace` 正是这个判定的瘦身镜像)。
- * - `diff`:唯一需要 IO 的一位,`attempt.diff()` 非空且至少改动/删除了一个文件——落盘没有
- *   与 `hasEvents` 对应的瘦身摘要位,只能懒加载判定;成本是每个 item 一次 `diff()` 读取,
- *   不是完整证据装配的四路 Promise.all。
- */
-async function attemptCapabilities(attempt) {
-    const result = attempt.result;
-    const diff = await attempt.diff();
-    const diffCapable = diff !== null && (Object.keys(diff.generatedFiles).length > 0 || diff.deletedFiles.length > 0);
-    return {
-        eval: result.hasSources === true,
-        execution: result.hasEvents === true,
-        timing: result.hasEvents === true && result.hasTrace === true,
-        diff: diffCapable,
-    };
-}
 /** 自由文本(error / 断言 detail / evidence)的发布消毒钩子;身份字段(name/severity/loc)不经它。 */
 function redactAssertions(assertions, redact) {
     if (assertions.length === 0)
@@ -212,7 +185,7 @@ function redactAssertions(assertions, redact) {
     }));
 }
 /** AttemptList / ExperimentList / EvalList 共用的叶子构造:一个 Item → 一个 AttemptListItem。 */
-async function attemptListItemOf(item, redact) {
+function attemptListItemOf(item, redact) {
     const result = item.attempt.result;
     const cost = attemptCostUSD(result);
     return {
@@ -226,7 +199,6 @@ async function attemptListItemOf(item, redact) {
         durationMs: result.durationMs,
         ...(cost !== null ? { costUSD: cost } : {}),
         locator: locatorOf(item),
-        capabilities: await attemptCapabilities(item.attempt),
     };
 }
 const identityRedact = (text) => text;
@@ -235,7 +207,7 @@ export async function attemptListData(input, opts) {
     const { snapshots } = resolveInput(input);
     const redact = opts?.redact ?? identityRedact;
     const items = collectItems(snapshots);
-    return Promise.all(items.map((item) => attemptListItemOf(item, redact)));
+    return items.map((item) => attemptListItemOf(item, redact));
 }
 /** `EvalList.data(selection)`:每个 `experimentId + evalId` 一项,按 evalId 再按 experimentId 升序。 */
 export async function evalListData(input) {
@@ -255,7 +227,7 @@ export async function evalListData(input) {
         const sorted = [...group].sort((a, b) => a.attempt.result.attempt - b.attempt.result.attempt);
         const verdict = foldEvalVerdict(sorted.map((item) => item.attempt.result));
         const representative = sorted.find((item) => item.attempt.result.verdict === verdict) ?? sorted[0];
-        const attempts = await Promise.all(sorted.map((item) => attemptListItemOf(item, identityRedact)));
+        const attempts = sorted.map((item) => attemptListItemOf(item, identityRedact));
         out.push({
             evalId: evalIdOf(sorted[0]),
             experimentId: experimentIdOf(sorted[0]),
@@ -285,7 +257,7 @@ export async function experimentListData(input) {
             const sorted = [...evalItems].sort((a, b) => a.attempt.result.attempt - b.attempt.result.attempt);
             const verdict = foldEvalVerdict(sorted.map((item) => item.attempt.result));
             const representative = sorted.find((item) => item.attempt.result.verdict === verdict) ?? sorted[0];
-            const attempts = await Promise.all(sorted.map((item) => attemptListItemOf(item, identityRedact)));
+            const attempts = sorted.map((item) => attemptListItemOf(item, identityRedact));
             evalRows.push({
                 evalId,
                 verdict,
@@ -315,7 +287,17 @@ export async function experimentListData(input) {
             evalRows,
         });
     }
-    out.sort((a, b) => a.experimentId.localeCompare(b.experimentId));
+    // ExperimentList 是默认实验比较表:初始态按成功率从高到低,缺数据沉底;
+    // 同分时按 experiment id 稳定排序。web 增强可临时重排,text 面沿用同一基准顺序。
+    out.sort((a, b) => {
+        if (a.passRate.value === null && b.passRate.value === null)
+            return a.experimentId.localeCompare(b.experimentId);
+        if (a.passRate.value === null)
+            return 1;
+        if (b.passRate.value === null)
+            return -1;
+        return b.passRate.value - a.passRate.value || a.experimentId.localeCompare(b.experimentId);
+    });
     return out;
 }
 export async function matrixData(input, opts) {

package/dist/report/format.d.ts

@@ -1,4 +1,3 @@
-import type { AttemptEvidenceCapabilities } from "../results/attempt-evidence.ts";
 import type { AssertionResult, Verdict } from "../types.ts";
 export declare function formatMetricValue(value: number, unit?: string): string;
 /** 无单位纯数字(scoreboard 总分等):一位小数,去尾零。 */
@@ -13,11 +12,6 @@ export declare function formatUSD(usd: number): string;
 export declare function formatPercent(ratio: number): string;
 /** passed / failed / errored / skipped 的判定符。 */
 export declare function verdictMark(verdict: Verdict): string;
-/**
- * 证据能力标记:`[E,X,⏱]` 式方括号列表,固定顺序 eval → execution → timing → diff,
- * 缺哪个不列哪个;四位全无时返回空串(没有证据可看,不画空 `[]`)。
- */
-export declare function capabilityBadge(capabilities: AttemptEvidenceCapabilities): string;
 /**
  * 一个 AttemptListItem 的失败原因摘要,按优先级取第一个在场的:`error` → 未通过的 gate 断言
  * (原始声明顺序,`name`,detail 在场则 `"name: detail"`,多条用「, 」连接)→ 缺席。与

package/dist/report/format.js

@@ -77,8 +77,7 @@ export function formatUSD(usd) {
 export function formatPercent(ratio) {
     return `${Math.round(ratio * 100)}%`;
 }
-// ── 实体列表(ExperimentList / EvalList / AttemptList)共用的判定符与证据能力位 ──
-// web / text 两面对同一个 AttemptListItem 给出同一个标记(docs-site「终端输出形成反馈闭环」)。
+// ── 实体列表(ExperimentList / EvalList / AttemptList)共用的判定符 ──
 /** passed / failed / errored / skipped 的判定符。 */
 export function verdictMark(verdict) {
     switch (verdict) {
@@ -92,22 +91,6 @@ export function verdictMark(verdict) {
             return "–";
     }
 }
-/**
- * 证据能力标记:`[E,X,⏱]` 式方括号列表,固定顺序 eval → execution → timing → diff,
- * 缺哪个不列哪个;四位全无时返回空串(没有证据可看,不画空 `[]`)。
- */
-export function capabilityBadge(capabilities) {
-    const marks = [];
-    if (capabilities.eval)
-        marks.push("E");
-    if (capabilities.execution)
-        marks.push("X");
-    if (capabilities.timing)
-        marks.push("⏱");
-    if (capabilities.diff)
-        marks.push("D");
-    return marks.length > 0 ? `[${marks.join(",")}]` : "";
-}
 /**
  * 一个 AttemptListItem 的失败原因摘要,按优先级取第一个在场的:`error` → 未通过的 gate 断言
  * (原始声明顺序,`name`,detail 在场则 `"name: detail"`,多条用「, 」连接)→ 缺席。与

package/dist/report/index.d.ts

@@ -8,12 +8,12 @@ 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 { 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, 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

@@ -20,7 +20,7 @@ export { Row, Col, Section, Text, Style, Table } from "./primitives.js";
 // 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";
+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

@@ -22,6 +22,19 @@ 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": "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";
     /** <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)";
@@ -46,8 +59,6 @@ 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";

package/dist/report/locale.js

@@ -25,6 +25,19 @@ 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": "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",
     /** <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)",
@@ -49,8 +62,6 @@ 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",
@@ -90,6 +101,19 @@ 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": "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",
     "table.columnsHidden.one": "(还有 {n} 列未列出)",
     "table.columnsHidden.other": "(还有 {n} 列未列出)",
     "overview.snapshots": "快照",
@@ -111,7 +135,6 @@ 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",

package/dist/report/metrics.js

@@ -13,7 +13,7 @@
 //
 // 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;

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

@@ -2,9 +2,9 @@ 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。 */

package/dist/report/react/AttemptList.js

@@ -1,11 +1,10 @@
 import { jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
 import { DEFAULT_REPORT_LOCALE, localeText } from "../locale.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 })] }))] }));

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

@@ -1,10 +1,11 @@
 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, }: {
     items: ExperimentListItem[];
     attemptHref: (locator: AttemptLocator) => string;
+    filter?: boolean;
     className?: string;
     locale?: ReportLocale;
 }): ReactElement;

package/dist/report/react/ExperimentList.js

@@ -1,15 +1,67 @@
 import { Fragment as _Fragment, jsx as _jsx, jsxs as _jsxs } from "react/jsx-runtime";
+import { attemptItemReason } 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-${attempt.verdict}`), children: [_jsx("span", { className: "nre-attempt-branch", "aria-hidden": "true", children: last ? "└─" : "├─" }), _jsx("span", { className: "nre-eval-attempt-badges", children: _jsx(AttemptLocatorBadge, { item: attempt, attemptHref: attemptHref }) }), _jsxs("span", { className: "nre-eval-attempt-metrics", children: [formatDurationMs(attempt.durationMs), attempt.costUSD !== undefined && _jsxs(_Fragment, { children: [" \u00B7 ", formatUSD(attempt.costUSD)] })] }), _jsx("span", { className: "nre-eval-reason", children: reason ?? "—" })] }));
+}
+function EvalAttempts({ row, attemptHref, locale, }) {
+    return (_jsxs("li", { className: "nre-experiment-eval", children: [_jsxs("div", { className: cx("nre-experiment-eval-header", `nre-eval-${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 }) }), _jsx("span", { className: "nre-eval-summary", children: row.verdict === "passed"
+                            ? [formatDurationMs(row.duration.value ?? 0), row.cost.value === null ? undefined : formatUSD(row.cost.value)]
+                                .filter((value) => value !== undefined)
+                                .join(" · ")
+                            : row.reason })] }), _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, }) {
+    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: item.experimentId }), _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, }) {
+    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 }, 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] }));
 }

package/dist/report/react/MetricScatter.js

@@ -40,10 +40,6 @@ export function MetricScatter({ data, pointHref, className, locale = DEFAULT_REP
     if (drawableRows.length === 0) {
         return (_jsxs("figure", { className: cx("nre", "nre-metric-scatter", className), children: [_jsx("p", { className: "nre-scatter-empty nre-missing", children: localeText(locale, "scatter.noData", { x: xLabel, y: yLabel }) }), missingNote] }));
     }
-    // 恰好 1 个可画点:成本 × 通过率的对比至少要两个实验,单点不成图 —— 显式说清而不是画个孤点。
-    if (drawableRows.length === 1) {
-        return (_jsxs("figure", { className: cx("nre", "nre-metric-scatter", className), children: [_jsx("p", { className: "nre-scatter-empty", children: localeText(locale, "scatter.needTwo", { x: xLabel, y: yLabel }) }), missingNote] }));
-    }
     const axisLabel = (label, col) => `${label}${col.unit ? `(${col.unit})` : ""}`;
     const xScale = axisScale(drawableRows.map((r) => r.x.value), data.x.better, MARGIN.left, MARGIN.left + PLOT_W);
     // y 像素轴向下增长:better:"higher" 高值在上 → 映射到 [bottom, top];"lower" 由 axisScale 反向后同样落到上方
@@ -57,7 +53,7 @@ export function MetricScatter({ data, pointHref, className, locale = DEFAULT_REP
             label: pointLabel(r.key),
             xValue,
             yValue,
-            // hover 内容:experiment(点键)+ 系列(series,CostPassRateComparison 传的是 agent 维度,
+            // hover 内容:experiment(点键)+ 系列(series,ExperimentComparison 传的是 agent 维度,
             // 有则加一行;无系列的散点没有这行)+ 两轴 display 与 samples/total(docs/feature/reports/library.md「MetricScatter」行为清单)
             title: `${r.key}${r.series !== undefined ? `\n${r.series}` : ""}\n${xLabel}: ${r.x.display}(${r.x.samples}/${r.x.total})\n${yLabel}: ${r.y.display}(${r.y.samples}/${r.y.total})`,
             px: xScale.scale(xValue),

package/dist/report/react/fixtures.js

@@ -300,7 +300,6 @@ const failedAttempt = {
     durationMs: 32_000,
     costUSD: 0.12,
     locator: locator("@1a4a4a4"),
-    capabilities: { eval: true, execution: true, timing: true, diff: false },
 };
 const erroredAttempt = {
     evalId: "geometry/angles",
@@ -312,7 +311,12 @@ const erroredAttempt = {
     assertions: [],
     durationMs: 4_500,
     locator: locator("@1c1c1c1"),
-    capabilities: { eval: false, execution: true, timing: false, diff: false },
+};
+const failedRetryAttempt = {
+    ...failedAttempt,
+    attempt: 4,
+    durationMs: 35_000,
+    locator: locator("@1b5b5b5"),
 };
 export const attemptListItems = [failedAttempt, erroredAttempt];
 export const evalListItems = [
@@ -342,13 +346,14 @@ export const experimentListItems = [
         experimentId: "compare/bub",
         agent: "bub",
         model: "gpt-5.4",
+        flags: { memory: true },
         verdicts: { passed: 1, failed: 1, errored: 0, skipped: 0 },
         passRate: { value: 0.5, display: "50%", samples: 2, total: 2, refs: [] },
         cost: { value: 0.12, display: "$0.12", samples: 1, total: 2, refs: [failedAttempt.locator] },
         duration: { value: 32_000, display: "32.0s", samples: 2, total: 2, refs: [] },
         tokens: { value: null, display: "—", samples: 0, total: 2, refs: [] },
         evals: 2,
-        attempts: 2,
+        attempts: 3,
         lastRunAt: "2026-07-01T10:00:00Z",
         evalRows: [
             {
@@ -357,7 +362,7 @@ export const experimentListItems = [
                 reason: "roots-correct: expected x=2, got x=3",
                 duration: { value: 32_000, display: "32.0s", samples: 1, total: 1, refs: [failedAttempt.locator] },
                 cost: { value: 0.12, display: "$0.12", samples: 1, total: 1, refs: [failedAttempt.locator] },
-                attempts: [failedAttempt],
+                attempts: [failedAttempt, failedRetryAttempt],
             },
             {
                 evalId: "algebra/simple",

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

@@ -1,3 +1,3 @@
-export { MISSING_TEXT, capabilityBadge, formatDurationMs, formatPercent, formatUSD, verdictMark } from "../format.ts";
+export { MISSING_TEXT, formatDurationMs, formatPercent, formatUSD, verdictMark } from "../format.ts";
 /** 拼 class 名:过滤空值,末尾接使用者透传的 className。 */
 export declare function cx(...parts: (string | undefined | false)[]): string;

package/dist/report/react/format.js

@@ -1,6 +1,6 @@
 // web 面的小工具:展示格式化统一住在计算侧的 ../format.ts(两个渲染面同一份),
 // 这里 re-export 并补 class 名拼接。MetricCell 一律自带 display,组件不重算。
-export { MISSING_TEXT, capabilityBadge, formatDurationMs, formatPercent, formatUSD, verdictMark } from "../format.js";
+export { MISSING_TEXT, formatDurationMs, formatPercent, formatUSD, verdictMark } from "../format.js";
 /** 拼 class 名:过滤空值,末尾接使用者透传的 className。 */
 export function cx(...parts) {
     return parts.filter(Boolean).join(" ");

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

@@ -1,6 +1,6 @@
 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 { AttemptEvidenceCapabilities, AttemptListItem, AttemptLocator, DeltaData, EvalListItem, ExperimentListEvalRow, ExperimentListItem, GroupSummaryData, LineAxis, LineData, MatrixData, MetricCell, MetricColumn, OverviewData, ScatterData, ScoreboardData, SelectionWarning, TableData, TableRowMeta, } from "../types.ts";
+export type { AttemptListItem, AttemptLocator, DeltaData, EvalListItem, ExperimentListEvalRow, ExperimentListItem, GroupSummaryData, LineAxis, LineData, MatrixData, MetricCell, MetricColumn, OverviewData, ScatterData, ScoreboardData, SelectionWarning, TableData, TableRowMeta, } from "../types.ts";
 export { DEFAULT_REPORT_LOCALE, resolveMetricLabel } from "../locale.ts";
 export type { LocalizedLabel, ReportLocale } from "../locale.ts";
 export { NRE_PALETTE, colorClassForKey, colorHexForKey, colorIndexForKey, seriesClassForKey } from "./colors.ts";