niceeval 0.8.0 → 0.9.0

package/dist/report/components.js

@@ -1,13 +1,18 @@
 import { jsx as _jsx } from "react/jsx-runtime";
 import { defineComponent, isHostWebContextActive, memoFetchOf, } from "./tree.js";
-import { attemptListData, deltaTableData, evalListData, experimentComparisonData, experimentListData, metricLineData, metricMatrixData, metricScatterData, metricTableData, scopeSummaryData, scoreboardData, } from "./compute.js";
+import { attemptListData, copyFixPromptData, deltaTableData, evalListData, experimentComparisonData, experimentListData, heroData, metricLineData, metricMatrixData, metricScatterData, metricTableData, scopeSummaryData, scopeWarningsData, scoreboardData, traceWaterfallData, } from "./compute.js";
 import { collectItems, locatorOf, resolveInput } from "./aggregate.js";
-import { attemptListText, deltaText, evalListText, experimentComparisonText, experimentListText, barsText, lineText, matrixText, scatterText, scoreboardText, scopeSummaryText, tableText, } from "./text/faces.js";
+import { attemptListText, deltaText, evalListText, experimentComparisonText, experimentListText, barsText, heroCardText, lineText, matrixText, scatterText, scopeWarningsText, scoreboardText, scopeSummaryText, tableText, traceWaterfallText, } from "./text/faces.js";
 import { ScopeSummary as ScopeSummaryWeb } from "./react/ScopeSummary.js";
 import { ExperimentComparisonView } from "./react/ExperimentComparison.js";
 import { ExperimentList as ExperimentListWeb } from "./react/ExperimentList.js";
 import { EvalList as EvalListWeb } from "./react/EvalList.js";
 import { AttemptList as AttemptListWeb } from "./react/AttemptList.js";
+import { HeroCard as HeroCardWeb } from "./react/HeroCard.js";
+import { PoweredBy as PoweredByWeb } from "./react/PoweredBy.js";
+import { ScopeWarnings as ScopeWarningsWeb } from "./react/ScopeWarnings.js";
+import { CopyFixPrompt as CopyFixPromptWeb } from "./react/CopyFixPrompt.js";
+import { TraceWaterfall as TraceWaterfallWeb } from "./react/TraceWaterfall.js";
 import { MetricTable as MetricTableWeb } from "./react/MetricTable.js";
 import { MetricMatrix as MetricMatrixWeb } from "./react/MetricMatrix.js";
 import { MetricBars as MetricBarsWeb } from "./react/MetricBars.js";
@@ -183,6 +188,56 @@ const validateAttemptListData = (data) => {
     }
     return null;
 };
+const validateHeroData = (data) => {
+    if (!isObject(data))
+        return "expected an object";
+    if (!("latestStartedAt" in data) || (data.latestStartedAt !== null && typeof data.latestStartedAt !== "string")) {
+        return 'missing "latestStartedAt" (string | null)';
+    }
+    if (typeof data.snapshots !== "number")
+        return 'missing "snapshots" (number)';
+    return null;
+};
+const validateScopeWarningsData = (data) => {
+    if (!Array.isArray(data))
+        return "expected an array of ScopeWarning";
+    for (const item of data) {
+        if (!isObject(item) || typeof item.kind !== "string" || typeof item.message !== "string") {
+            return "each warning needs { kind, message, … }";
+        }
+    }
+    return null;
+};
+const validateCopyFixPromptData = (data) => {
+    if (!isObject(data))
+        return "expected an object";
+    if (typeof data.prompt !== "string")
+        return 'missing "prompt" (string)';
+    if (typeof data.failures !== "number")
+        return 'missing "failures" (number)';
+    return null;
+};
+const validateTraceWaterfallData = (data) => {
+    if (!Array.isArray(data))
+        return "expected an array of TraceWaterfallRow";
+    for (const row of data) {
+        if (!isObject(row) ||
+            typeof row.experimentId !== "string" ||
+            typeof row.evalId !== "string" ||
+            typeof row.locator !== "string" ||
+            !("durationMs" in row) ||
+            (row.durationMs !== null && typeof row.durationMs !== "number") ||
+            !Array.isArray(row.spans)) {
+            return "each row needs { experimentId, evalId, locator, durationMs: number | null, spans }";
+        }
+        for (const span of row.spans) {
+            if (!isObject(span) || typeof span.name !== "string" || typeof span.startOffsetMs !== "number") {
+                return "each span needs { name, kind, startOffsetMs, durationMs, failed }";
+            }
+        }
+    }
+    return null;
+};
 function makeDataComponent(def) {
     const assertData = (data) => {
         const problem = def.validate(data);
@@ -261,8 +316,8 @@ export const ExperimentList = makeDataComponent({
     name: "ExperimentList",
     dataFnName: "experimentListData",
     shapeName: "ExperimentListItem[]",
-    dataFn: (input, options) => experimentListData(input, options),
-    specKeys: ["redact"],
+    dataFn: (input) => experimentListData(input),
+    specKeys: [],
     validate: validateExperimentListData,
     web: (props, ctx) => (_jsx(ExperimentListWeb, { data: props.data, filter: props.filter, relativeTo: props.relativeTo, locale: props.locale ?? ctx.locale, attemptHref: hrefOf(props, ctx) ?? ctx.attemptHref, className: props.className })),
     text: (props, ctx) => experimentListText(props.data, ctx, props.relativeTo),
@@ -272,8 +327,8 @@ export const EvalList = makeDataComponent({
     name: "EvalList",
     dataFnName: "evalListData",
     shapeName: "EvalListItem[]",
-    dataFn: (input, options) => evalListData(input, options),
-    specKeys: ["redact"],
+    dataFn: (input) => evalListData(input),
+    specKeys: [],
     validate: validateEvalListData,
     web: (props, ctx) => (_jsx(EvalListWeb, { data: props.data, locale: props.locale ?? ctx.locale, attemptHref: hrefOf(props, ctx) ?? ctx.attemptHref, className: props.className })),
     text: (props, ctx) => evalListText(props.data, ctx),
@@ -283,10 +338,10 @@ export const AttemptList = makeDataComponent({
     name: "AttemptList",
     dataFnName: "attemptListData",
     shapeName: "AttemptListItem[]",
-    dataFn: (input, options) => attemptListData(input, options),
-    specKeys: ["redact"],
+    dataFn: (input) => attemptListData(input),
+    specKeys: [],
     validate: validateAttemptListData,
-    web: (props, ctx) => (_jsx(AttemptListWeb, { data: props.data, total: props.total, locale: props.locale ?? ctx.locale, attemptHref: hrefOf(props, ctx) ?? ctx.attemptHref, className: props.className })),
+    web: (props, ctx) => (_jsx(AttemptListWeb, { data: props.data, total: props.total, filter: props.filter, locale: props.locale ?? ctx.locale, attemptHref: hrefOf(props, ctx) ?? ctx.attemptHref, className: props.className })),
     text: (props, ctx) => attemptListText(props.data, props.total, ctx),
 });
 /**
@@ -297,7 +352,7 @@ export const AttemptList = makeDataComponent({
  */
 export const FailureList = defineComponent(async (props, ctx) => {
     const input = props.input ?? ctx.scope;
-    const all = await attemptListData(input, props.redact !== undefined ? { redact: props.redact } : undefined);
+    const all = await attemptListData(input);
     // attempt 开始时间不在列表条目里(它不是列表展示字段);从同一 input 的读取面按 locator 对回。
     const startedAtByLocator = new Map();
     for (const item of collectItems(resolveInput(input).snapshots)) {
@@ -316,6 +371,100 @@ export const FailureList = defineComponent(async (props, ctx) => {
     return (_jsx(AttemptList, { data: failures.slice(0, limit), total: failures.length, attemptHref: props.attemptHref, locale: props.locale, className: props.className }));
 });
 FailureList.displayName = "FailureList";
+/** HeroCard 的 data 校验入口(它不经 makeDataComponent,数据形态是唯一形态)。 */
+const assertHeroData = (data) => {
+    const problem = validateHeroData(data);
+    if (problem !== null)
+        throw dataShapeError("HeroCard", "heroData", "HeroData", problem);
+    return data;
+};
+/**
+ * `HeroCard`:Hero 的渲染件,双面组件,只收 data 形态——标题输入是站点声明与 Scope 的
+ * 合成物,没有单独的 spec 等价形。web 面渲染 hero 标题(h1)、按渲染 locale 格式化的运行
+ * meta(latestStartedAt 为 null 时内置「暂无运行」文案)与品牌行(等同 PoweredBy,恒含、
+ * 无拆除 prop);text 面输出标题行与 meta 行,不含品牌行
+ * (docs/feature/reports/library/site-components.md「HeroCard」)。
+ */
+export const HeroCard = defineComponent({
+    web: (props, ctx) => {
+        assertHeroData(props.data);
+        return _jsx(HeroCardWeb, { title: props.title, data: props.data, className: props.className, locale: ctx.locale });
+    },
+    text: (props, ctx) => {
+        assertHeroData(props.data);
+        return heroCardText(props.title, props.data, ctx);
+    },
+});
+HeroCard.displayName = "HeroCard";
+/**
+ * `Hero`:页首的站点标题区——标题、最后运行时间、快照合成来源,恒含品牌行。官方组合组件,
+ * 与手写 `<HeroCard title={title ?? ctx.report.title} data={await heroData(ctx.scope)} />`
+ * 严格等价、没有私有能力;读 `ctx.report` 意味着输出跟随站点,要站点无关的标题区直接用
+ * `HeroCard` 显式传值(docs/feature/reports/library/site-components.md「Hero」)。
+ */
+export const Hero = defineComponent(async ({ title, className }, ctx) => (_jsx(HeroCard, { title: title ?? ctx.report.title, data: await heroData(ctx.scope), className: className })));
+Hero.displayName = "Hero";
+/**
+ * `PoweredBy`:唯一的品牌件,无 props 双面组件。web 面渲染指向 niceeval 官网的一行品牌色
+ * 小字(`utm_source=report&utm_medium=powered-by`,`rel` 仅 `noopener` 以保留 Referer);
+ * text 面零输出。没有任何配置——品牌契约是「提供一个组件,不给开关」:不想要品牌就不用
+ * 这些组件、自己写替代组件(docs/feature/reports/library/site-components.md「PoweredBy」)。
+ */
+export const PoweredBy = defineComponent({
+    web: () => _jsx(PoweredByWeb, {}),
+    text: () => "",
+});
+PoweredBy.displayName = "PoweredBy";
+/**
+ * `ScopeWarnings`:选择警告区,警告的唯一呈现组件。把 Scope 携带的 `ScopeWarning[]`
+ * 按「下一步动作」聚合渲染(带 experimentId 的按实验聚合、非实验作用域按 kind 聚合;
+ * integrity 组在前);web 面组头带去重后的可复制命令、明细收原生 `<details>`(总条数 ≤ 3
+ * 默认展开),text 面同构但不折叠。空警告集与裸 `Snapshot[]` 输入两面零输出
+ * (docs/feature/reports/library/site-components.md「ScopeWarnings」)。
+ */
+export const ScopeWarnings = makeDataComponent({
+    name: "ScopeWarnings",
+    dataFnName: "scopeWarningsData",
+    shapeName: "ScopeWarning[]",
+    dataFn: (input) => scopeWarningsData(input),
+    specKeys: [],
+    validate: validateScopeWarningsData,
+    web: (props, ctx) => props.data.length === 0 ? null : (_jsx(ScopeWarningsWeb, { data: props.data, locale: props.locale ?? ctx.locale, className: props.className })),
+    text: (props, ctx) => scopeWarningsText(props.data, ctx),
+});
+/**
+ * `CopyFixPrompt`:把当前范围的全部失败整理成一段可交给 coding agent 的修复 prompt。
+ * prompt 在 resolve 阶段算好、烘进静态 HTML,无 JS 时在折叠块里完整可读,「复制」是增强层
+ * 行为;`failures` 为 0 时两面零输出;text 面恒零输出——终端里的等价能力是 `show` 的
+ * attempt 下钻命令本身(docs/feature/reports/library/site-components.md「CopyFixPrompt」)。
+ */
+export const CopyFixPrompt = makeDataComponent({
+    name: "CopyFixPrompt",
+    dataFnName: "copyFixPromptData",
+    shapeName: "CopyFixPromptData",
+    dataFn: (input) => copyFixPromptData(input),
+    specKeys: [],
+    validate: validateCopyFixPromptData,
+    web: (props, ctx) => props.data.failures === 0 ? null : (_jsx(CopyFixPromptWeb, { data: props.data, locale: props.locale ?? ctx.locale, className: props.className })),
+    text: () => "",
+});
+/**
+ * `TraceWaterfall`:每个 attempt 一行的执行时间瀑布,用 canonical OTel 字段显示被测 agent
+ * 的原始 span(agent / model / tool)。web 面静态渲染顶层 span 分解条(失败 span 带失败
+ * 标记),行链接 attempt 详情;text 面每 attempt 一行(locator、总耗时、span 计数与失败
+ * 标记)+ 可复制的 `--timing` 下钻命令。trace 缺失的行照常出现并如实显示缺失;runner
+ * 生命周期节点不进瀑布(docs/feature/reports/library/site-components.md「TraceWaterfall」)。
+ */
+export const TraceWaterfall = makeDataComponent({
+    name: "TraceWaterfall",
+    dataFnName: "traceWaterfallData",
+    shapeName: "TraceWaterfallRow[]",
+    dataFn: (input) => traceWaterfallData(input),
+    specKeys: [],
+    validate: validateTraceWaterfallData,
+    web: (props, ctx) => (_jsx(TraceWaterfallWeb, { data: props.data, attemptHref: hrefOf(props, ctx) ?? ctx.attemptHref, locale: props.locale ?? ctx.locale, className: props.className })),
+    text: (props, ctx) => traceWaterfallText(props.data, ctx),
+});
 /** 榜单:一行一个维度值、一列一个指标,回答「谁整体更好」。 */
 export const MetricTable = makeDataComponent({
     name: "MetricTable",

package/dist/report/compute.d.ts

@@ -1,4 +1,4 @@
-import type { AttemptListItem, DeltaData, DeltaPair, DimensionInput, EntityListDataOptions, EvalListItem, ExperimentComparisonData, ExperimentListItem, FlagPairs, LineData, MatrixData, Metric, NumericAxis, ReportInput, ScatterData, ScopeSummaryData, ScoreboardData, TableData } from "./types.ts";
+import type { AttemptListItem, CopyFixPromptData, DeltaData, DeltaPair, DimensionInput, EvalListItem, ExperimentComparisonData, ExperimentListItem, FlagPairs, HeroData, LineData, MatrixData, Metric, NumericAxis, ReportInput, ScatterData, ScopeSummaryData, ScopeWarning, ScoreboardData, TableData, TraceWaterfallRow } from "./types.ts";
 import type { JsonValue } from "../types.ts";
 export interface MetricTableOptions {
     /** 行维度(内置 / 自定义 / flag() / runConfig())。 */
@@ -23,9 +23,9 @@ export interface MetricMatrixOptions {
 }
 export declare function metricMatrixData(input: ReportInput, options: MetricMatrixOptions): Promise<MatrixData>;
 /** `attemptListData(input)`:每个 Attempt 一项,顺序取自 Scope 展平顺序(不重排)。 */
-export declare function attemptListData(input: ReportInput, options?: EntityListDataOptions): Promise<AttemptListItem[]>;
+export declare function attemptListData(input: ReportInput): Promise<AttemptListItem[]>;
 /** `evalListData(input)`:每个 `experimentId + evalId` 一项,按 evalId 再按 experimentId 升序。 */
-export declare function evalListData(input: ReportInput, options?: EntityListDataOptions): Promise<EvalListItem[]>;
+export declare function evalListData(input: ReportInput): Promise<EvalListItem[]>;
 /**
  * `experimentListData(input)`:每个 experiment 一项,展开到每道 Eval;初始按端到端成功率
  * 从高到低(缺数据沉底,同分按 id)。一行只有一套 agent / model / flags 是输入约束:
@@ -33,7 +33,7 @@ export declare function evalListData(input: ReportInput, options?: EntityListDat
  * Snapshot[] 时若同一 experiment 混入不一致的可比性配置,按完整用户反馈失败并指引——
  * 看跨配置演化用 snapshot 维度或 MetricLine,不把两套配置拼成一行冒充单一配置。
  */
-export declare function experimentListData(input: ReportInput, options?: EntityListDataOptions): Promise<ExperimentListItem[]>;
+export declare function experimentListData(input: ReportInput): Promise<ExperimentListItem[]>;
 /**
  * `scopeSummaryData(input)`:范围摘要——快照时间窗、experiment / eval / attempt 数、
  * 两级判定计票、端到端成功率与总成本(docs/feature/reports/library/summaries.md)。
@@ -110,3 +110,30 @@ export interface DeltaTableOptions {
     evals?: string | readonly string[];
 }
 export declare function deltaTableData(input: ReportInput, options: DeltaTableOptions): Promise<DeltaData>;
+/**
+ * `heroData(input)`:站点标题区的运行 meta——`latestStartedAt` 取范围内最新快照的开始时间
+ * (空范围为 null,不编造当前时间),`snapshots` 计贡献当前水位的快照数
+ * (docs/feature/reports/library/site-components.md「HeroCard」)。
+ */
+export declare function heroData(input: ReportInput): Promise<HeroData>;
+/**
+ * `scopeWarningsData(input)`:Scope 携带的挑选警告原样透出;`input` 是裸 `Snapshot[]` 时
+ * 没有挑选过程、没有警告,返回空数组,也如实
+ * (docs/feature/reports/library/site-components.md「ScopeWarnings」)。
+ */
+export declare function scopeWarningsData(input: ReportInput): Promise<readonly ScopeWarning[]>;
+/**
+ * `copyFixPromptData(input)`:把范围内全部失败(verdict 为 failed / errored 的 attempt)
+ * 整理成一段可交给 coding agent 的修复 prompt——逐失败含 eval id、主失败摘要与 attempt
+ * 下钻命令(`niceeval show @<locator>`)。prompt 面向 agent,固定英文
+ * (docs/feature/reports/library/site-components.md「CopyFixPrompt」)。
+ */
+export declare function copyFixPromptData(input: ReportInput): Promise<CopyFixPromptData>;
+/**
+ * `traceWaterfallData(input)`:每个 attempt 一行的执行时间瀑布摘要。span 事实只来自
+ * trace artifact(经 AttemptHandle 懒加载的 canonical OTel span);runner 生命周期节点
+ * (`result.phases`)不进瀑布。行内只汇总顶层 span(parentSpanId 缺失或不在本 trace 内),
+ * 按 startOffsetMs 升序;trace 缺失或为空时 `durationMs` 为 null、行照常出现
+ * (docs/feature/reports/library/site-components.md「TraceWaterfall」)。
+ */
+export declare function traceWaterfallData(input: ReportInput): Promise<readonly TraceWaterfallRow[]>;

package/dist/report/compute.js

@@ -117,9 +117,8 @@ function failureSummaryOf(result) {
     }
     return { summary: null, more: 0 };
 }
-const identityRedact = (text) => text;
 /** AttemptList / ExperimentList / EvalList 共用的叶子构造:一个 Item → 一个 AttemptListItem。 */
-async function attemptListItemOf(item, redact) {
+async function attemptListItemOf(item) {
     const result = item.attempt.result;
     const { summary, more } = failureSummaryOf(result);
     return {
@@ -128,7 +127,7 @@ async function attemptListItemOf(item, redact) {
         attempt: result.attempt,
         agent: result.agent,
         verdict: result.verdict,
-        failureSummary: summary === null ? null : redact(summary),
+        failureSummary: summary,
         moreFailures: more,
         examScore: await computeCell(examScore, [item]),
         durationMs: result.durationMs,
@@ -137,16 +136,14 @@ async function attemptListItemOf(item, redact) {
     };
 }
 /** `attemptListData(input)`:每个 Attempt 一项,顺序取自 Scope 展平顺序(不重排)。 */
-export async function attemptListData(input, options) {
+export async function attemptListData(input) {
     const { snapshots } = resolveInput(input);
-    const redact = options?.redact ?? identityRedact;
     const items = collectItems(snapshots);
-    return Promise.all(items.map((item) => attemptListItemOf(item, redact)));
+    return Promise.all(items.map((item) => attemptListItemOf(item)));
 }
 /** `evalListData(input)`:每个 `experimentId + evalId` 一项,按 evalId 再按 experimentId 升序。 */
-export async function evalListData(input, options) {
+export async function evalListData(input) {
     const { snapshots } = resolveInput(input);
-    const redact = options?.redact ?? identityRedact;
     const items = collectItems(snapshots);
     const groups = new Map();
     for (const item of items) {
@@ -161,7 +158,7 @@ export async function evalListData(input, options) {
     for (const group of groups.values()) {
         const sorted = [...group].sort((a, b) => a.attempt.result.attempt - b.attempt.result.attempt);
         const verdict = foldEvalVerdict(sorted.map((item) => item.attempt.result));
-        const attempts = await Promise.all(sorted.map((item) => attemptListItemOf(item, redact)));
+        const attempts = await Promise.all(sorted.map((item) => attemptListItemOf(item)));
         out.push({
             experimentId: experimentIdOf(sorted[0]),
             evalId: evalIdOf(sorted[0]),
@@ -182,9 +179,8 @@ export async function evalListData(input, options) {
  * Snapshot[] 时若同一 experiment 混入不一致的可比性配置,按完整用户反馈失败并指引——
  * 看跨配置演化用 snapshot 维度或 MetricLine,不把两套配置拼成一行冒充单一配置。
  */
-export async function experimentListData(input, options) {
+export async function experimentListData(input) {
     const { snapshots } = resolveInput(input);
-    const redact = options?.redact ?? identityRedact;
     // 可比性配置单义检查:同一 experiment 的输入快照必须共享一套可比性配置。
     const configByExperiment = new Map();
     for (const snapshot of snapshots) {
@@ -211,7 +207,7 @@ export async function experimentListData(input, options) {
         for (const [evalId, evalItems] of evalGroups) {
             const sorted = [...evalItems].sort((a, b) => a.attempt.result.attempt - b.attempt.result.attempt);
             const verdict = foldEvalVerdict(sorted.map((item) => item.attempt.result));
-            const attempts = await Promise.all(sorted.map((item) => attemptListItemOf(item, redact)));
+            const attempts = await Promise.all(sorted.map((item) => attemptListItemOf(item)));
             evalRows.push({
                 evalId,
                 verdict,
@@ -778,3 +774,126 @@ function deltaOutcome(metric, delta) {
     const better = metric.better ?? "higher";
     return (delta > 0) === (better === "higher") ? "improved" : "regressed";
 }
+// ───────────────────────── 站点组件的计算函数(hero / warnings / fix prompt / trace)─────────────────────────
+/**
+ * `heroData(input)`:站点标题区的运行 meta——`latestStartedAt` 取范围内最新快照的开始时间
+ * (空范围为 null,不编造当前时间),`snapshots` 计贡献当前水位的快照数
+ * (docs/feature/reports/library/site-components.md「HeroCard」)。
+ */
+export async function heroData(input) {
+    const { snapshots } = resolveInput(input);
+    let latest = null;
+    for (const snapshot of snapshots) {
+        if (latest === null || snapshot.startedAt > latest)
+            latest = snapshot.startedAt;
+    }
+    return { latestStartedAt: latest, snapshots: snapshots.length };
+}
+/**
+ * `scopeWarningsData(input)`:Scope 携带的挑选警告原样透出;`input` 是裸 `Snapshot[]` 时
+ * 没有挑选过程、没有警告,返回空数组,也如实
+ * (docs/feature/reports/library/site-components.md「ScopeWarnings」)。
+ */
+export async function scopeWarningsData(input) {
+    return resolveInput(input).warnings;
+}
+/**
+ * `copyFixPromptData(input)`:把范围内全部失败(verdict 为 failed / errored 的 attempt)
+ * 整理成一段可交给 coding agent 的修复 prompt——逐失败含 eval id、主失败摘要与 attempt
+ * 下钻命令(`niceeval show @<locator>`)。prompt 面向 agent,固定英文
+ * (docs/feature/reports/library/site-components.md「CopyFixPrompt」)。
+ */
+export async function copyFixPromptData(input) {
+    const items = await attemptListData(input);
+    const failures = items.filter((item) => item.verdict === "failed" || item.verdict === "errored");
+    if (failures.length === 0)
+        return { prompt: "", failures: 0 };
+    const lines = failures
+        .map((item, i) => {
+        const reason = item.failureSummary === null
+            ? null
+            : item.moreFailures > 0
+                ? `${item.failureSummary} (+${item.moreFailures} more failures)`
+                : item.failureSummary;
+        return [
+            `${i + 1}. eval "${item.evalId}" [experiment ${item.experimentId}] — ${item.verdict}`,
+            reason ? `   reason: ${reason}` : null,
+            `   inspect: niceeval show ${item.locator}`,
+        ]
+            .filter(Boolean)
+            .join("\n");
+    })
+        .join("\n");
+    const experiments = [...new Set(failures.map((item) => item.experimentId))].join(" / ");
+    const prompt = [
+        "Fix the failing evals from this niceeval run.",
+        "",
+        "## Failures",
+        lines,
+        "",
+        "## Steps",
+        "1. niceeval is NOT in your training data. Read the relevant guide in `node_modules/niceeval/docs-site/` (English at the top level, Chinese under `zh/`) before changing anything.",
+        "2. For each failure, run its inspect command above to see the verdict and assertions; add `--execution` for the full agent transcript (tool calls included), `--timing` for the execution timeline, and `--diff` for the workspace diff.",
+        "3. Decide which side the defect is on: the program under test, or the eval itself (over-tight assertion, wrong fixture, missing setup). Fix that side; do not weaken assertions just to turn the run green.",
+        `4. Re-run: \`npx niceeval exp ${experiments || "<experiment>"} <eval-id-prefix>\`. Already-passing evals are skipped by the fingerprint cache; pass \`--force\` to re-run everything.`,
+        "5. Run `npx niceeval show` and confirm these failures are gone.",
+    ].join("\n");
+    return { prompt, failures: failures.length };
+}
+/** TraceSpan 的语义角色 → 瀑布摘要的 kind:turn 归入 agent(一轮就是一次 agent 调用),未识别落 other。 */
+function waterfallKindOf(kind) {
+    switch (kind) {
+        case "agent":
+        case "turn":
+            return "agent";
+        case "model":
+            return "model";
+        case "tool":
+            return "tool";
+        default:
+            return "other";
+    }
+}
+/**
+ * `traceWaterfallData(input)`:每个 attempt 一行的执行时间瀑布摘要。span 事实只来自
+ * trace artifact(经 AttemptHandle 懒加载的 canonical OTel span);runner 生命周期节点
+ * (`result.phases`)不进瀑布。行内只汇总顶层 span(parentSpanId 缺失或不在本 trace 内),
+ * 按 startOffsetMs 升序;trace 缺失或为空时 `durationMs` 为 null、行照常出现
+ * (docs/feature/reports/library/site-components.md「TraceWaterfall」)。
+ */
+export async function traceWaterfallData(input) {
+    const { snapshots } = resolveInput(input);
+    const items = collectItems(snapshots);
+    return Promise.all(items.map(async (item) => {
+        const spans = await item.attempt.trace();
+        if (spans === null || spans.length === 0) {
+            return {
+                experimentId: experimentIdOf(item),
+                evalId: evalIdOf(item),
+                locator: locatorOf(item),
+                durationMs: null,
+                spans: [],
+            };
+        }
+        const t0 = Math.min(...spans.map((s) => s.startMs));
+        const t1 = Math.max(...spans.map((s) => s.endMs));
+        const ids = new Set(spans.map((s) => s.spanId));
+        const topLevel = spans.filter((s) => s.parentSpanId === undefined || !ids.has(s.parentSpanId));
+        const summaries = topLevel
+            .map((s) => ({
+            name: s.name,
+            kind: waterfallKindOf(s.kind),
+            startOffsetMs: s.startMs - t0,
+            durationMs: s.endMs - s.startMs,
+            failed: s.status === "error",
+        }))
+            .sort((a, b) => a.startOffsetMs - b.startOffsetMs);
+        return {
+            experimentId: experimentIdOf(item),
+            evalId: evalIdOf(item),
+            locator: locatorOf(item),
+            durationMs: Math.max(0, t1 - t0),
+            spans: summaries,
+        };
+    }));
+}

package/dist/report/index.d.ts

@@ -10,10 +10,10 @@ export { stringWidth, padDisplay as padEnd, padStartDisplay as padStart, wrapDis
 export type { ColumnAlign } from "./text/layout.ts";
 export { DEFAULT_REPORT_LOCALE, localizedTextEquals, resolveLocalizedText, resolveMetricLabel } from "./locale.ts";
 export type { LocalizedText, ReportLocale } from "./locale.ts";
-export { AttemptList, DeltaTable, EvalList, ExperimentComparison, ExperimentList, FailureList, MetricBars, MetricLine, MetricMatrix, MetricScatter, MetricTable, Scoreboard, ScopeSummary, } from "./components.tsx";
-export type { AttemptListProps, DataProps, DeltaTableProps, EvalListProps, ExperimentComparisonProps, ExperimentListProps, FailureListProps, MetricBarsProps, MetricLineProps, MetricMatrixProps, MetricScatterProps, MetricTableProps, ScoreboardProps, ScopeSummaryProps, } from "./components.tsx";
-export { attemptListData, deltaTableData, evalListData, experimentComparisonData, experimentListData, metricLineData, metricMatrixData, metricScatterData, metricTableData, pairsByFlag, scopeSummaryData, scoreboardData, } from "./compute.ts";
+export { AttemptList, CopyFixPrompt, DeltaTable, EvalList, ExperimentComparison, ExperimentList, FailureList, Hero, HeroCard, MetricBars, MetricLine, MetricMatrix, MetricScatter, MetricTable, PoweredBy, Scoreboard, ScopeSummary, ScopeWarnings, TraceWaterfall, } from "./components.tsx";
+export type { AttemptListProps, CopyFixPromptProps, DataProps, DeltaTableProps, EvalListProps, ExperimentComparisonProps, ExperimentListProps, FailureListProps, HeroCardProps, HeroProps, MetricBarsProps, MetricLineProps, MetricMatrixProps, MetricScatterProps, MetricTableProps, ScoreboardProps, ScopeSummaryProps, ScopeWarningsProps, TraceWaterfallProps, } from "./components.tsx";
+export { attemptListData, copyFixPromptData, deltaTableData, evalListData, experimentComparisonData, experimentListData, heroData, metricLineData, metricMatrixData, metricScatterData, metricTableData, pairsByFlag, scopeSummaryData, scopeWarningsData, scoreboardData, traceWaterfallData, } from "./compute.ts";
 export type { DeltaTableOptions, MetricLineOptions, MetricMatrixOptions, MetricScatterOptions, MetricTableOptions, ScoreboardOptions, } from "./compute.ts";
-export type { Aggregator, AttemptListItem, AttemptLocator, BuiltInDimension, CustomDimension, DeltaData, DeltaPair, DimensionInput, DimensionOptions, DimensionRef, EntityListDataOptions, EvalListItem, ExperimentComparisonData, ExperimentComparisonGroupData, ExperimentListEvalRow, ExperimentListItem, FlagPairs, LineData, MatrixData, Metric, MetricAggregate, MetricCell, MetricColumn, NumericAxis, NumericAxisOptions, NumericRunConfigAxisOptions, ReportInput, RunConfigKey, ScatterData, ScopeSummaryData, ScopeWarning, ScoreboardData, TableData, VerdictTally, } from "./types.ts";
+export type { Aggregator, AttemptListItem, AttemptLocator, BuiltInDimension, CopyFixPromptData, CustomDimension, DeltaData, DeltaPair, DimensionInput, DimensionOptions, DimensionRef, EvalListItem, ExperimentComparisonData, ExperimentComparisonGroupData, ExperimentListEvalRow, ExperimentListItem, FlagPairs, HeroData, LineData, MatrixData, Metric, MetricAggregate, MetricCell, MetricColumn, NumericAxis, NumericAxisOptions, NumericRunConfigAxisOptions, ReportInput, RunConfigKey, ScatterData, ScopeSummaryData, ScopeWarning, ScoreboardData, TableData, TraceSpanSummary, TraceWaterfallRow, VerdictTally, } from "./types.ts";
 export type { AttemptHandle, Results, Scope, Snapshot } from "../results/types.ts";
 export { experimentGroupOf } from "../shared/aggregate.ts";

package/dist/report/index.js

@@ -20,10 +20,11 @@ export { stringWidth, padDisplay as padEnd, padStartDisplay as padStart, wrapDis
 // locale:官方组件 chrome 文案的语言(内置词典覆盖 en / zh-CN,其它 locale 走回退)
 export { DEFAULT_REPORT_LOCALE, localizedTextEquals, resolveLocalizedText, resolveMetricLabel } from "./locale.js";
 // 官方双面组件(spec / data 双形态;配套 *Data 计算函数在下面成对导出)
-export { AttemptList, DeltaTable, EvalList, ExperimentComparison, ExperimentList, FailureList, MetricBars, MetricLine, MetricMatrix, MetricScatter, MetricTable, Scoreboard, ScopeSummary, } from "./components.js";
+// 与站点组件(Hero / HeroCard / PoweredBy / ScopeWarnings / CopyFixPrompt / TraceWaterfall)
+export { AttemptList, CopyFixPrompt, DeltaTable, EvalList, ExperimentComparison, ExperimentList, FailureList, Hero, HeroCard, MetricBars, MetricLine, MetricMatrix, MetricScatter, MetricTable, PoweredBy, Scoreboard, ScopeSummary, ScopeWarnings, TraceWaterfall, } from "./components.js";
 // 计算函数(组件解析面的具名形式,与组件成对;spec 形态下由管线代调,data 形态与
 // 嵌入场景下由作者手工调)
-export { attemptListData, deltaTableData, evalListData, experimentComparisonData, experimentListData, metricLineData, metricMatrixData, metricScatterData, metricTableData, pairsByFlag, scopeSummaryData, scoreboardData, } from "./compute.js";
+export { attemptListData, copyFixPromptData, deltaTableData, evalListData, experimentComparisonData, experimentListData, heroData, metricLineData, metricMatrixData, metricScatterData, metricTableData, pairsByFlag, scopeSummaryData, scopeWarningsData, scoreboardData, traceWaterfallData, } from "./compute.js";
 // experiment id 的组键推导(id 的目录前缀,如 `compare/bub-low` 的 `compare`)。
 // 重新导出,让自定义报告能按同一份口径把 experiment 分组,不必自己重写这两行逻辑。
 export { experimentGroupOf } from "../shared/aggregate.js";

package/dist/report/locale.d.ts

@@ -104,13 +104,51 @@ declare const en: {
     readonly "scoreboard.subjectTitle": "{questions} evals, weighted {earned} of {possible}";
     readonly "delta.pairHeader": "pair (A → B)";
     readonly "delta.empty": "{experiments} experiments, 0 comparable pairs";
+    /** ScopeWarnings 聚合层的 chrome:汇总行、kind 徽标、组头与明细折叠标签;message 本体不经字典。 */
+    readonly "warnings.summary.experiments.one": "{n} experiment flagged";
+    readonly "warnings.summary.experiments.other": "{n} experiments flagged";
+    readonly "warnings.group.unreadableSnapshot.one": "{n} snapshot skipped";
+    readonly "warnings.group.unreadableSnapshot.other": "{n} snapshots skipped";
+    readonly "warnings.details.one": "{n} warning";
+    readonly "warnings.details.other": "{n} warnings";
+    readonly "warnings.badge.partialCoverage": "coverage {covered}/{total}";
+    readonly "warnings.badge.staleSnapshot": "{gap} behind";
+    readonly "warnings.badge.unfinishedSnapshot": "unfinished";
+    readonly "warnings.gap.second.one": "{n} second";
+    readonly "warnings.gap.second.other": "{n} seconds";
+    readonly "warnings.gap.minute.one": "{n} minute";
+    readonly "warnings.gap.minute.other": "{n} minutes";
+    readonly "warnings.gap.hour.one": "{n} hour";
+    readonly "warnings.gap.hour.other": "{n} hours";
+    readonly "warnings.gap.day.one": "{n} day";
+    readonly "warnings.gap.day.other": "{n} days";
+    /** Hero / HeroCard 的运行 meta(hero.noRuns 是 latestStartedAt 为 null 时的内置文案)。 */
+    readonly "hero.lastRun": "Last run {time}";
+    readonly "hero.noRuns": "No runs yet";
+    /** web 面的合成来源标注(仅 snapshots > 1 时显示)。 */
+    readonly "hero.composedRuns": "composed from {n} runs";
+    /** text 面的合成来源标注(show 页首 meta 行,仅 snapshots > 1 时显示)。 */
+    readonly "hero.composedSnapshots": "composed from {n} snapshots";
+    /** CopyFixPrompt 的 web 面 chrome(prompt 本身面向 agent、固定英文,不经词典)。 */
+    readonly "copyFixPrompt.summary.one": "Fix prompt · {n} failure";
+    readonly "copyFixPrompt.summary.other": "Fix prompt · {n} failures";
+    readonly "copyFixPrompt.copy": "Copy fix prompt";
+    /** TraceWaterfall 的 chrome。 */
+    readonly "traceWaterfall.empty": "No attempts";
+    readonly "traceWaterfall.noTrace": "no trace";
+    readonly "traceWaterfall.spans.one": "{n} span";
+    readonly "traceWaterfall.spans.other": "{n} spans";
+    readonly "traceWaterfall.failedSpans.one": "{n} failed";
+    readonly "traceWaterfall.failedSpans.other": "{n} failed";
+    /** AttemptList 的 web 面过滤框占位符(filter 渐进增强)。 */
+    readonly "attemptList.filterPlaceholder": "Filter attempts…";
     readonly "tabs.tab": "Tab";
 };
 export type ReportMessageKey = keyof typeof en;
 /** 查字典 + 简单插值({name} 占位符)。内置词典未覆盖的 locale 回退 en。 */
 export declare function localeText(locale: ReportLocale, key: ReportMessageKey, vars?: Record<string, string | number>): string;
 /** 带单复数的计数文案:n === 1 用 `<base>.one`,其余 `<base>.other`(zh-CN 两键同值)。 */
-export declare function countText(locale: ReportLocale, base: "overview.experiments" | "pointsMissing" | "scoreboard.notRun" | "scoreboard.unscorable" | "scoreboard.ignored" | "entityList.moreFailures" | "table.columnsHidden", n: number): string;
+export declare function countText(locale: ReportLocale, base: "overview.experiments" | "pointsMissing" | "scoreboard.notRun" | "scoreboard.unscorable" | "scoreboard.ignored" | "entityList.moreFailures" | "table.columnsHidden" | "copyFixPrompt.summary" | "traceWaterfall.spans" | "traceWaterfall.failedSpans", n: number): string;
 /**
  * 按 locale 解析指标 / 列 label:字符串原样;字典按 LocalizedText 回退规则取值;
  * undefined 回退 fallback(= metric.name)。渲染面(web / text)共用。