niceeval 0.4.5 → 0.5.0

package/src/view/app/App.tsx

@@ -2,9 +2,9 @@ import { useCallback, useEffect, useMemo, useRef, useState } from "react";
 import { detectLocale, makeTranslator, persistLocale, setDocumentLocale } from "./i18n.ts";
 import type { MessageKey } from "./i18n.ts";
 import type { Locale, LocalizedText, SortKey, SortState, Tab, ViewData, ViewResult, ViewRow } from "./types.ts";
-import { buildGroupMap, compareRows, resultFromUrl } from "./lib/rows.ts";
+import { CELL_KEYS, buildGroupMap, buildRows, compareRows, flattenAttempts, resultFromUrl } from "./lib/rows.ts";
 import { formatAttemptHash, parseAttemptHash, resolveAttemptRef, unresolvedAttemptWarning } from "./lib/attempt-route.ts";
-import { formatCost, formatDateTime, formatDuration, formatPercent } from "./lib/format.ts";
+import { formatCost, formatDateTime, formatDuration } from "./lib/format.ts";
 import { Metric } from "./components/primitives.tsx";
 import { GroupSelector } from "./components/GroupSelector.tsx";
 import { CostScoreChart } from "./components/CostScoreChart.tsx";
@@ -15,12 +15,20 @@ import { Tabs, TabsContent, TabsList, TabsTrigger } from "./components/ui/tabs.t
 import { RunsView } from "./pages/RunsPage.tsx";
 import { TracesView } from "./pages/TracesPage.tsx";
 
+// 报告槽只有一个:默认装官方榜单(Experiments tab);--report 整槽替换成用户报告。
+// 证据室(Runs / Traces / AttemptModal)是 view 本体,两种模式下原样保留。
 export const navItems: { id: Tab; label: MessageKey }[] = [
   { id: "experiments", label: "nav.experiments" },
   { id: "runs", label: "nav.runs" },
   { id: "traces", label: "nav.traces" },
 ];
 
+export const reportNavItems: { id: Tab; label: MessageKey }[] = [
+  { id: "report", label: "nav.report" },
+  { id: "runs", label: "nav.runs" },
+  { id: "traces", label: "nav.traces" },
+];
+
 /** 把 config.name 解析成当前界面语言的一条文案:字符串原样返回,多语言映射按 locale 挑,挑不到回退 en / 第一条。 */
 export function localizedText(text: LocalizedText | undefined, locale: Locale): string | undefined {
   if (!text) return undefined;
@@ -29,23 +37,26 @@ export function localizedText(text: LocalizedText | undefined, locale: Locale):
 }
 
 /** 初始 URL → 直接打开的 attempt:先认 #/attempt/<run>/<result> 深链,回退旧版 ?modal= 参数。 */
-function modalResultFromLocation(rows: ViewRow[]): ViewResult | null {
+function modalResultFromLocation(snapshots: ViewData["snapshots"]): ViewResult | null {
   const ref = parseAttemptHash(location.hash);
   if (ref) {
-    const found = resolveAttemptRef(rows, ref);
+    const found = resolveAttemptRef(snapshots, ref);
     if (found) return found;
     // 定位不到(run 不在、下标越界、旧格式数据):不开空 modal,页面照常渲染。
     console.warn(unresolvedAttemptWarning(location.hash));
     return null;
   }
-  return resultFromUrl(rows);
+  return resultFromUrl(snapshots);
 }
 
-export function App({ data }: { data: ViewData }) {
-  const rows = data.rows ?? [];
+export function App({ data, reportHtml }: { data: ViewData; reportHtml?: string }) {
+  const snapshots = data.snapshots ?? [];
+  const hasReport = reportHtml !== undefined;
+  const rows = useMemo(() => buildRows(data), [data]);
+  const attempts = useMemo(() => flattenAttempts(snapshots), [snapshots]);
   const [locale, setLocale] = useState<Locale>(() => detectLocale());
   const t = useMemo(() => makeTranslator(locale), [locale]);
-  const [tab, setTab] = useState<Tab>("experiments");
+  const [tab, setTab] = useState<Tab>(hasReport ? "report" : "experiments");
   const [sort, setSort] = useState<SortState>({ key: "passRate", dir: -1 });
   const [query, setQuery] = useState("");
   const [openRows, setOpenRows] = useState<Set<string>>(() => new Set());
@@ -53,7 +64,7 @@ export function App({ data }: { data: ViewData }) {
     const groups = [...new Set(rows.map((r) => r.group).filter(Boolean))].sort();
     return groups[0] ?? null;
   });
-  const [modalResult, setModalResult] = useState<ViewResult | null>(() => modalResultFromLocation(rows));
+  const [modalResult, setModalResult] = useState<ViewResult | null>(() => modalResultFromLocation(snapshots));
   // 当前 modal 的 hash 历史条目前面是否还有本页条目(本页 push 的 / 前进键回到的):
   // true → UI 关闭走 history.back(),前进键还能重新打开;false(深链直接落地)→ 原地抹 hash,
   // 免得 back 把用户弹出站外。
@@ -107,7 +118,7 @@ export function App({ data }: { data: ViewData }) {
         setModalResult(null);
         return;
       }
-      const found = resolveAttemptRef(rows, ref);
+      const found = resolveAttemptRef(snapshots, ref);
       if (!found) {
         console.warn(unresolvedAttemptWarning(location.hash));
         setModalResult(null);
@@ -119,7 +130,7 @@ export function App({ data }: { data: ViewData }) {
     };
     window.addEventListener("hashchange", onHashChange);
     return () => window.removeEventListener("hashchange", onHashChange);
-  }, [rows]);
+  }, [snapshots]);
 
   const groupMap = useMemo<Map<string, ViewRow[]>>(() => buildGroupMap(rows), [rows]);
   const pool = selectedGroup ? groupMap.get(selectedGroup) ?? [] : rows;
@@ -158,6 +169,11 @@ export function App({ data }: { data: ViewData }) {
     });
   };
 
+  // hero 的官方数字:整体通过率是 MetricTable.data 常量维度的单行格子,其余来自 RunOverview.data。
+  const overallPassRate = data.overall?.rows[0]?.cells[CELL_KEYS.passRate];
+  const totals = data.overview?.totals;
+  const warnings = data.overview?.warnings ?? [];
+
   return (
     <Tabs value={tab} onValueChange={(v) => setTab(v as Tab)}>
       <header className="topbar">
@@ -166,7 +182,7 @@ export function App({ data }: { data: ViewData }) {
           <span>NiceEval</span>
         </a>
         <TabsList aria-label={t("nav.label")}>
-          {navItems.map((item) => (
+          {(hasReport ? reportNavItems : navItems).map((item) => (
             <TabsTrigger key={item.id} value={item.id}>
               {t(item.label)}
             </TabsTrigger>
@@ -194,27 +210,62 @@ export function App({ data }: { data: ViewData }) {
               {/* viewData 只带原始值(ISO / number),这里按当前界面 locale 格式化。 */}
               <b>{t("hero.lastRun")}</b> {data.lastRunAt ? formatDateTime(data.lastRunAt, locale) : t("hero.noRuns")}
             </span>
+            {data.composedRuns > 0 ? (
+              // 榜单是跨 run 合成的现刻水位,表头如实标注合成来源(几个 run)。
+              <span>{t("hero.composedFrom", { count: data.composedRuns })}</span>
+            ) : null}
           </div>
         </section>
 
-        <section className="summary" aria-label="Run summary">
-          <Metric label={t("metric.passRate")} value={formatPercent(data.passRate)} />
-          <Metric label={t("metric.evalResults")} value={String(data.resultCount)} />
-          <Metric label={t("metric.duration")} value={formatDuration(data.durationMs)} />
-          <Metric label={t("metric.cost")} value={formatCost(data.estimatedCostUSD)} />
-        </section>
+        {/* 官方水位 KPI 与挑选警告属于报告槽:--report 整槽替换后由用户报告自己决定
+            摆不摆(<DefaultReport /> / <RunOverview /> 都能把它们摆回来)。 */}
+        {!hasReport && (
+          <section className="summary" aria-label="Run summary">
+            <Metric label={t("metric.passRate")} value={overallPassRate?.display ?? "—"} />
+            <Metric label={t("metric.evalResults")} value={String(totals?.evals ?? 0)} />
+            <Metric label={t("metric.duration")} value={formatDuration(totals?.durationMs ?? 0)} />
+            <Metric label={t("metric.cost")} value={formatCost(totals?.costUSD ?? undefined)} />
+          </section>
+        )}
+
+        {!hasReport && warnings.length > 0 && (
+          // 选集警告(partial-coverage / stale-snapshot / synthetic-experiment-id):
+          // message 是挑选器渲染好的英文句子,原样打;data-kind 供样式与测试定位。
+          <section className="incompatible-banner selection-warnings" role="alert">
+            <b>{t("banner.warningsTitle")}</b>
+            <ul>
+              {warnings.map((w, i) => (
+                <li key={`${w.kind}-${i}`} data-kind={w.kind}>
+                  <span className="ib-meta">{w.message}</span>
+                </li>
+              ))}
+            </ul>
+          </section>
+        )}
 
         {(data.skippedRuns?.length ?? 0) > 0 && (
           <section className="incompatible-banner" role="alert">
             <b>{t("banner.skippedTitle")}</b>
             <ul>
               {data.skippedRuns!.map((run) => (
-                <li key={run.dir}>
+                <li key={run.dir} data-reason={run.reason}>
                   <span className="ib-dir">{run.dir}</span>
                   <span className="ib-meta">
-                    {run.reason === "incompatible-version"
-                      ? t("banner.skipped.incompatible", { producer: run.producerVersion ?? "?", schemaVersion: run.schemaVersion })
-                      : t("banner.skipped.malformed", { detail: run.detail ?? "?" })}
+                    {run.reason === "malformed"
+                      ? t("banner.skipped.malformed", { detail: run.detail ?? "?" })
+                      : run.reason === "incomplete"
+                        ? t("banner.skipped.incomplete")
+                        : run.producerName && run.producerName !== "niceeval"
+                          ? // 第三方 harness 写的落盘:如实报名字和版本,不拼 npx 命令。
+                            t("banner.skipped.incompatibleForeign", {
+                              name: run.producerName,
+                              version: run.producerVersion ?? "?",
+                              schemaVersion: run.schemaVersion,
+                            })
+                          : t("banner.skipped.incompatible", {
+                              producer: run.producerVersion ?? "?",
+                              schemaVersion: run.schemaVersion,
+                            })}
                   </span>
                   {run.command ? <code>{run.command}</code> : null}
                 </li>
@@ -223,6 +274,15 @@ export function App({ data }: { data: ViewData }) {
           </section>
         )}
 
+        {hasReport && (
+          <TabsContent value="report" id="tab-report">
+            {/* 报告槽:server 侧渲染好的静态 HTML(含 <Style> 产物),这里只摆放。
+                attempt 深链是普通 <a href="#/attempt/…">,经 hashchange 打开证据室弹窗。 */}
+            <div className="report-slot" dangerouslySetInnerHTML={{ __html: reportHtml }} />
+          </TabsContent>
+        )}
+
+        {!hasReport && (
         <TabsContent value="experiments" id="tab-experiments">
           <div className="section-head">
             <h2>{t("section.experiments")}</h2>
@@ -261,12 +321,13 @@ export function App({ data }: { data: ViewData }) {
             </div>
           )}
         </TabsContent>
+        )}
 
         <TabsContent value="runs">
-          <RunsView rows={rows} t={t} />
+          <RunsView attempts={attempts} t={t} />
         </TabsContent>
         <TabsContent value="traces">
-          <TracesView rows={rows} t={t} />
+          <TracesView attempts={attempts} t={t} />
         </TabsContent>
       </main>
       {modalResult && <AttemptModal result={modalResult} onClose={closeModal} t={t} />}

package/src/view/app/components/CostScoreChart.tsx

@@ -1,6 +1,7 @@
 import { useState } from "react";
 import type { T } from "../shared.ts";
 import type { ViewRow } from "../types.ts";
+import { CELL_KEYS } from "../lib/rows.ts";
 import { formatCost, formatPercent } from "../lib/format.ts";
 import { layoutLabelOffsets, niceTicks, seriesColor } from "../lib/chart.ts";
 
@@ -13,22 +14,40 @@ function formatAxisCost(value: number): string {
   return value === 0 ? "$0" : formatCost(value);
 }
 
+interface ChartPoint {
+  key: string;
+  label: string;
+  agent: string;
+  model?: string;
+  /** 官方 cost 格子的值(平均每 eval 成本,与轴标签同口径)。 */
+  cost: number;
+  /** 官方 pass-rate 格子的值(0–1)。 */
+  rate: number;
+}
+
 /**
  * 一个 group 内「成本 vs 通过率」散点图:一行(一个实验/config)= 一个点。
- * 只在 group 内至少两条 row 有 estimatedCostUSD 时渲染,否则没有比较意义。
+ * 点位取官方 MetricTable.data 的格子值;任一轴缺数据(null)的行不画。
+ * 只在 group 内至少两条 row 有成本时渲染,否则没有比较意义。
  */
 export function CostScoreChart({ rows, t }: { rows: ViewRow[]; t: T }) {
   const [hoverKey, setHoverKey] = useState<string | null>(null);
-  const points = rows.filter((r): r is ViewRow & { estimatedCostUSD: number } => typeof r.estimatedCostUSD === "number" && Number.isFinite(r.estimatedCostUSD));
+  const points: ChartPoint[] = [];
+  for (const row of rows) {
+    const cost = row.cells[CELL_KEYS.cost]?.value ?? null;
+    const rate = row.cells[CELL_KEYS.passRate]?.value ?? null;
+    if (cost === null || rate === null || !Number.isFinite(cost)) continue;
+    points.push({ key: row.key, label: row.label, agent: row.agent, model: row.model, cost, rate });
+  }
   if (points.length < 2) return null;
 
   const plotW = WIDTH - MARGIN.left - MARGIN.right;
   const plotH = HEIGHT - MARGIN.top - MARGIN.bottom;
 
-  const xTicks = niceTicks(0, Math.max(...points.map((p) => p.estimatedCostUSD)), 5);
+  const xTicks = niceTicks(0, Math.max(...points.map((p) => p.cost)), 5);
   const xMax = xTicks[xTicks.length - 1] || 1;
 
-  const rates = points.map((p) => p.passRate);
+  const rates = points.map((p) => p.rate);
   const rateMin = Math.min(...rates);
   const rateMax = Math.max(...rates);
   const pad = Math.max(0.03, (rateMax - rateMin) * 0.25);
@@ -44,8 +63,8 @@ export function CostScoreChart({ rows, t }: { rows: ViewRow[]; t: T }) {
   const hovered = points.find((p) => p.key === hoverKey);
 
   const positions = points.map((p) => {
-    const cx = xScale(p.estimatedCostUSD);
-    const cy = yScale(p.passRate);
+    const cx = xScale(p.cost);
+    const cy = yScale(p.rate);
     const anchorLeft = cx >= MARGIN.left + plotW * 0.72;
     return { cx, cy, anchorLeft, width: p.label.length * 6.4 + 10 };
   });
@@ -93,7 +112,7 @@ export function CostScoreChart({ rows, t }: { rows: ViewRow[]; t: T }) {
                   className="csc-point"
                   tabIndex={0}
                   role="button"
-                  aria-label={`${p.label}: ${formatCost(p.estimatedCostUSD)}, ${formatPercent(p.passRate)}`}
+                  aria-label={`${p.label}: ${formatCost(p.cost)}, ${formatPercent(p.rate)}`}
                   onMouseEnter={() => setHoverKey(p.key)}
                   onMouseLeave={() => setHoverKey((k) => (k === p.key ? null : k))}
                   onFocus={() => setHoverKey(p.key)}
@@ -114,11 +133,11 @@ export function CostScoreChart({ rows, t }: { rows: ViewRow[]; t: T }) {
         {hovered ? (
           <div
             className="csc-tooltip"
-            style={{ left: `${(xScale(hovered.estimatedCostUSD) / WIDTH) * 100}%`, top: `${(yScale(hovered.passRate) / HEIGHT) * 100}%` }}
+            style={{ left: `${(xScale(hovered.cost) / WIDTH) * 100}%`, top: `${(yScale(hovered.rate) / HEIGHT) * 100}%` }}
           >
-            <b>{formatPercent(hovered.passRate)}</b> {t("chart.axisScore")}
+            <b>{formatPercent(hovered.rate)}</b> {t("chart.axisScore")}
             <div className="csc-tooltip-meta">
-              {formatCost(hovered.estimatedCostUSD)} · {hovered.model || hovered.agent}
+              {formatCost(hovered.cost)} · {hovered.model || hovered.agent}
             </div>
           </div>
         ) : null}

package/src/view/app/components/ExperimentTable.tsx

@@ -3,9 +3,17 @@ import { ChevronRight } from "lucide-react";
 import type { OpenModal, T } from "../shared.ts";
 import type { Assertion, SortKey, SortState, ViewResult, ViewRow } from "../types.ts";
 import { EvalGroup, failingAssertions, groupByEval, outcomeClass, outcomeLabel, outcomeSummary, reasonFor, scoresSummary } from "../lib/outcome.ts";
-import { configChips } from "../lib/rows.ts";
-import { formatClock, formatCost, formatDateTime, formatDuration, formatPercent, formatTokens, totalTokens } from "../lib/format.ts";
+import { CELL_KEYS, configChips } from "../lib/rows.ts";
+import { formatClock, formatCost, formatDateTime, formatDuration, formatTokens, totalTokens } from "../lib/format.ts";
 import { Kpi, SortHeader } from "./primitives.tsx";
+import type { MetricCell } from "../types.ts";
+
+/** 官方格子的渲染:display 已格式化;samples < total 时 title 如实报覆盖率(有 attempt 测不了这个指标)。 */
+function CellValue({ cell }: { cell: MetricCell | undefined }) {
+  if (!cell || cell.value === null) return <>—</>;
+  const partial = cell.samples < cell.total;
+  return <span title={partial ? `${cell.samples}/${cell.total} attempts measured` : undefined}>{cell.display}</span>;
+}
 
 export function ExperimentTable({
   rows,
@@ -32,7 +40,7 @@ export function ExperimentTable({
             <SortHeader name={t("table.experiment")} sortKey="experiment" sort={sort} onSort={setSortKey} />
             <SortHeader name={t("table.model")} sortKey="model" sort={sort} onSort={setSortKey} />
             <SortHeader name={t("table.agent")} sortKey="agent" sort={sort} onSort={setSortKey} />
-            <SortHeader name={t("table.avgDuration")} sortKey="avgDurationMs" sort={sort} onSort={setSortKey} />
+            <SortHeader name={t("table.avgDuration")} sortKey="duration" sort={sort} onSort={setSortKey} />
             <SortHeader name={t("table.successRate")} sortKey="passRate" sort={sort} onSort={setSortKey} />
             <SortHeader name={t("table.tokens")} sortKey="tokens" sort={sort} onSort={setSortKey} />
             <SortHeader name={t("table.estCost")} sortKey="cost" sort={sort} onSort={setSortKey} />
@@ -53,7 +61,8 @@ export function ExperimentTable({
 }
 
 export function ExperimentRow({ row, open, onToggle, t }: { row: ViewRow; open: boolean; onToggle: () => void; t: T }) {
-  const tone = row.passRate >= 0.8 ? "good" : row.passRate >= 0.5 ? "warn" : "bad";
+  const passRate = row.cells[CELL_KEYS.passRate]?.value ?? null;
+  const tone = passRate === null ? "" : passRate >= 0.8 ? "good" : passRate >= 0.5 ? "warn" : "bad";
   return (
     <tr
       className={`main-row${open ? " is-open" : ""}`}
@@ -77,10 +86,19 @@ export function ExperimentRow({ row, open, onToggle, t }: { row: ViewRow; open:
       </td>
       <td>{row.model || t("config.default")}</td>
       <td>{row.agent}</td>
-      <td className="num">{formatDuration(row.avgDurationMs)}</td>
-      <td className={`num ${tone}`}>{formatPercent(row.passRate)}</td>
-      <td className="num">{formatTokens(totalTokens(row.usage))}</td>
-      <td className="num">{formatCost(row.estimatedCostUSD)}</td>
+      {/* 官方 MetricTable.data 的格子:display 直接渲染,数字口径与 show 榜单同源。 */}
+      <td className="num">
+        <CellValue cell={row.cells[CELL_KEYS.duration]} />
+      </td>
+      <td className={`num ${tone}`}>
+        <CellValue cell={row.cells[CELL_KEYS.passRate]} />
+      </td>
+      <td className="num">
+        <CellValue cell={row.cells[CELL_KEYS.tokens]} />
+      </td>
+      <td className="num">
+        <CellValue cell={row.cells[CELL_KEYS.cost]} />
+      </td>
       <td>
         <span className="pill">{outcomeSummary(row, t)}</span>
       </td>
@@ -115,7 +133,7 @@ export function ExperimentDetail({ row, openModal, t }: { row: ViewRow; openModa
             <Kpi label={t("detail.errored")} value={row.errored} className={row.errored ? "infra-err" : ""} />
             {row.runs > row.evals ? <Kpi label={t("detail.runs")} value={row.runs} /> : null}
             <Kpi label={t("detail.totalTime")} value={formatDuration(totalDuration)} />
-            <Kpi label={t("detail.totalCost")} value={formatCost(row.estimatedCostUSD)} />
+            <Kpi label={t("detail.totalCost")} value={formatCost(row.totalCostUSD)} />
             <Kpi label={t("detail.ran")} value={formatDateTime(row.lastRunAt)} title={row.lastRunAt || ""} />
           </div>
           <h3>{t("detail.evaluationAttempts")}</h3>

package/src/view/app/components/GroupSelector.tsx

@@ -25,7 +25,7 @@ export function GroupSelector({
         const errored = evalGroups.filter((g) => g.outcome === "errored").length;
         const passRate = evalPassRate(allResults);
         const tone = passRate >= 0.8 ? "good" : passRate >= 0.5 ? "warn" : "bad";
-        const totalCost = groupRows.reduce((s: number, r: ViewRow) => s + (r.estimatedCostUSD || 0), 0);
+        const totalCost = groupRows.reduce((s: number, r: ViewRow) => s + (r.totalCostUSD || 0), 0);
         const lastRun = groupRows
           .map((r: ViewRow) => r.lastRunAt)
           .filter((value): value is string => Boolean(value))

package/src/view/app/i18n.ts

@@ -6,6 +6,7 @@ import { interpolate, normalizeLocale, type Locale, type Vars } from "../../i18n
 export type MessageKey =
   | "app.title"
   | "nav.label"
+  | "nav.report"
   | "nav.experiments"
   | "nav.runs"
   | "nav.traces"
@@ -57,7 +58,7 @@ export type MessageKey =
   | "detail.rawSample"
   | "detail.rawNote"
   | "config.experiment"
-  | "config.flagsNone"
+  | "config.paramsNone"
   | "config.default"
   | "config.none"
   | "config.notApplicable"
@@ -109,7 +110,11 @@ export type MessageKey =
   | "outcome.skipped"
   | "banner.skippedTitle"
   | "banner.skipped.incompatible"
+  | "banner.skipped.incompatibleForeign"
   | "banner.skipped.malformed"
+  | "banner.skipped.incomplete"
+  | "banner.warningsTitle"
+  | "hero.composedFrom"
   | "chart.costVsScore"
   | "chart.axisCost"
   | "chart.axisScore";
@@ -120,6 +125,7 @@ const dictionaries: Record<Locale, Dictionary> = {
   en: {
     "app.title": "niceeval experiment view",
     "nav.label": "Report",
+    "nav.report": "Report",
     "nav.experiments": "Experiments",
     "nav.runs": "Runs",
     "nav.traces": "Traces",
@@ -171,7 +177,7 @@ const dictionaries: Record<Locale, Dictionary> = {
     "detail.rawSample": "Raw sample result",
     "detail.rawNote": "debug JSON, defaults to first error/failure when available",
     "config.experiment": "experiment",
-    "config.flagsNone": "none",
+    "config.paramsNone": "none",
     "config.default": "default",
     "config.none": "none",
     "config.notApplicable": "n/a",
@@ -223,7 +229,11 @@ const dictionaries: Record<Locale, Dictionary> = {
     "outcome.skipped": "skipped",
     "banner.skippedTitle": "Some runs could not be loaded and are not shown here:",
     "banner.skipped.incompatible": "written by niceeval {{producer}} (schemaVersion {{schemaVersion}}) — view it with the command on the right",
+    "banner.skipped.incompatibleForeign": "written by {{name}} {{version}} (schemaVersion {{schemaVersion}}) — this viewer cannot read it; open it with the tool that produced it",
     "banner.skipped.malformed": "unreadable report ({{detail}}) — it may be corrupted; re-run the eval or delete this run directory",
+    "banner.skipped.incomplete": "the run crashed before writing summary.json — completed attempt artifacts remain on disk for manual inspection; delete the directory if you no longer need them",
+    "banner.warningsTitle": "Heads-up about the current leaderboard selection:",
+    "hero.composedFrom": "Composed from {{count}} run(s)",
     "chart.costVsScore": "Cost vs. Score",
     "chart.axisCost": "Avg cost per eval",
     "chart.axisScore": "Pass rate",
@@ -231,6 +241,7 @@ const dictionaries: Record<Locale, Dictionary> = {
   "zh-CN": {
     "app.title": "niceeval 实验查看器",
     "nav.label": "报告",
+    "nav.report": "报告",
     "nav.experiments": "实验",
     "nav.runs": "运行",
     "nav.traces": "追踪",
@@ -282,7 +293,7 @@ const dictionaries: Record<Locale, Dictionary> = {
     "detail.rawSample": "原始样例结果",
     "detail.rawNote": "调试 JSON，默认选择第一条错误/失败",
     "config.experiment": "实验",
-    "config.flagsNone": "无",
+    "config.paramsNone": "无",
     "config.default": "默认",
     "config.none": "无",
     "config.notApplicable": "不适用",
@@ -334,7 +345,11 @@ const dictionaries: Record<Locale, Dictionary> = {
     "outcome.skipped": "跳过",
     "banner.skippedTitle": "以下 run 读取失败,此处不展示:",
     "banner.skipped.incompatible": "由 niceeval {{producer}} 写入(schemaVersion {{schemaVersion}})—— 用右侧命令查看",
+    "banner.skipped.incompatibleForeign": "由 {{name}} {{version}} 写入(schemaVersion {{schemaVersion}})—— 当前查看器读不了;请用写出它的工具查看",
     "banner.skipped.malformed": "报告读不了({{detail}})—— 可能已损坏;重跑该 eval 或删除这个 run 目录",
+    "banner.skipped.incomplete": "该 run 中途崩溃、没写出 summary.json —— 已完成的 attempt 工件仍在盘上供手工排查;不需要就删除该目录",
+    "banner.warningsTitle": "当前榜单挑选的提醒:",
+    "hero.composedFrom": "合成自 {{count}} 个 run",
     "chart.costVsScore": "成本 × 通过率",
     "chart.axisCost": "平均每个 eval 成本",
     "chart.axisScore": "通过率",

package/src/view/app/lib/attempt-route.test.ts

@@ -2,7 +2,7 @@
 
 import { describe, expect, it } from "vitest";
 import { formatAttemptHash, parseAttemptHash, resolveAttemptRef } from "./attempt-route.ts";
-import type { ViewResult, ViewRow } from "../types.ts";
+import type { ViewResult, ViewSnapshot } from "../types.ts";
 
 const attempt = (run: string, index: number): ViewResult => ({
   id: `demo/eval-${index}`,
@@ -14,8 +14,8 @@ const attempt = (run: string, index: number): ViewResult => ({
   attemptRef: { run, result: index },
 });
 
-// 榜单行只有 results 参与定位,其余聚合字段与路由无关。
-const row = (results: ViewResult[]): ViewRow => ({ results }) as ViewRow;
+// 快照只有 results 参与定位,其余元信息字段与路由无关。
+const snap = (results: ViewResult[]): ViewSnapshot => ({ results }) as ViewSnapshot;
 
 describe("parseAttemptHash", () => {
   it("parses the canonical run-dir + index shape", () => {
@@ -69,21 +69,21 @@ describe("parseAttemptHash", () => {
 describe("resolveAttemptRef", () => {
   const runA = "2026-07-01T10-00-00-000Z";
   const runB = "2026-07-02T10-00-00-000Z";
-  const rows = [row([attempt(runA, 0), attempt(runB, 0)]), row([attempt(runA, 1)])];
+  const snapshots = [snap([attempt(runA, 0), attempt(runB, 0)]), snap([attempt(runA, 1)])];
 
   it("finds the attempt whose injected ref matches run + index", () => {
-    expect(resolveAttemptRef(rows, { run: runA, result: 1 })).toBe(rows[1]!.results[0]);
-    expect(resolveAttemptRef(rows, { run: runB, result: 0 })).toBe(rows[0]!.results[1]);
+    expect(resolveAttemptRef(snapshots, { run: runA, result: 1 })).toBe(snapshots[1]!.results[0]);
+    expect(resolveAttemptRef(snapshots, { run: runB, result: 0 })).toBe(snapshots[0]!.results[1]);
   });
 
   it("returns null for unknown runs and out-of-range indexes", () => {
-    expect(resolveAttemptRef(rows, { run: "no-such-run", result: 0 })).toBeNull();
-    expect(resolveAttemptRef(rows, { run: runA, result: 99 })).toBeNull();
+    expect(resolveAttemptRef(snapshots, { run: "no-such-run", result: 0 })).toBeNull();
+    expect(resolveAttemptRef(snapshots, { run: runA, result: 99 })).toBeNull();
   });
 
   it("returns null when results predate attemptRef injection (old baked data)", () => {
     const legacy = attempt(runA, 0);
     delete (legacy as { attemptRef?: unknown }).attemptRef;
-    expect(resolveAttemptRef([row([legacy])], { run: runA, result: 0 })).toBeNull();
+    expect(resolveAttemptRef([snap([legacy])], { run: runA, result: 0 })).toBeNull();
   });
 });

package/src/view/app/lib/attempt-route.ts

@@ -3,7 +3,7 @@
 // 这里只做纯解析 / 格式化 / 匹配,不碰 location / history,方便单测。
 // hash 目前只有这一种路由:tab 切换是纯组件 state,旧版 modal 深链走 ?modal= 查询参数,互不占用。
 
-import type { AttemptRef, ViewResult, ViewRow } from "../types.ts";
+import type { AttemptRef, ViewResult, ViewSnapshot } from "../types.ts";
 
 export const ATTEMPT_HASH_PREFIX = "#/attempt/";
 
@@ -33,10 +33,10 @@ export function parseAttemptHash(hash: string): AttemptRef | null {
   return { run, result: parseInt(indexPart, 10) };
 }
 
-/** 在榜单行里找 AttemptRef 指向的 attempt;旧格式烘焙的数据没有 attemptRef,自然找不到。 */
-export function resolveAttemptRef(rows: ViewRow[], ref: AttemptRef): ViewResult | null {
-  for (const row of rows) {
-    for (const result of row.results ?? []) {
+/** 在全部快照(含历史)里找 AttemptRef 指向的 attempt;旧格式烘焙的数据没有 attemptRef,自然找不到。 */
+export function resolveAttemptRef(snapshots: ViewSnapshot[], ref: AttemptRef): ViewResult | null {
+  for (const snapshot of snapshots) {
+    for (const result of snapshot.results ?? []) {
       if (result.attemptRef?.run === ref.run && result.attemptRef.result === ref.result) return result;
     }
   }

package/src/view/app/lib/outcome.ts

@@ -75,9 +75,7 @@ export function scoresSummary(assertions: Assertion[]): string {
 }
 
 export function outcomeSummary(row: ViewRow, t: T): string {
-  // Fold legacy "scored" rows into the passed count.
-  const passed = (row.passed || 0) + (row.scored || 0);
-  const parts = [`${passed} ${t("outcome.passed")}`, `${row.failed} ${t("outcome.failed")}`];
+  const parts = [`${row.passed} ${t("outcome.passed")}`, `${row.failed} ${t("outcome.failed")}`];
   if (row.errored) parts.push(`${row.errored} ${t("outcome.errored")}`);
   if (row.skipped) parts.push(`${row.skipped} ${t("outcome.skipped")}`);
   return parts.join(" / ");