feat: add persisted batch text testing

2026-08-10 13:49:24 +08:00
parent 5b8b9fd097
commit 19e8c8c108
27 changed files with 3855 additions and 1450 deletions
--- a/frontend/src/components/assistant-editor/debug-auto-test.tsx
+++ b/frontend/src/components/assistant-editor/debug-auto-test.tsx
@@ -66,15 +66,10 @@ export type DebugTestCase = {
  failAtTurn?: number;
 };

-/** 测试集：Debug Drawer 只按「当前助手关联」浏览，不做管理 */
+/** 测试集：Debug Drawer 只浏览和选择，不做管理。 */
 export type DebugTestSuite = {
  id: string;
  name: string;
-  /**
-   * 关联助手 id。MVP 用 `"*"` 表示「当前正在调试的助手都可见」；
-   * 接上真实 API 后改成具体 assistantId 列表。
-   */
-  assistantIds: string[];
  caseIds: string[];
 };

@@ -295,15 +290,11 @@ export const MOCK_TEST_CASES: DebugTestCase[] = [
  ),
 ];

-/**
- * 当前助手关联的测试集（MVP mock）。
- * `"*"` = 任意正在调试的助手都可见，避免把「系统全部测试集」甩进 Drawer。
- */
+/** 测试集（MVP mock）；测试集本身不绑定助手。 */
 export const MOCK_TEST_SUITES: DebugTestSuite[] = [
  {
    id: "suite-accident-basic",
    name: "事故快处基础流程",
-    assistantIds: ["*"],
    caseIds: [
      "tc-dual-car",
      "tc-no-injury",
@@ -314,7 +305,6 @@ export const MOCK_TEST_SUITES: DebugTestSuite[] = [
  {
    id: "suite-clarify",
    name: "异常输入与澄清",
-    assistantIds: ["*"],
    caseIds: [
      "tc-hello",
      "tc-noise",
@@ -326,13 +316,11 @@ export const MOCK_TEST_SUITES: DebugTestSuite[] = [
  {
    id: "suite-tools",
    name: "工具与系统联动",
-    assistantIds: ["*"],
    caseIds: ["tc-injury-transfer", "tc-transfer-tool", "tc-end-call-tool"],
  },
  {
    id: "suite-realtime-voice",
    name: "实时语音交互",
-    assistantIds: ["*"],
    caseIds: ["tc-hello", "tc-noise", "tc-barge-in", "tc-long-silence"],
  },
 ];
@@ -347,15 +335,6 @@ function findSuite(id: string | null): DebugTestSuite | null {
  return MOCK_TEST_SUITES.find((item) => item.id === id) ?? null;
 }

-/** 只返回当前助手关联的测试集 */
-function suitesForAssistant(assistantId: string | null): DebugTestSuite[] {
-  return MOCK_TEST_SUITES.filter((suite) => {
-    if (suite.assistantIds.includes("*")) return true;
-    if (!assistantId) return false;
-    return suite.assistantIds.includes(assistantId);
-  });
-}
-
 function casesInSuites(suites: DebugTestSuite[]): DebugTestCase[] {
  const seen = new Set<string>();
  const cases: DebugTestCase[] = [];
@@ -657,23 +636,18 @@ function TestCasePickerPopover({
  onOpenChange,
  selectedId,
  onSelect,
-  assistantId,
  trigger,
 }: {
  open: boolean;
  onOpenChange: (open: boolean) => void;
  selectedId: string | null;
  onSelect: (id: string) => void;
-  assistantId: string | null;
  trigger: React.ReactNode;
 }) {
  const [query, setQuery] = useState("");
  const [activeSuiteId, setActiveSuiteId] = useState<string | null>(null);

-  const suites = useMemo(
-    () => suitesForAssistant(assistantId),
-    [assistantId],
-  );
+  const suites = MOCK_TEST_SUITES;
  const scopedCases = useMemo(() => casesInSuites(suites), [suites]);
  const activeSuite = findSuite(activeSuiteId);

@@ -810,7 +784,7 @@ function TestCasePickerPopover({
                    </div>
                    {suites.length === 0 ? (
                      <div className="rounded-xl border border-dashed border-hairline-strong px-3 py-6 text-center text-xs text-muted-foreground">
-                        当前助手还没有关联测试集
+                        暂无测试集
                      </div>
                    ) : (
                      suites.map((suite) => (
@@ -851,7 +825,6 @@ function TestCasePickerPopover({

 export function AutoTestRunBar({
  state,
-  assistantId,
  onSelectCase,
  onClearCase,
  onStart,
@@ -859,7 +832,6 @@ export function AutoTestRunBar({
  onRerun,
 }: {
  state: AutoTestRunState;
-  assistantId: string | null;
  onSelectCase: (id: string) => void;
  onClearCase: () => void;
  onStart: () => void;
@@ -877,7 +849,6 @@ export function AutoTestRunBar({
          onOpenChange={setPickerOpen}
          selectedId={state.selectedId}
          onSelect={onSelectCase}
-          assistantId={assistantId}
          trigger={
            <button
              type="button"
@@ -985,7 +956,6 @@ export function AutoTestRunBar({
          onOpenChange={setPickerOpen}
          selectedId={state.selectedId}
          onSelect={onSelectCase}
-          assistantId={assistantId}
          trigger={
            <button
              type="button"
--- a/frontend/src/components/assistant-editor/debug-preview.tsx
+++ b/frontend/src/components/assistant-editor/debug-preview.tsx
@@ -334,7 +334,7 @@ export function DebugDrawer({
        )}
      </div>
      {debugMode === "auto" ? (
-        <AutoTestPanel assistantId={assistantId} autoTest={autoTest} />
+        <AutoTestPanel autoTest={autoTest} />
      ) : (
        <DebugVoicePanel
          view={view}
@@ -355,10 +355,8 @@ export function DebugDrawer({
 }

 function AutoTestPanel({
-  assistantId,
  autoTest,
 }: {
-  assistantId: string | null;
  autoTest: ReturnType<typeof useAutoTestRunner>;
 }) {
  const {
@@ -379,7 +377,6 @@ function AutoTestPanel({
      <div className="shrink-0 border-t border-hairline bg-card p-3">
        <AutoTestRunBar
          state={state}
-          assistantId={assistantId}
          onSelectCase={selectCase}
          onClearCase={clearCase}
          onStart={start}
--- a/frontend/src/components/batch-test/batch-case-detail-drawer.tsx
+++ b/frontend/src/components/batch-test/batch-case-detail-drawer.tsx
@@ -3,7 +3,6 @@
 /**
 * 批量测试用例详情面板。
 * 与 prompt mode DebugDrawer overlay 相同：作为右侧 flex 半屏，无遮罩/模糊。
- * 当前展示前端 mock 数据，后续可直接替换为真实执行事件与校验结果。
 */

 import {
@@ -79,7 +78,7 @@ export function BatchCaseDetailDrawer({
            {item.name}
          </h2>
          <p className="truncate text-xs text-muted-soft">
-            被测助手：{assistantName} · Mock 结果
+            被测助手：{assistantName} · Pipeline 结果
          </p>
        </div>
        <BatchCaseStatusBadge status={item.status} />
@@ -99,12 +98,7 @@ export function BatchCaseDetailDrawer({
              <section className="rounded-2xl border border-hairline bg-canvas-soft px-4 py-3.5">
                <p className="text-xs font-medium text-foreground">执行契约</p>
                <p className="mt-1 text-xs leading-5 text-muted-soft">
-                  {item.turns.length} 轮固定输入 ·{" "}
-                  {item.turns.reduce(
-                    (sum, turn) => sum + turn.evaluations.length,
-                    0,
-                  ) + item.overallCriteria.length}{" "}
-                  项评估标准 · 全部通过才算通过
+                  正在执行固定文字脚本；完成后将展示每一轮回复、工具调用和全部评估标准。
                </p>
              </section>
              <ExecutionTimeline status={item.status} />
@@ -288,7 +282,7 @@ function ToolCallCard({ toolCall }: { toolCall: BatchToolCallRecord }) {
      </div>
      <div className="mt-2 grid gap-2 sm:grid-cols-2">
        <CodeResult label="参数" value={toolCall.argumentsJson} />
-        <CodeResult label="Mock 返回" value={toolCall.resultJson} />
+        <CodeResult label="工具返回" value={toolCall.resultJson} />
      </div>
    </div>
  );
@@ -401,7 +395,7 @@ function ExecutionTimeline({ status }: { status: "waiting" | "running" }) {
    },
    {
      label: "请求助手响应",
-      detail: isRunning ? "生成中（Mock）" : "尚未开始",
+      detail: isRunning ? "Pipeline 生成中" : "尚未开始",
      state: isRunning ? "active" : "pending",
    },
    {
--- a/frontend/src/components/batch-test/batch-run-completed.tsx
+++ b/frontend/src/components/batch-test/batch-run-completed.tsx
@@ -1,16 +1,13 @@
 "use client";

-/**
- * 批量测试 — 已完成结果视图（MVP mock）。
- */
+/** 批量测试已完成结果视图。 */

-import { ChevronRight } from "lucide-react";
+import { ChevronDown, ChevronRight, Settings2 } from "lucide-react";
 import { useRef, useState } from "react";

 import { BatchCaseDetailDrawer } from "@/components/batch-test/batch-case-detail-drawer";
 import {
  BatchCaseStatusBadge,
-  BatchPhasePill,
  getBatchCaseStatusLabel,
 } from "@/components/batch-test/batch-case-status";
 import {
@@ -50,13 +47,8 @@ export function BatchRunCompletedView({
  const detailOpen = Boolean(selectedCase);

  const counts = countByStatus(run.cases);
-  const judged = counts.pass + counts.fail + counts.error;
  const total = run.cases.length;
-  const passRate = judged === 0 ? 0 : Math.round((counts.pass / judged) * 100);
-  const failRate =
-    judged === 0 ? 0 : Math.round((counts.fail / judged) * 100);
-  const errorRate =
-    judged === 0 ? 0 : Math.round((counts.error / judged) * 100);
+  const outcome = getRunOutcome(run, counts);
  const visibleCases =
    filter === "fail"
      ? run.cases.filter((item) => item.status === "fail")
@@ -93,26 +85,96 @@ export function BatchRunCompletedView({
        >
          {/* 结果总览 */}
          <section className="rounded-2xl border border-hairline bg-card p-5 shadow-sm">
-            <div className="flex flex-wrap items-center gap-2">
-              <h2 className="text-base font-medium text-foreground">
-                {run.title}
-              </h2>
-              <BatchPhasePill phase="completed" />
+            <div className="flex flex-wrap items-start justify-between gap-3">
+              <div>
+                <h2 className="text-base font-medium text-foreground">
+                  {run.title}
+                </h2>
+                {run.finishedAt && (
+                  <p className="mt-1.5 text-xs text-muted-soft">
+                    完成于 {formatRunTime(run.finishedAt)}
+                    {run.stopReason ? ` · ${stopReasonLabel(run)}` : ""}
+                  </p>
+                )}
+              </div>
+              <span
+                className={cn(
+                  "inline-flex shrink-0 items-center rounded-full px-3 py-1 text-xs font-medium",
+                  outcome.className,
+                )}
+              >
+                {outcome.label}
+              </span>
            </div>
-            {run.finishedAt && (
-              <p className="mt-1.5 text-xs text-muted-soft">
-                完成时间：{formatRunTime(run.finishedAt)}
-                {run.stopReason ? ` · ${stopReasonLabel(run)}` : ""}
-              </p>
-            )}

-            <dl className="mt-4 grid gap-3 rounded-xl border border-hairline bg-canvas-soft/60 px-4 py-3 text-sm sm:grid-cols-2 lg:grid-cols-4">
+            <div className="mt-5 rounded-xl border border-hairline bg-canvas-soft/60 p-4">
+              <div className="flex flex-col gap-4 sm:flex-row sm:items-end sm:justify-between">
+                <div>
+                  <p className="text-xs font-medium tracking-wide text-muted-soft uppercase">
+                    测试结果
+                  </p>
+                  <p className="mt-1 font-display text-3xl text-ink">
+                    <span className="tabular-nums">{counts.pass}</span>
+                    <span className="mx-1.5 text-muted-soft">/</span>
+                    <span className="tabular-nums">{total}</span>
+                    <span className="ml-2 text-2xl">通过</span>
+                  </p>
+                </div>
+
+                <div className="grid grid-cols-3 gap-x-5 gap-y-2 text-sm sm:shrink-0">
+                  <ResultCount tone="success" label="通过" count={counts.pass} />
+                  <ResultCount
+                    tone="destructive"
+                    label="断言失败"
+                    count={counts.fail}
+                  />
+                  <ResultCount
+                    tone="warning"
+                    label="执行错误"
+                    count={counts.error}
+                  />
+                  {counts.skipped > 0 && (
+                    <ResultCount
+                      tone="muted"
+                      label="未执行"
+                      count={counts.skipped}
+                    />
+                  )}
+                </div>
+              </div>
+
+              <ResultDistribution counts={counts} total={total} />
+            </div>
+
+            <dl
+              className={cn(
+                "mt-4 grid gap-x-6 gap-y-4 text-sm sm:grid-cols-2",
+                !detailOpen && "lg:grid-cols-4",
+              )}
+            >
              <div>
                <dt className="text-xs text-muted-foreground">被测助手</dt>
-                <dd className="mt-1 truncate font-medium text-foreground">
+                <dd className="mt-1 break-words font-medium text-foreground">
                  {run.assistantName}
                </dd>
              </div>
+              <div className="min-w-0">
+                <dt className="text-xs text-muted-foreground">评估模型</dt>
+                <dd
+                  className="mt-1 break-words font-medium text-foreground"
+                  title={run.config.evaluatorModelResourceName}
+                >
+                  {run.config.evaluatorModelResourceName}
+                </dd>
+                {run.config.evaluatorModel && (
+                  <dd
+                    className="mt-0.5 break-all font-mono text-[11px] text-muted-soft"
+                    title={run.config.evaluatorModel}
+                  >
+                    {run.config.evaluatorModel}
+                  </dd>
+                )}
+              </div>
              <div>
                <dt className="text-xs text-muted-foreground">测试范围</dt>
                <dd className="mt-1 font-medium text-foreground">
@@ -122,70 +184,34 @@ export function BatchRunCompletedView({
              <div>
                <dt className="text-xs text-muted-foreground">运行参数</dt>
                <dd className="mt-1 font-medium text-foreground">
-                  {run.config.concurrency} 并发 · {run.config.timeoutSecs} 秒
-                </dd>
-              </div>
-              <div>
-                <dt className="text-xs text-muted-foreground">耗时 / 处理策略</dt>
-                <dd className="mt-1 text-xs font-medium leading-5 text-foreground">
-                  {formatRunDuration(run.startedAt, run.finishedAt)} · 断言失败：
-                  {BATCH_FAILURE_STRATEGY_LABEL[run.config.failureStrategy]}
-                  <br />
-                  执行错误：重试 {run.config.errorRetryCount} 次，
-                  {BATCH_ERROR_STRATEGY_LABEL[run.config.errorStrategy]}
+                  {run.config.concurrency} 并发 · {run.config.timeoutSecs} 秒超时
+                  <span className="mx-1.5 text-muted-soft">·</span>
+                  {formatRunDuration(run.startedAt, run.finishedAt)}
                </dd>
              </div>
            </dl>

-            <div className="mt-5 flex flex-col items-center gap-6 sm:flex-row sm:justify-between sm:gap-8">
-              <div className="text-center sm:text-left">
-                <div className="font-display text-3xl text-ink">
-                  {counts.pass} / {judged || total} 通过
-                </div>
-                <div className="mt-1 text-sm font-medium text-success">
-                  通过率 {passRate}%
-                </div>
+            <details className="group mt-4 border-t border-hairline pt-3">
+              <summary className="flex w-fit cursor-pointer list-none items-center gap-2 rounded-full px-2 py-1 text-xs font-medium text-muted-foreground transition-colors hover:bg-canvas-soft hover:text-foreground focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring/40 [&::-webkit-details-marker]:hidden">
+                <Settings2 size={13} />
+                查看运行设置
+                <ChevronDown
+                  size={13}
+                  className="transition-transform group-open:rotate-180"
+                />
+              </summary>
+              <div className="mt-3 grid gap-3 rounded-xl border border-hairline bg-canvas-soft/50 px-4 py-3 text-xs leading-5 text-muted-foreground sm:grid-cols-2">
+                <p>
+                  <span className="font-medium text-foreground">断言失败：</span>
+                  {BATCH_FAILURE_STRATEGY_LABEL[run.config.failureStrategy]}
+                </p>
+                <p>
+                  <span className="font-medium text-foreground">执行错误：</span>
+                  重试 {run.config.errorRetryCount} 次，
+                  {BATCH_ERROR_STRATEGY_LABEL[run.config.errorStrategy]}
+                </p>
              </div>
-
-              <PassRateDonut
-                pass={counts.pass}
-                fail={counts.fail}
-                error={counts.error}
-              />
-
-              <div className="space-y-2 text-sm">
-                <LegendRow
-                  tone="success"
-                  label="通过"
-                  count={counts.pass}
-                  percent={passRate}
-                />
-                <LegendRow
-                  tone="destructive"
-                  label="断言失败"
-                  count={counts.fail}
-                  percent={failRate}
-                />
-                <LegendRow
-                  tone="warning"
-                  label="执行错误"
-                  count={counts.error}
-                  percent={errorRate}
-                />
-                {counts.skipped > 0 && (
-                  <LegendRow
-                    tone="muted"
-                    label="未执行"
-                    count={counts.skipped}
-                    percent={
-                      total === 0
-                        ? 0
-                        : Math.round((counts.skipped / total) * 100)
-                    }
-                  />
-                )}
-              </div>
-            </div>
+            </details>
          </section>

          {/* 用例列表 */}
@@ -319,106 +345,53 @@ export function BatchRunCompletedView({
  );
 }

-function PassRateDonut({
-  pass,
-  fail,
-  error,
-}: {
-  pass: number;
-  fail: number;
-  error: number;
-}) {
-  const total = pass + fail + error;
-  const passRatio = total === 0 ? 0 : pass / total;
-  const size = 112;
-  const stroke = 14;
-  const radius = (size - stroke) / 2;
-  const circumference = 2 * Math.PI * radius;
-  const passLength = circumference * passRatio;
-  const failLength = circumference * (total === 0 ? 0 : fail / total);
-  const errorLength = circumference * (total === 0 ? 0 : error / total);
+type BatchCounts = ReturnType<typeof countByStatus>;

-  return (
-    <div
-      className="relative shrink-0"
-      style={{ width: size, height: size }}
-      aria-hidden
-    >
-      <svg width={size} height={size} className="-rotate-90">
-        <circle
-          cx={size / 2}
-          cy={size / 2}
-          r={radius}
-          fill="none"
-          stroke="currentColor"
-          strokeWidth={stroke}
-          className="text-surface-strong"
-        />
-        {total > 0 && (
-          <>
-            {pass > 0 && (
-              <circle
-                cx={size / 2}
-                cy={size / 2}
-                r={radius}
-                fill="none"
-                stroke="currentColor"
-                strokeWidth={stroke}
-                strokeDasharray={`${passLength} ${circumference}`}
-                strokeLinecap="butt"
-                className="text-success"
-              />
-            )}
-            {fail > 0 && (
-              <circle
-                cx={size / 2}
-                cy={size / 2}
-                r={radius}
-                fill="none"
-                stroke="currentColor"
-                strokeWidth={stroke}
-                strokeDasharray={`${failLength} ${circumference}`}
-                strokeDashoffset={-passLength}
-                strokeLinecap="butt"
-                className="text-destructive"
-              />
-            )}
-            {error > 0 && (
-              <circle
-                cx={size / 2}
-                cy={size / 2}
-                r={radius}
-                fill="none"
-                stroke="currentColor"
-                strokeWidth={stroke}
-                strokeDasharray={`${errorLength} ${circumference}`}
-                strokeDashoffset={-(passLength + failLength)}
-                strokeLinecap="butt"
-                className="text-amber-500"
-              />
-            )}
-          </>
-        )}
-      </svg>
-      <div className="absolute inset-0 flex flex-col items-center justify-center">
-        <span className="text-lg font-medium tabular-nums text-foreground">
-          {total === 0 ? 0 : Math.round((pass / total) * 100)}%
-        </span>
-      </div>
-    </div>
-  );
+function getRunOutcome(run: BatchRunSnapshot, counts: BatchCounts) {
+  if (run.status === "cancelled") {
+    return {
+      label: "运行已停止",
+      className: "bg-surface-strong text-muted-foreground",
+    };
+  }
+  if (counts.error > 0) {
+    return {
+      label: "存在执行错误",
+      className: "bg-amber-500/10 text-amber-600 dark:text-amber-400",
+    };
+  }
+  if (counts.fail > 0) {
+    return {
+      label: "存在断言失败",
+      className: "bg-destructive/10 text-destructive",
+    };
+  }
+  if (counts.skipped > 0) {
+    return {
+      label: "部分未执行",
+      className: "bg-surface-strong text-muted-foreground",
+    };
+  }
+  if (counts.pass > 0) {
+    return {
+      label: "全部通过",
+      className: "bg-success/10 text-success",
+    };
+  }
+  return {
+    label: "暂无结果",
+    className: "bg-surface-strong text-muted-foreground",
+  };
 }

-function LegendRow({
+function ResultCount({
  tone,
  label,
  count,
-  percent,
 }: {
  tone: "success" | "destructive" | "warning" | "muted";
  label: string;
  count: number;
-  percent: number;
 }) {
  const dotClass = {
    success: "bg-success",
@@ -428,11 +401,48 @@ function LegendRow({
  }[tone];

  return (
-    <div className="flex items-center gap-2 text-muted-foreground">
-      <span className={cn("size-2.5 rounded-full", dotClass)} />
-      <span>
-        {label} {count}（{percent}%）
+    <div className="flex min-w-0 items-center gap-2">
+      <span
+        className={cn("size-2 shrink-0 rounded-full", dotClass)}
+        aria-hidden
+      />
+      <span className="truncate text-muted-foreground">{label}</span>
+      <span className="ml-auto font-medium tabular-nums text-foreground">
+        {count}
      </span>
    </div>
  );
 }
+
+function ResultDistribution({
+  counts,
+  total,
+}: {
+  counts: BatchCounts;
+  total: number;
+}) {
+  const segments = [
+    { key: "pass", count: counts.pass, className: "bg-success" },
+    { key: "fail", count: counts.fail, className: "bg-destructive" },
+    { key: "error", count: counts.error, className: "bg-amber-500" },
+    { key: "skipped", count: counts.skipped, className: "bg-muted-soft" },
+  ].filter((item) => item.count > 0);
+
+  return (
+    <div
+      className="mt-4 flex h-2 overflow-hidden rounded-full bg-surface-strong"
+      role="img"
+      aria-label={`结果分布：通过 ${counts.pass}，断言失败 ${counts.fail}，执行错误 ${counts.error}，未执行 ${counts.skipped}`}
+    >
+      {total > 0 &&
+        segments.map((segment) => (
+          <span
+            key={segment.key}
+            className={segment.className}
+            style={{ flexBasis: 0, flexGrow: segment.count }}
+            aria-hidden
+          />
+        ))}
+    </div>
+  );
+}
--- a/frontend/src/components/batch-test/batch-run-running.tsx
+++ b/frontend/src/components/batch-test/batch-run-running.tsx
@@ -1,8 +1,6 @@
 "use client";

-/**
- * 批量测试 — 运行中视图（MVP mock）。
- */
+/** 批量测试运行中的持久化快照视图。 */

 import { ChevronRight } from "lucide-react";
 import { useRef, useState } from "react";
@@ -67,6 +65,10 @@ export function BatchRunRunningView({ run }: { run: BatchRunSnapshot }) {
                  </h2>
                  <BatchPhasePill phase="running" />
                </div>
+                <p className="text-xs text-muted-soft">
+                  被测助手：{run.assistantName} · 评估模型：
+                  {run.config.evaluatorModelResourceName}
+                </p>

                <div className="flex items-center gap-3">
                  <div className="h-2.5 min-w-0 flex-1 overflow-hidden rounded-full bg-surface-strong">
--- a/frontend/src/components/pages/BatchTestPage.tsx
+++ b/frontend/src/components/pages/BatchTestPage.tsx
@@ -2,7 +2,7 @@

 /**
 * 批量测试页（MVP）
- * 三阶段：配置 → 运行中 → 已完成。执行进度为前端 mock，后续可换真实引擎。
+ * 三阶段：配置 → 运行中 → 已完成。运行快照由后端持久化并轮询更新。
 */

 import {
@@ -35,21 +35,21 @@ import {
 import {
  BATCH_ERROR_STRATEGY_LABEL,
  BATCH_FAILURE_STRATEGY_LABEL,
-  createMockBatchRun,
-  hasEvaluationFailure,
  type BatchErrorStrategy,
  type BatchFailureStrategy,
-  type BatchMockExecutionPlan,
  type BatchRunPhase,
  type BatchRunSnapshot,
 } from "@/data/batch-run";
+import type { TestCase, TestSuite } from "@/data/test-suites";
 import {
-  listTestCases,
-  listTestSuites,
-  type TestCase,
-  type TestSuite,
-} from "@/data/test-suites";
-import { assistantsApi, type Assistant } from "@/lib/api";
+  assistantsApi,
+  batchRunsApi,
+  modelResourcesApi,
+  testCasesApi,
+  testSuitesApi,
+  type Assistant,
+  type ModelResource,
+} from "@/lib/api";

 const CONCURRENCY_OPTIONS = ["1", "2", "3", "5", "10"] as const;

@@ -82,7 +82,7 @@ const BATCH_SECTIONS = [
  { id: "settings", label: "运行设置" },
 ] as const;

-const TICK_MS = 900;
+const RUN_POLL_MS = 1000;

 function getAppScrollContainer(): HTMLElement | null {
  return document.querySelector<HTMLElement>(".app-content");
@@ -104,29 +104,38 @@ function buildRunTitle(
  return "批量测试";
 }

-function loadTestScope() {
-  const suites = listTestSuites();
+function groupCasesBySuite(suites: TestSuite[], cases: TestCase[]) {
  const casesBySuite: Record<string, TestCase[]> = {};

  for (const suite of suites) {
-    const items = listTestCases(suite.id);
-    casesBySuite[suite.id] = items;
+    casesBySuite[suite.id] = cases
+      .filter((item) => item.suiteId === suite.id)
+      .sort((left, right) => left.sortOrder - right.sortOrder);
  }
-
-  return { suites, casesBySuite };
+  return casesBySuite;
 }

 export function BatchTestPage() {
  const router = useRouter();
  const [phase, setPhase] = useState<BatchRunPhase>("config");
  const [run, setRun] = useState<BatchRunSnapshot | null>(null);
-  const mockExecutionPlanRef = useRef<BatchMockExecutionPlan>({});
+  const [runError, setRunError] = useState("");
+  const [starting, setStarting] = useState(false);
+  const [stopping, setStopping] = useState(false);

  const [assistants, setAssistants] = useState<Assistant[]>([]);
  const [loadingAssistants, setLoadingAssistants] = useState(true);
  const [assistantId, setAssistantId] = useState("");
+  const [evaluatorModels, setEvaluatorModels] = useState<ModelResource[]>([]);
+  const [loadingEvaluatorModels, setLoadingEvaluatorModels] = useState(true);
+  const [evaluatorModelResourceId, setEvaluatorModelResourceId] = useState("");

-  const [{ suites, casesBySuite }] = useState(loadTestScope);
+  const [suites, setSuites] = useState<TestSuite[]>([]);
+  const [casesBySuite, setCasesBySuite] = useState<Record<string, TestCase[]>>(
+    {},
+  );
+  const [loadingScope, setLoadingScope] = useState(true);
+  const [scopeError, setScopeError] = useState("");
  const [selectedCaseIds, setSelectedCaseIds] = useState<Set<string>>(
    () => new Set(),
  );
@@ -249,102 +258,84 @@ export function BatchTestPage() {
    })();
  }, []);

-  // mock 执行引擎：按并发推进用例状态
  useEffect(() => {
-    if (phase !== "running" || !run?.startedAt) return;
-
-    function tick(settleRunning: boolean) {
-      setRun((prev) => {
-        if (!prev || prev.finishedAt) return prev;
-
-        let cases = prev.cases.map((item) => ({ ...item }));
-
-        // 1) 结算上一拍仍在运行的用例（首拍只启动，不结算）
-        let sawFail = false;
-        let sawError = false;
-        if (settleRunning) {
-          for (const item of cases) {
-            if (item.status !== "running") continue;
-            const plannedError = mockExecutionPlanRef.current[item.id];
-            if (plannedError) {
-              if (
-                plannedError.retryable &&
-                item.attemptCount < item.maxAttempts
-              ) {
-                item.status = "waiting";
-              } else {
-                item.status = "error";
-                item.executionError = plannedError;
-                sawError = true;
-              }
-              continue;
-            }
-            item.status = hasEvaluationFailure(item) ? "fail" : "pass";
-            if (item.status === "fail") sawFail = true;
-          }
-        }
-
-        const stopReason =
-          sawError && prev.config.errorStrategy === "stop_on_error"
-            ? "execution_error"
-            : sawFail && prev.config.failureStrategy === "stop_on_fail"
-              ? "assertion_failure"
-              : null;
-
-        // 配置为停止时，剩余等待/运行中的用例统一标记为未执行。
-        if (stopReason) {
-          cases = cases.map((item) =>
-            item.status === "waiting" || item.status === "running"
-              ? { ...item, status: "skipped" as const }
-              : item,
-          );
-          return {
-            ...prev,
-            cases,
-            finishedAt: new Date().toISOString(),
-            stopReason,
-          };
-        }
-
-        // 2) 按并发补齐运行中
-        const limit = prev.config.concurrency;
-        let running = cases.filter((item) => item.status === "running").length;
-        for (const item of cases) {
-          if (running >= limit) break;
-          if (item.status !== "waiting") continue;
-          item.status = "running";
-          item.attemptCount += 1;
-          running += 1;
-        }
-
-        const pending = cases.some(
-          (item) => item.status === "waiting" || item.status === "running",
+    void (async () => {
+      try {
+        const resources = (await modelResourcesApi.list()).filter(
+          (item) => item.capability === "LLM" && item.enabled,
        );
-        if (!pending) {
-          return {
-            ...prev,
-            cases,
-            finishedAt: new Date().toISOString(),
-            stopReason: null,
-          };
-        }
+        setEvaluatorModels(resources);
+        const defaultResource =
+          resources.find((item) => item.isDefault) ?? resources[0];
+        if (defaultResource) setEvaluatorModelResourceId(defaultResource.id);
+      } catch {
+        // 模型资源为空或加载失败时，运行前检查会阻止启动。
+      } finally {
+        setLoadingEvaluatorModels(false);
+      }
+    })();
+  }, []);

-        return { ...prev, cases };
-      });
+  useEffect(() => {
+    void (async () => {
+      try {
+        setScopeError("");
+        const [loadedSuites, loadedCases] = await Promise.all([
+          testSuitesApi.list(),
+          testCasesApi.list(),
+        ]);
+        setSuites(loadedSuites);
+        setCasesBySuite(groupCasesBySuite(loadedSuites, loadedCases));
+      } catch (error) {
+        setScopeError(
+          error instanceof Error ? error.message : "加载测试范围失败",
+        );
+      } finally {
+        setLoadingScope(false);
+      }
+    })();
+  }, []);
+
+  useEffect(() => {
+    if (phase !== "running" || !run?.id) return;
+    let disposed = false;
+    let timer = 0;
+
+    async function poll() {
+      let finished = false;
+      try {
+        const snapshot = await batchRunsApi.get(run!.id);
+        if (disposed) return;
+        setRun(snapshot);
+        setRunError("");
+        if (
+          snapshot.finishedAt ||
+          snapshot.status === "completed" ||
+          snapshot.status === "cancelled"
+        ) {
+          finished = true;
+          setStopping(false);
+          setPhase("completed");
+        }
+      } catch (error) {
+        if (!disposed) {
+          setRunError(
+            error instanceof Error ? error.message : "刷新运行结果失败",
+          );
+        }
+      } finally {
+        if (!disposed && !finished) {
+          timer = window.setTimeout(() => void poll(), RUN_POLL_MS);
+        }
+      }
    }

-    // 立刻拉起第一批，再按节拍结算/推进
-    tick(false);
-    const timer = window.setInterval(() => tick(true), TICK_MS);
-    return () => window.clearInterval(timer);
-  }, [phase, run?.startedAt]);
-
-  // 执行引擎只负责写入完成快照；页面阶段在快照稳定后统一切换。
-  useEffect(() => {
-    if (phase !== "running" || !run?.finishedAt) return;
-    const timer = window.setTimeout(() => setPhase("completed"), 0);
-    return () => window.clearTimeout(timer);
-  }, [phase, run?.finishedAt]);
+    void poll();
+    return () => {
+      disposed = true;
+      window.clearTimeout(timer);
+    };
+  }, [phase, run?.id]);

  const selectedCases = useMemo(() => {
    const ordered: TestCase[] = [];
@@ -364,9 +355,13 @@ export function BatchTestPage() {

  const canStart =
    Boolean(assistantId) &&
+    Boolean(evaluatorModelResourceId) &&
    selectedCount > 0 &&
    timeoutValid &&
    !loadingAssistants &&
+    !loadingEvaluatorModels &&
+    !loadingScope &&
+    !starting &&
    phase === "config";

  const runReadiness: {
@@ -375,60 +370,90 @@ export function BatchTestPage() {
    actionable?: boolean;
  } = loadingAssistants
    ? { label: "正在加载助手…" }
-    : assistants.length === 0
-      ? { label: "暂无可用助手", target: "target", actionable: true }
-      : !assistantId
-        ? { label: "请选择被测助手", target: "target", actionable: true }
-        : selectedCount === 0
-          ? { label: "请选择测试用例", target: "scope", actionable: true }
-          : !timeoutValid
-            ? {
-                label: "请修正超时时间",
-                target: "settings",
-                actionable: true,
-              }
-            : {
-                label: `${selectedCount} 个用例 · ${concurrency} 并发 · 超时 ${timeoutValue} 秒`,
-              };
+    : loadingEvaluatorModels
+      ? { label: "正在加载评估模型…" }
+      : loadingScope
+        ? { label: "正在加载测试范围…" }
+        : scopeError
+          ? { label: "测试范围加载失败", target: "scope", actionable: true }
+          : assistants.length === 0
+            ? { label: "暂无可用助手", target: "target", actionable: true }
+            : !assistantId
+              ? { label: "请选择被测助手", target: "target", actionable: true }
+              : evaluatorModels.length === 0
+                ? { label: "暂无可用评估模型", target: "target", actionable: true }
+                : !evaluatorModelResourceId
+                  ? { label: "请选择评估模型", target: "target", actionable: true }
+                  : selectedCount === 0
+                    ? { label: "请选择测试用例", target: "scope", actionable: true }
+                    : !timeoutValid
+                      ? {
+                          label: "请修正超时时间",
+                          target: "settings",
+                          actionable: true,
+                        }
+                      : {
+                          label: `${selectedCount} 个用例 · ${concurrency} 并发 · 超时 ${timeoutValue} 秒`,
+                        };

-  function startRun(cases: TestCase[]) {
-    const assistantName =
-      assistants.find((item) => item.id === assistantId)?.name ?? "助手";
-    const { snapshot, executionPlan } = createMockBatchRun({
-      title: buildRunTitle(cases, suites),
-      assistantName,
-      cases,
-      concurrency: Number(concurrency) || 3,
-      timeoutSecs: timeoutValue,
-      failureStrategy: failStrategy,
-      errorRetryCount: errorRetryValue,
-      errorStrategy,
-    });
-    mockExecutionPlanRef.current = executionPlan;
-    setRun(snapshot);
-    setPhase("running");
+  async function startRun(cases: TestCase[]) {
+    if (
+      !assistantId ||
+      !evaluatorModelResourceId ||
+      cases.length === 0 ||
+      starting
+    ) {
+      return;
+    }
+    setStarting(true);
+    setRunError("");
+    try {
+      const snapshot = await batchRunsApi.create({
+        assistantId,
+        evaluatorModelResourceId,
+        caseIds: cases.map((item) => item.id),
+        title: buildRunTitle(cases, suites),
+        config: {
+          concurrency: Number(concurrency) || 3,
+          timeoutSecs: timeoutValue,
+          failureStrategy: failStrategy,
+          errorRetryCount: errorRetryValue,
+          errorStrategy,
+        },
+      });
+      setRun(snapshot);
+      setPhase("running");
+    } catch (error) {
+      setRunError(error instanceof Error ? error.message : "启动批量测试失败");
+    } finally {
+      setStarting(false);
+    }
  }

  function handleStart() {
    if (!canStart) return;
-    startRun(selectedCases);
+    void startRun(selectedCases);
  }

-  function handleStop() {
-    setRun((prev) => {
-      if (!prev) return prev;
-      return {
-        ...prev,
-        stopReason: "manual",
-        finishedAt: new Date().toISOString(),
-        cases: prev.cases.map((item) =>
-          item.status === "waiting" || item.status === "running"
-            ? { ...item, status: "skipped" }
-            : item,
-        ),
-      };
-    });
-    setPhase("completed");
+  async function handleStop() {
+    if (!run || stopping) return;
+    setStopping(true);
+    setRunError("");
+    try {
+      const snapshot = await batchRunsApi.cancel(run.id);
+      setRun(snapshot);
+      if (
+        snapshot.finishedAt ||
+        snapshot.status === "completed" ||
+        snapshot.status === "cancelled"
+      ) {
+        setPhase("completed");
+        setStopping(false);
+      }
+    } catch (error) {
+      setRunError(error instanceof Error ? error.message : "停止批量测试失败");
+      setStopping(false);
+    }
  }

  function sourceCasesByIds(ids: string[]): TestCase[] {
@@ -445,7 +470,7 @@ export function BatchTestPage() {
      setRun(null);
      return;
    }
-    startRun(cases);
+    void startRun(cases);
  }

  function handleRerunFailed() {
@@ -457,12 +482,12 @@ export function BatchTestPage() {
        )
        .map((item) => item.id),
    );
-    if (failedCases.length > 0) startRun(failedCases);
+    if (failedCases.length > 0) void startRun(failedCases);
  }

  function handleRerunCase(caseId: string) {
    const [item] = sourceCasesByIds([caseId]);
-    if (item) startRun([item]);
+    if (item) void startRun([item]);
  }

  function handleEditCase(caseId: string) {
@@ -474,6 +499,7 @@ export function BatchTestPage() {
  function handleBackToConfig() {
    setPhase("config");
    setRun(null);
+    setRunError("");
  }

  if (phase === "running" && run) {
@@ -485,13 +511,26 @@ export function BatchTestPage() {
          <Button
            variant="outline"
            className="gap-2 rounded-full border-destructive/40 text-destructive hover:bg-destructive/10 hover:text-destructive"
-            onClick={handleStop}
+            disabled={stopping}
+            onClick={() => void handleStop()}
          >
-            <Square size={14} className="fill-current" />
-            停止运行
+            {stopping ? (
+              <Loader2 size={14} className="animate-spin" />
+            ) : (
+              <Square size={14} className="fill-current" />
+            )}
+            {stopping ? "正在停止…" : "停止运行"}
          </Button>
        }
      >
+        {runError && (
+          <p
+            className="border-b border-destructive/20 bg-destructive/5 px-5 py-2 text-sm text-destructive"
+            role="alert"
+          >
+            {runError}
+          </p>
+        )}
        <BatchRunRunningView run={run} />
      </ListPageLayout>
    );
@@ -517,7 +556,7 @@ export function BatchTestPage() {
            <Button
              variant="outline"
              className="gap-2 rounded-full border-hairline-strong"
-              disabled={failedCount === 0}
+              disabled={failedCount === 0 || starting}
              onClick={handleRerunFailed}
            >
              <RotateCcw size={15} />
@@ -525,14 +564,27 @@ export function BatchTestPage() {
            </Button>
            <Button
              className="gap-2 rounded-full px-4"
+              disabled={starting}
              onClick={handleRerunAll}
            >
-              <RotateCcw size={15} />
-              重跑全部
+              {starting ? (
+                <Loader2 size={15} className="animate-spin" />
+              ) : (
+                <RotateCcw size={15} />
+              )}
+              {starting ? "正在启动…" : "重跑全部"}
            </Button>
          </div>
        }
      >
+        {runError && (
+          <p
+            className="border-b border-destructive/20 bg-destructive/5 px-5 py-2 text-sm text-destructive"
+            role="alert"
+          >
+            {runError}
+          </p>
+        )}
        <BatchRunCompletedView
          run={run}
          onEditCase={handleEditCase}
@@ -545,7 +597,7 @@ export function BatchTestPage() {
  return (
    <ListPageLayout
      title="批量测试"
-      description="配置一次批量运行：选择被测助手、测试范围与基础运行参数。"
+      description="配置一次批量运行：选择被测助手、评估模型、测试范围与运行参数。"
      className="max-w-[880px]"
      topbarAction={
        <div className="flex items-center gap-3">
@@ -578,8 +630,12 @@ export function BatchTestPage() {
            aria-describedby="batch-run-readiness"
            onClick={handleStart}
          >
-            <Play size={16} />
-            开始批量测试
+            {starting ? (
+              <Loader2 size={16} className="animate-spin" />
+            ) : (
+              <Play size={16} />
+            )}
+            {starting ? "正在启动…" : "开始批量测试"}
          </Button>
        </div>
      }
@@ -593,6 +649,15 @@ export function BatchTestPage() {
        }
      />

+      {runError && (
+        <p
+          className="rounded-xl border border-destructive/20 bg-destructive/5 px-3 py-2 text-sm text-destructive"
+          role="alert"
+        >
+          {runError}
+        </p>
+      )}
+
      <div className="space-y-4">
        <section
          ref={(element) => {
@@ -603,36 +668,77 @@ export function BatchTestPage() {
          <SectionCard
            icon={<Target size={15} />}
            title="运行目标"
-            description="选择本次批量测试要对齐的助手配置"
+            description="分别选择生成回复的助手和判断结果的 LLM"
          >
-            <label className="block space-y-2">
-              <span className="text-sm font-medium text-foreground">
-                被测助手
-              </span>
-              {loadingAssistants ? (
-                <div className="flex items-center gap-2 text-sm text-muted-foreground">
-                  <Loader2 size={16} className="animate-spin" />
-                  正在加载助手…
-                </div>
-              ) : assistants.length === 0 ? (
-                <p className="rounded-xl border border-dashed border-hairline-strong bg-canvas-soft/50 px-3 py-4 text-sm text-muted-foreground">
-                  暂无可用助手，请先在「创建助手」中新建。
-                </p>
-              ) : (
-                <Select value={assistantId} onValueChange={setAssistantId}>
-                  <SelectTrigger className="w-full border-hairline-strong bg-background">
-                    <SelectValue placeholder="选择被测助手" />
-                  </SelectTrigger>
-                  <SelectContent>
-                    {assistants.map((item) => (
-                      <SelectItem key={item.id} value={item.id}>
-                        {item.name}
-                      </SelectItem>
-                    ))}
-                  </SelectContent>
-                </Select>
-              )}
-            </label>
+            <div className="grid gap-5 sm:grid-cols-2">
+              <label className="block space-y-2">
+                <span className="text-sm font-medium text-foreground">
+                  被测助手
+                </span>
+                {loadingAssistants ? (
+                  <div className="flex items-center gap-2 text-sm text-muted-foreground">
+                    <Loader2 size={16} className="animate-spin" />
+                    正在加载助手…
+                  </div>
+                ) : assistants.length === 0 ? (
+                  <p className="rounded-xl border border-dashed border-hairline-strong bg-canvas-soft/50 px-3 py-4 text-sm text-muted-foreground">
+                    暂无可用助手，请先在「创建助手」中新建。
+                  </p>
+                ) : (
+                  <Select value={assistantId} onValueChange={setAssistantId}>
+                    <SelectTrigger className="w-full border-hairline-strong bg-background">
+                      <SelectValue placeholder="选择被测助手" />
+                    </SelectTrigger>
+                    <SelectContent>
+                      {assistants.map((item) => (
+                        <SelectItem key={item.id} value={item.id}>
+                          {item.name}
+                        </SelectItem>
+                      ))}
+                    </SelectContent>
+                  </Select>
+                )}
+              </label>
+
+              <label className="block space-y-2">
+                <span className="text-sm font-medium text-foreground">
+                  评估模型
+                </span>
+                {loadingEvaluatorModels ? (
+                  <div className="flex items-center gap-2 text-sm text-muted-foreground">
+                    <Loader2 size={16} className="animate-spin" />
+                    正在加载评估模型…
+                  </div>
+                ) : evaluatorModels.length === 0 ? (
+                  <p className="rounded-xl border border-dashed border-hairline-strong bg-canvas-soft/50 px-3 py-4 text-sm text-muted-foreground">
+                    暂无已启用的 LLM 模型资源，请先完成模型配置。
+                  </p>
+                ) : (
+                  <Select
+                    value={evaluatorModelResourceId}
+                    onValueChange={setEvaluatorModelResourceId}
+                  >
+                    <SelectTrigger className="w-full border-hairline-strong bg-background">
+                      <SelectValue placeholder="选择评估模型" />
+                    </SelectTrigger>
+                    <SelectContent>
+                      {evaluatorModels.map((item) => (
+                        <SelectItem key={item.id} value={item.id}>
+                          {item.name}
+                          {typeof item.values.modelId === "string" &&
+                          item.values.modelId
+                            ? ` · ${item.values.modelId}`
+                            : ""}
+                        </SelectItem>
+                      ))}
+                    </SelectContent>
+                  </Select>
+                )}
+                <span className="block text-xs leading-5 text-muted-soft">
+                  仅用于 LLM 语义断言，不参与助手回复生成。
+                </span>
+              </label>
+            </div>
          </SectionCard>
        </section>

@@ -647,12 +753,26 @@ export function BatchTestPage() {
            title="选择测试范围"
            description="搜索测试集或用例，按需组合本次运行范围"
          >
-            <TestScopeSelector
-              suites={suites}
-              casesBySuite={casesBySuite}
-              selectedCaseIds={selectedCaseIds}
-              onSelectionChange={setSelectedCaseIds}
-            />
+            {loadingScope ? (
+              <div className="flex items-center gap-2 py-6 text-sm text-muted-foreground">
+                <Loader2 size={16} className="animate-spin" />
+                正在加载测试范围…
+              </div>
+            ) : scopeError ? (
+              <p
+                className="rounded-xl border border-destructive/20 bg-destructive/5 px-3 py-3 text-sm text-destructive"
+                role="alert"
+              >
+                {scopeError}
+              </p>
+            ) : (
+              <TestScopeSelector
+                suites={suites}
+                casesBySuite={casesBySuite}
+                selectedCaseIds={selectedCaseIds}
+                onSelectionChange={setSelectedCaseIds}
+              />
+            )}
          </SectionCard>
        </section>

--- a/frontend/src/components/pages/TestCasesPage.tsx
+++ b/frontend/src/components/pages/TestCasesPage.tsx
@@ -48,41 +48,24 @@ import {
 import { Input } from "@/components/ui/input";
 import { ListToolbar } from "@/components/ui/list-toolbar";
 import { SearchInput } from "@/components/ui/search-input";
-import {
-  Select,
-  SelectContent,
-  SelectItem,
-  SelectTrigger,
-  SelectValue,
-} from "@/components/ui/select";
 import {
  cloneOverallCriteria,
  cloneTurns,
  createEmptyFixedInputTurn,
-  createTestCase,
-  createTestSuite,
  DEFAULT_INPUT_MODE,
-  duplicateTestCase,
-  duplicateTestSuite,
  getTestCaseValidationMessage,
-  getTestSuite,
-  listTestCases,
-  listTestSuites,
  normalizeOverallCriteria,
-  removeTestCase,
-  removeTestCases,
-  removeTestSuite,
-  reorderTestCases,
-  suiteCaseStats,
  TEST_CASE_INPUT_MODE_LABEL,
  TEST_CASE_INPUT_MODE_SHORT_LABEL,
-  updateTestCase,
-  updateTestSuite,
  type TestCase,
  type TestCaseInputMode,
  type TestSuite,
 } from "@/data/test-suites";
-import { assistantsApi, type Assistant } from "@/lib/api";
+import {
+  testCasesApi,
+  testSuitesApi,
+  type TestCaseWrite,
+} from "@/lib/api";

 // 路由驱动:
 // /test/cases → list
@@ -109,16 +92,33 @@ export function TestCasesPage(props: TestCasesPageProps) {

 function SuiteListView() {
  const router = useRouter();
-  const [suites, setSuites] = useState<TestSuite[]>(() => listTestSuites());
+  const [suites, setSuites] = useState<TestSuite[]>([]);
+  const [loading, setLoading] = useState(true);
+  const [loadError, setLoadError] = useState("");
  const [search, setSearch] = useState("");
  const [currentPage, setCurrentPage] = useState(1);
  const [deletingId, setDeletingId] = useState<string | null>(null);

+  async function reloadSuites() {
+    try {
+      setLoadError("");
+      setSuites(await testSuitesApi.list());
+    } catch (error) {
+      setLoadError(error instanceof Error ? error.message : "加载测试集失败");
+    } finally {
+      setLoading(false);
+    }
+  }
+
+  useEffect(() => {
+    void reloadSuites();
+  }, []);
+
  const filtered = useMemo(() => {
    const keyword = search.trim().toLowerCase();
    return suites.filter((suite) => {
      if (!keyword) return true;
-      return [suite.name, suite.assistantName, suite.id]
+      return [suite.name, suite.description, suite.id]
        .join(" ")
        .toLowerCase()
        .includes(keyword);
@@ -136,13 +136,16 @@ function SuiteListView() {
    router.push(`/test/cases/${suite.id}`);
  }

-  function duplicateSuite(suite: TestSuite) {
-    const copied = duplicateTestSuite(suite.id);
-    if (!copied) return;
-    setSuites(listTestSuites());
+  async function duplicateSuite(suite: TestSuite) {
+    try {
+      await testSuitesApi.duplicate(suite.id);
+      await reloadSuites();
+    } catch (error) {
+      setLoadError(error instanceof Error ? error.message : "复制测试集失败");
+    }
  }

-  function removeSuite(suite: TestSuite) {
+  async function removeSuite(suite: TestSuite) {
    if (
      !window.confirm(
        `确定删除测试集“${suite.name}”及其全部测试用例吗？`,
@@ -151,9 +154,14 @@ function SuiteListView() {
      return;
    }
    setDeletingId(suite.id);
-    removeTestSuite(suite.id);
-    setSuites(listTestSuites());
-    setDeletingId(null);
+    try {
+      await testSuitesApi.remove(suite.id);
+      await reloadSuites();
+    } catch (error) {
+      setLoadError(error instanceof Error ? error.message : "删除测试集失败");
+    } finally {
+      setDeletingId(null);
+    }
  }

  return (
@@ -186,14 +194,26 @@ function SuiteListView() {
          }
        />

+        {loadError && (
+          <p className="mb-3 text-sm text-destructive" role="alert">
+            {loadError}
+          </p>
+        )}
+
        <DataList<TestSuite>
          rows={paginated}
          rowKey={(suite) => suite.id}
          onRowClick={openSuite}
          empty={{
-            title: suites.length === 0 ? "暂无测试集" : "未找到匹配的测试集",
+            title: loading
+              ? "正在加载测试集…"
+              : suites.length === 0
+                ? "暂无测试集"
+                : "未找到匹配的测试集",
            description:
-              suites.length === 0
+              loading
+                ? "请稍候。"
+                : suites.length === 0
                ? "点击右上角「新建测试集」开始。"
                : "请调整关键词后再试。",
          }}
@@ -222,19 +242,12 @@ function SuiteListView() {
                </>
              ),
            },
-            {
-              key: "assistant",
-              header: "关联助手",
-              width: "md:w-[160px]",
-              cellClassName: "text-muted-foreground",
-              cell: (suite) => suite.assistantName || "—",
-            },
            {
              key: "caseCount",
              header: "用例数",
              width: "md:w-[96px]",
              cellClassName: "tabular-nums text-muted-foreground",
-              cell: (suite) => suiteCaseStats(suite.id).total,
+              cell: (suite) => suite.caseCount ?? 0,
            },
            {
              key: "actions",
@@ -313,42 +326,31 @@ function SuiteListView() {

 function SuiteCreateView() {
  const router = useRouter();
-  const [assistants, setAssistants] = useState<Assistant[]>([]);
-  const [loadingAssistants, setLoadingAssistants] = useState(true);
  const [name, setName] = useState("");
-  const [assistantName, setAssistantName] = useState("");
  const [creating, setCreating] = useState(false);
+  const [createError, setCreateError] = useState("");

-  useEffect(() => {
-    void (async () => {
-      try {
-        const list = await assistantsApi.list();
-        setAssistants(list);
-        if (list[0]) setAssistantName(list[0].name);
-      } catch {
-        setAssistantName("视频快处助手");
-      } finally {
-        setLoadingAssistants(false);
-      }
-    })();
-  }, []);
-
-  function confirmCreate() {
+  async function confirmCreate() {
    if (!name.trim() || creating) return;
    setCreating(true);
-    const saved = createTestSuite({
-      name,
-      description: "",
-      assistantName,
-    });
-    router.push(`/test/cases/${saved.id}`);
+    setCreateError("");
+    try {
+      const saved = await testSuitesApi.create({
+        name: name.trim(),
+        description: "",
+      });
+      router.push(`/test/cases/${saved.id}`);
+    } catch (error) {
+      setCreateError(error instanceof Error ? error.message : "创建测试集失败");
+      setCreating(false);
+    }
  }

  return (
    <ListPageLayout
      title="新建测试集"
      className="max-w-[1180px]"
-      description="测试集是固定输入用例的业务分组。确认后进入用例编辑。"
+      description="测试集只保存固定输入和预期；被测助手在批量运行时选择。"
      action={
        <Button
          variant="outline"
@@ -375,39 +377,11 @@ function SuiteCreateView() {
        </label>
      </ListPageSection>

-      <ListPageSection>
-        <div>
-          <div className="mb-2 text-sm font-medium text-foreground">
-            关联助手
-          </div>
-          {loadingAssistants ? (
-            <div className="flex items-center gap-2 text-sm text-muted-foreground">
-              <Loader2 size={16} className="animate-spin" />
-              正在加载助手…
-            </div>
-          ) : assistants.length > 0 ? (
-            <Select value={assistantName} onValueChange={setAssistantName}>
-              <SelectTrigger className="w-full border-hairline-strong bg-background">
-                <SelectValue placeholder="选择关联助手" />
-              </SelectTrigger>
-              <SelectContent>
-                {assistants.map((item) => (
-                  <SelectItem key={item.id} value={item.name}>
-                    {item.name}
-                  </SelectItem>
-                ))}
-              </SelectContent>
-            </Select>
-          ) : (
-            <Input
-              value={assistantName}
-              onChange={(event) => setAssistantName(event.target.value)}
-              placeholder="助手名称"
-              className="border-hairline-strong bg-background text-foreground placeholder:text-muted-soft"
-            />
-          )}
-        </div>
-      </ListPageSection>
+      {createError && (
+        <p className="text-sm text-destructive" role="alert">
+          {createError}
+        </p>
+      )}

      <div className="flex items-center justify-end gap-3">
        <Button
@@ -476,64 +450,67 @@ function SuiteDetailView({
  initialCaseId?: string;
 }) {
  const router = useRouter();
-  const [initial] = useState(() => {
-    const initialCases = listTestCases(suiteId);
-    const initialCase =
-      initialCases.find((item) => item.id === initialCaseId) ??
-      initialCases[0] ??
-      null;
-    const initialDraft = initialCase ? caseToDraft(initialCase) : null;
-    return {
-      suite: getTestSuite(suiteId),
-      cases: initialCases,
-      selectedId: initialCase?.id ?? null,
-      draft: initialDraft,
-    };
-  });
-  const [suite, setSuite] = useState<TestSuite | null>(initial.suite);
-  const [cases, setCases] = useState<TestCase[]>(initial.cases);
-  const [selectedId, setSelectedId] = useState<string | null>(
-    initial.selectedId,
-  );
+  const [suite, setSuite] = useState<TestSuite | null>(null);
+  const [cases, setCases] = useState<TestCase[]>([]);
+  const [selectedId, setSelectedId] = useState<string | null>(null);
  const [search, setSearch] = useState("");
-  const [draft, setDraft] = useState<CaseEditorDraft | null>(initial.draft);
-  const [savedSnapshot, setSavedSnapshot] = useState<CaseEditorDraft | null>(
-    initial.draft,
-  );
+  const [draft, setDraft] = useState<CaseEditorDraft | null>(null);
+  const [savedSnapshot, setSavedSnapshot] = useState<CaseEditorDraft | null>(null);
  const [statusMessage, setStatusMessage] = useState("");
  const [selectionMode, setSelectionMode] = useState(false);
  const [checkedIds, setCheckedIds] = useState<Set<string>>(new Set());
+  const [loading, setLoading] = useState(true);
+  const [saving, setSaving] = useState(false);

-  function reload(preferId?: string | null) {
-    const nextSuite = getTestSuite(suiteId);
-    const nextCases = listTestCases(suiteId);
-    setSuite(nextSuite);
-    setCases(nextCases);
+  async function reload(preferId?: string | null) {
+    try {
+      const [nextSuite, nextCases] = await Promise.all([
+        testSuitesApi.get(suiteId),
+        testCasesApi.list(suiteId),
+      ]);
+      setSuite(nextSuite);
+      setCases(nextCases);

-    const nextSelected =
-      (preferId && nextCases.some((item) => item.id === preferId)
-        ? preferId
-        : null) ??
-      (selectedId && nextCases.some((item) => item.id === selectedId)
-        ? selectedId
-        : null) ??
-      nextCases[0]?.id ??
-      null;
+      const nextSelected =
+        (preferId && nextCases.some((item) => item.id === preferId)
+          ? preferId
+          : null) ??
+        (selectedId && nextCases.some((item) => item.id === selectedId)
+          ? selectedId
+          : null) ??
+        nextCases[0]?.id ??
+        null;

-    setSelectedId(nextSelected);
-    if (nextSelected) {
-      const item = nextCases.find((caseItem) => caseItem.id === nextSelected);
-      if (item) {
-        const nextDraft = caseToDraft(item);
-        setDraft(nextDraft);
-        setSavedSnapshot(nextDraft);
+      setSelectedId(nextSelected);
+      if (nextSelected) {
+        const item = nextCases.find((caseItem) => caseItem.id === nextSelected);
+        if (item) {
+          const nextDraft = caseToDraft(item);
+          setDraft(nextDraft);
+          setSavedSnapshot(nextDraft);
+        }
+      } else {
+        setDraft(null);
+        setSavedSnapshot(null);
      }
-    } else {
+      setStatusMessage("");
+    } catch (error) {
+      setSuite(null);
+      setCases([]);
      setDraft(null);
      setSavedSnapshot(null);
+      setStatusMessage(error instanceof Error ? error.message : "加载测试集失败");
+    } finally {
+      setLoading(false);
    }
  }

+  useEffect(() => {
+    void reload(initialCaseId ?? null);
+    // suiteId changes by remount; the initial request only runs once per view.
+    // eslint-disable-next-line react-hooks/exhaustive-deps
+  }, [suiteId, initialCaseId]);
+
  const filtered = useMemo(() => {
    const keyword = search.trim().toLowerCase();
    return cases.filter((item) => {
@@ -558,7 +535,10 @@ function SuiteDetailView({
    ? getTestCaseValidationMessage(draft)
    : "请先选择测试用例";
  const canSave =
-    Boolean(draft?.name.trim()) && dirty && validationMessage === null;
+    Boolean(draft?.name.trim()) &&
+    dirty &&
+    validationMessage === null &&
+    !saving;

  useEffect(() => {
    if (!dirty) return;
@@ -629,39 +609,36 @@ function SuiteDetailView({
    setStatusMessage("");
  }

-  function handleSave() {
-    if (!draft || validationMessage) return;
-    const patch = {
+  async function handleSave() {
+    if (!draft || validationMessage || saving) return;
+    const body: TestCaseWrite = {
      name: draft.name.trim() || "未命名用例",
+      description:
+        cases.find((item) => item.id === selectedId)?.description ?? "",
      inputMode: draft.inputMode,
      contextTurns: draft.contextTurns,
      turns: draft.turns,
      overallCriteria: normalizeOverallCriteria(draft.overallCriteria),
    };

-    let saved: TestCase | null;
-    if (selectedId) {
-      saved = updateTestCase(selectedId, patch);
-    } else {
-      const created = createTestCase({
-        suiteId,
-        name: patch.name,
-      });
-      saved = created ? updateTestCase(created.id, patch) : null;
+    setSaving(true);
+    setStatusMessage("保存中…");
+    try {
+      const saved = selectedId
+        ? await testCasesApi.update(selectedId, body)
+        : await testCasesApi.create(suiteId, body);
+      await reload(saved.id);
+      setStatusMessage("已保存");
+      window.setTimeout(() => setStatusMessage(""), 2000);
+    } catch (error) {
+      setStatusMessage(error instanceof Error ? error.message : "保存测试用例失败");
+    } finally {
+      setSaving(false);
    }
-    if (!saved) return;
-    const nextDraft = caseToDraft(saved);
-    setSelectedId(saved.id);
-    setCases(listTestCases(suiteId));
-    setSuite(getTestSuite(suiteId));
-    setDraft(nextDraft);
-    setSavedSnapshot(nextDraft);
-    setStatusMessage("已保存");
-    window.setTimeout(() => setStatusMessage(""), 2000);
  }

  function handleCancelNewCase() {
-    reload();
+    void reload();
    setStatusMessage("");
  }

@@ -672,28 +649,34 @@ function SuiteDetailView({
    router.push("/test/cases");
  }

-  function handleDuplicateCase(item: TestCase) {
+  async function handleDuplicateCase(item: TestCase) {
    if (dirty && !window.confirm("当前用例有未保存修改，复制将丢弃。继续？")) {
      return;
    }
-    const copied = duplicateTestCase(item.id);
-    if (!copied) return;
-    reload(copied.id);
-    setStatusMessage("");
+    try {
+      const copied = await testCasesApi.duplicate(item.id);
+      await reload(copied.id);
+    } catch (error) {
+      setStatusMessage(error instanceof Error ? error.message : "复制测试用例失败");
+    }
  }

-  function handleDeleteCase(item: TestCase) {
+  async function handleDeleteCase(item: TestCase) {
    if (!window.confirm(`确定删除测试用例“${item.name}”吗？`)) return;
-    removeTestCase(item.id);
-    setCheckedIds((prev) => {
-      const next = new Set(prev);
-      next.delete(item.id);
-      return next;
-    });
-    reload(selectedId === item.id ? null : selectedId);
+    try {
+      await testCasesApi.remove(item.id);
+      setCheckedIds((prev) => {
+        const next = new Set(prev);
+        next.delete(item.id);
+        return next;
+      });
+      await reload(selectedId === item.id ? null : selectedId);
+    } catch (error) {
+      setStatusMessage(error instanceof Error ? error.message : "删除测试用例失败");
+    }
  }

-  function handleDeleteSelected() {
+  async function handleDeleteSelected() {
    if (!selectedId || !draft) return;
    if (
      !window.confirm(
@@ -702,8 +685,12 @@ function SuiteDetailView({
    ) {
      return;
    }
-    removeTestCase(selectedId);
-    reload(null);
+    try {
+      await testCasesApi.remove(selectedId);
+      await reload(null);
+    } catch (error) {
+      setStatusMessage(error instanceof Error ? error.message : "删除测试用例失败");
+    }
  }

  function handleEnterSelectionMode() {
@@ -743,32 +730,66 @@ function SuiteDetailView({
    });
  }

-  function handleBulkDelete() {
+  async function handleBulkDelete() {
    const ids = filtered
      .filter((item) => checkedIds.has(item.id))
      .map((item) => item.id);
    if (ids.length === 0) return;
    if (!window.confirm(`确定删除选中的 ${ids.length} 个测试用例吗？`)) return;
-    removeTestCases(ids);
-    handleExitSelectionMode();
-    reload(selectedId && ids.includes(selectedId) ? null : selectedId);
+    try {
+      await testCasesApi.bulkRemove(ids);
+      handleExitSelectionMode();
+      await reload(selectedId && ids.includes(selectedId) ? null : selectedId);
+    } catch (error) {
+      setStatusMessage(error instanceof Error ? error.message : "批量删除测试用例失败");
+    }
  }

-  function handleReorder(orderedIds: string[]) {
-    reorderTestCases(suiteId, orderedIds);
-    setCases(listTestCases(suiteId));
+  async function handleReorder(orderedIds: string[]) {
+    const order = new Map(orderedIds.map((id, index) => [id, index]));
+    setCases((current) =>
+      [...current].sort(
+        (left, right) =>
+          (order.get(left.id) ?? Number.MAX_SAFE_INTEGER) -
+          (order.get(right.id) ?? Number.MAX_SAFE_INTEGER),
+      ),
+    );
+    try {
+      await testCasesApi.reorder(suiteId, orderedIds);
+    } catch (error) {
+      setStatusMessage(error instanceof Error ? error.message : "调整用例顺序失败");
+      await reload(selectedId);
+    }
  }

-  function renameSuite(nextName: string) {
+  async function renameSuite(nextName: string) {
    if (!suite || nextName === suite.name) return;
-    const saved = updateTestSuite(suite.id, { name: nextName });
-    if (saved) setSuite(saved);
+    try {
+      const saved = await testSuitesApi.update(suite.id, {
+        name: nextName,
+        description: suite.description,
+      });
+      setSuite(saved);
+    } catch (error) {
+      setStatusMessage(error instanceof Error ? error.message : "重命名测试集失败");
+    }
+  }
+
+  if (loading) {
+    return (
+      <div className="flex h-full items-center justify-center gap-2 text-sm text-muted-foreground">
+        <Loader2 size={16} className="animate-spin" />
+        正在加载测试集…
+      </div>
+    );
  }

  if (!suite) {
    return (
      <div className="mx-auto flex w-full max-w-[1280px] flex-col gap-4 py-16">
-        <div className="font-medium text-destructive">测试集不存在</div>
+        <div className="font-medium text-destructive">
+          {statusMessage || "测试集不存在"}
+        </div>
        <Button
          variant="outline"
          size="sm"
@@ -791,7 +812,7 @@ function SuiteDetailView({
          />
          <EditableTitle
            value={suite.name}
-            onChange={renameSuite}
+            onChange={(value) => void renameSuite(value)}
            placeholder="未命名测试集"
            editLabel="测试集名称"
          />
@@ -813,14 +834,14 @@ function SuiteDetailView({
          onSearchChange={setSearch}
          onSelectCase={selectCase}
          onCreateCase={handleCreateCase}
-          onDuplicateCase={handleDuplicateCase}
-          onDeleteCase={handleDeleteCase}
+          onDuplicateCase={(item) => void handleDuplicateCase(item)}
+          onDeleteCase={(item) => void handleDeleteCase(item)}
          onEnterSelectionMode={handleEnterSelectionMode}
          onExitSelectionMode={handleExitSelectionMode}
          onToggleChecked={handleToggleChecked}
          onToggleSelectAll={handleToggleSelectAll}
-          onBulkDelete={handleBulkDelete}
-          onReorder={handleReorder}
+          onBulkDelete={() => void handleBulkDelete()}
+          onReorder={(ids) => void handleReorder(ids)}
        />

        <main className="flex min-h-0 min-w-0 flex-1 flex-col overflow-hidden bg-background">
@@ -868,7 +889,7 @@ function SuiteDetailView({
                    className="gap-1.5"
                    disabled={!canSave}
                    title={validationMessage ?? undefined}
-                    onClick={handleSave}
+                    onClick={() => void handleSave()}
                  >
                    <Save size={14} />
                    保存
@@ -882,7 +903,9 @@ function SuiteDetailView({
                        : "border-hairline-strong text-muted-foreground hover:text-foreground"
                    }
                    onClick={
-                      selectedId ? handleDeleteSelected : handleCancelNewCase
+                      selectedId
+                        ? () => void handleDeleteSelected()
+                        : handleCancelNewCase
                    }
                  >
                    {selectedId ? <Trash2 size={14} /> : <X size={14} />}
--- a/frontend/src/components/test-cases/next-reply-editor.tsx
+++ b/frontend/src/components/test-cases/next-reply-editor.tsx
@@ -122,25 +122,6 @@ function editorSections(): {
  ];
 }

-/** API 不可用时的兜底工具，保证下拉可选 */
-const FALLBACK_TOOLS: EditorToolOption[] = [
-  {
-    id: "mock_transfer_to_human",
-    functionName: "transfer_to_human",
-    label: "转人工",
-    parameters: [
-      { name: "reason", description: "转人工原因" },
-      { name: "target", description: "转接目标" },
-    ],
-  },
-  {
-    id: "mock_end_conversation",
-    functionName: "end_conversation",
-    label: "结束会话",
-    parameters: [{ name: "reason", description: "结束原因" }],
-  },
-];
-
 function toolParametersFromDefinition(tool: Tool): EditorToolOption["parameters"] {
  const definition = tool.definition;
  if (definition.type === "http" || definition.type === "client") {
@@ -208,9 +189,8 @@ export function NextReplyEditorBody({
  const sections = editorSections();

  const tools = useMemo(() => {
-    if (availableTools && availableTools.length > 0) return availableTools;
-    if (loadedTools.length > 0) return loadedTools;
-    return FALLBACK_TOOLS;
+    if (availableTools) return availableTools;
+    return loadedTools;
  }, [availableTools, loadedTools]);

  const sensors = useSensors(
@@ -221,16 +201,15 @@ export function NextReplyEditorBody({
  );

  useEffect(() => {
-    if (availableTools && availableTools.length > 0) return;
+    if (availableTools) return;
    let cancelled = false;
    void (async () => {
      try {
        const list = await toolsApi.list();
        if (cancelled) return;
-        const mapped = list.map(toEditorToolOption);
-        setLoadedTools(mapped.length > 0 ? mapped : FALLBACK_TOOLS);
+        setLoadedTools(list.map(toEditorToolOption));
      } catch {
-        if (!cancelled) setLoadedTools(FALLBACK_TOOLS);
+        if (!cancelled) setLoadedTools([]);
      }
    })();
    return () => {
@@ -1464,14 +1443,20 @@ function ToolCallBehaviorEditor({
            <SelectValue placeholder="选择工具…" />
          </SelectTrigger>
          <SelectContent>
-            {tools.map((tool) => (
-              <SelectItem key={tool.id} value={tool.id}>
-                {tool.functionName}
-                {tool.label && tool.label !== tool.functionName
-                  ? ` · ${tool.label}`
-                  : ""}
-              </SelectItem>
-            ))}
+            {tools.length === 0 ? (
+              <div className="px-2 py-3 text-xs text-muted-foreground">
+                暂无可用工具
+              </div>
+            ) : (
+              tools.map((tool) => (
+                <SelectItem key={tool.id} value={tool.id}>
+                  {tool.functionName}
+                  {tool.label && tool.label !== tool.functionName
+                    ? ` · ${tool.label}`
+                    : ""}
+                </SelectItem>
+              ))
+            )}
          </SelectContent>
        </Select>
      </div>
--- a/frontend/src/data/batch-run.ts
+++ b/frontend/src/data/batch-run.ts
@@ -1,14 +1,4 @@
-/**
- * 批量测试运行 — 前端 mock。
- * 真实执行引擎接入前，用本地状态模拟进度与结果。
- */
-
-import type {
-  ExpectedBehavior,
-  ReplyExpectedBehavior,
-  TestCase,
-  ToolCallExpectedBehavior,
-} from "@/data/test-suites";
+/** 批量测试运行的前后端共享结果契约与纯展示辅助函数。 */

 export type BatchCaseStatus =
  | "waiting"
@@ -34,24 +24,13 @@ export const BATCH_ERROR_STRATEGY_LABEL: Record<BatchErrorStrategy, string> = {
  stop_on_error: "标记错误并停止",
 };

-type BatchExecutionError = {
+export type BatchExecutionError = {
  code: string;
  message: string;
  stage: "pipeline" | "model" | "tool" | "evaluation";
  retryable: boolean;
 };

-export type BatchRunCase = {
-  id: string;
-  name: string;
-  status: BatchCaseStatus;
-  turns: BatchTurnResult[];
-  overallCriteria: BatchEvaluationResult[];
-  attemptCount: number;
-  maxAttempts: number;
-  executionError: BatchExecutionError | null;
-};
-
 export type BatchEvaluationResult = {
  id: string;
  label: string;
@@ -80,13 +59,30 @@ export type BatchTurnResult = {
  evaluations: BatchEvaluationResult[];
 };

+export type BatchRunCase = {
+  id: string;
+  name: string;
+  status: BatchCaseStatus;
+  turns: BatchTurnResult[];
+  overallCriteria: BatchEvaluationResult[];
+  attemptCount: number;
+  maxAttempts: number;
+  executionError: BatchExecutionError | null;
+};
+
 export type BatchRunPhase = "config" | "running" | "completed";
+export type BatchRunStatus = "queued" | "running" | "completed" | "cancelled";

 export type BatchRunSnapshot = {
+  id: string;
+  status: BatchRunStatus;
  title: string;
  assistantName: string;
  config: {
    suiteCount: number;
+    evaluatorModelResourceId: string;
+    evaluatorModelResourceName: string;
+    evaluatorModel: string;
    concurrency: number;
    timeoutSecs: number;
    failureStrategy: BatchFailureStrategy;
@@ -99,246 +95,6 @@ export type BatchRunSnapshot = {
  stopReason: "manual" | "assertion_failure" | "execution_error" | null;
 };

-export type BatchMockExecutionPlan = Record<string, BatchExecutionError>;
-
-/** 预定部分用例失败，方便演示失败展开态 */
-function shouldFail(index: number, item: TestCase): boolean {
-  if (item.lastResult === "fail") return true;
-  return index % 4 === 2;
-}
-
-/** 稳定产生少量执行错误，便于检查重试和继续/停止流程。 */
-function shouldError(index: number): boolean {
-  return index % 5 === 3;
-}
-
-function replyExpectation(behavior: ReplyExpectedBehavior): string {
-  if (behavior.assertionType === "llm") return behavior.llmCriteria;
-  const relation = behavior.negateKeywords
-    ? "不应包含"
-    : behavior.keywordMatchMode === "all"
-      ? "应包含全部"
-      : "应至少包含其一";
-  return `${relation}：${behavior.keywords.join("、")}`;
-}
-
-function toolExpectation(behavior: ToolCallExpectedBehavior): string {
-  if (behavior.expectation === "not_called") {
-    return `不应调用 ${behavior.functionName}`;
-  }
-  const count =
-    behavior.maxCalls === null
-      ? `${behavior.minCalls} 次以上`
-      : behavior.minCalls === behavior.maxCalls
-        ? `${behavior.minCalls} 次`
-        : `${behavior.minCalls}–${behavior.maxCalls} 次`;
-  return `应调用 ${behavior.functionName} ${count}`;
-}
-
-function mockAssistantReply(behaviors: ExpectedBehavior[]): string {
-  const positiveKeyword = behaviors.find(
-    (behavior): behavior is ReplyExpectedBehavior =>
-      behavior.type === "reply" &&
-      behavior.assertionType === "keyword" &&
-      !behavior.negateKeywords &&
-      behavior.keywords.length > 0,
-  );
-  if (positiveKeyword) {
-    return `好的，我已了解。请继续说明${positiveKeyword.keywords.slice(0, 2).join("和")}。`;
-  }
-  return "好的，我已记录当前信息，我们继续处理。";
-}
-
-function mockToolArguments(behavior: ToolCallExpectedBehavior): string {
-  const entries = behavior.paramAssertions.map((item) => [
-    item.name,
-    item.matchMode === "exact" ? item.value : `mock_${item.name}`,
-  ]);
-  return JSON.stringify(Object.fromEntries(entries), null, 2);
-}
-
-function createBehaviorEvaluation(
-  behavior: ExpectedBehavior,
-  fail: boolean,
-): {
-  evaluation: BatchEvaluationResult;
-  toolCalls: BatchToolCallRecord[];
-} {
-  if (behavior.type === "reply") {
-    const reason = fail
-      ? behavior.assertionType === "keyword"
-        ? "实际回复未满足关键词规则。"
-        : "LLM 判断认为回复未达到该项语义要求。"
-      : "";
-    return {
-      evaluation: {
-        id: behavior.id,
-        label:
-          behavior.assertionType === "keyword"
-            ? "回复 · 关键词"
-            : "回复 · LLM 判断",
-        kind: "reply",
-        status: fail ? "fail" : "pass",
-        expected: replyExpectation(behavior),
-        actual: fail
-          ? "已为您转接人工处理。"
-          : "实际回复满足本条内容要求。",
-        reason,
-      },
-      toolCalls: [],
-    };
-  }
-
-  const shouldRecordCall =
-    behavior.expectation === "called" ? !fail : fail;
-  const toolCalls: BatchToolCallRecord[] = shouldRecordCall
-    ? [
-        {
-          id: `call_${behavior.id}`,
-          functionName: behavior.functionName,
-          argumentsJson: mockToolArguments(behavior),
-          resultJson: behavior.mockResponse.body,
-          outcome: behavior.mockResponse.outcome,
-          durationMs: behavior.mockResponse.delayMs + 84,
-        },
-      ]
-    : [];
-  const actual = shouldRecordCall
-    ? `实际调用 1 次 ${behavior.functionName}`
-    : `实际调用 0 次 ${behavior.functionName}`;
-  const reason = fail
-    ? behavior.expectation === "not_called"
-      ? "检测到本轮不应发生的工具调用。"
-      : "未检测到满足次数要求的工具调用。"
-    : "";
-  return {
-    evaluation: {
-      id: behavior.id,
-      label: `工具 · ${behavior.functionName}`,
-      kind: "tool_call",
-      status: fail ? "fail" : "pass",
-      expected: toolExpectation(behavior),
-      actual,
-      reason,
-    },
-    toolCalls,
-  };
-}
-
-function createCaseResult(
-  item: TestCase,
-  caseIndex: number,
-  maxAttempts: number,
-): { result: BatchRunCase; plannedError: BatchExecutionError | null } {
-  const plannedError = shouldError(caseIndex);
-  const plannedFailure = !plannedError && shouldFail(caseIndex, item);
-  let failureAssigned = false;
-  const turns = item.turns.map((turn, turnIndex) => {
-    const toolCalls: BatchToolCallRecord[] = [];
-    const evaluations = turn.behaviors.map((behavior) => {
-      const fail = plannedFailure && !failureAssigned;
-      if (fail) failureAssigned = true;
-      const result = createBehaviorEvaluation(behavior, fail);
-      toolCalls.push(...result.toolCalls);
-      return result.evaluation;
-    });
-    return {
-      id: turn.id,
-      index: turnIndex,
-      userInput: turn.userInput,
-      assistantReply: mockAssistantReply(turn.behaviors),
-      toolCalls,
-      evaluations,
-    };
-  });
-  const overallCriteria = item.overallCriteria.map((criterion) => {
-    const fail = plannedFailure && !failureAssigned;
-    if (fail) failureAssigned = true;
-    return {
-      id: criterion.id,
-      label: criterion.name,
-      kind: "overall" as const,
-      status: fail ? ("fail" as const) : ("pass" as const),
-      expected: criterion.criteria,
-      actual: fail
-        ? "整段对话未完整达到该业务目标。"
-        : "整段对话达到该业务目标。",
-      reason: fail ? "LLM 判断认为整段对话未满足该项标准。" : "",
-    };
-  });
-  return {
-    result: {
-      id: item.id,
-      name: item.name,
-      status: "waiting",
-      turns,
-      overallCriteria,
-      attemptCount: 0,
-      maxAttempts,
-      executionError: null,
-    },
-    plannedError: plannedError
-      ? {
-          code: "PIPELINE_TIMEOUT",
-          message: "等待 pipeline 完成回复时超过单用例超时时间。",
-          stage: "pipeline",
-          retryable: true,
-        }
-      : null,
-  };
-}
-
-export function createMockBatchRun(input: {
-  title: string;
-  assistantName: string;
-  cases: TestCase[];
-  concurrency: number;
-  timeoutSecs: number;
-  failureStrategy: BatchFailureStrategy;
-  errorRetryCount: number;
-  errorStrategy: BatchErrorStrategy;
-}): {
-  snapshot: BatchRunSnapshot;
-  executionPlan: BatchMockExecutionPlan;
-} {
-  const preparedCases = input.cases.map((item, index) =>
-    createCaseResult(item, index, input.errorRetryCount + 1),
-  );
-  const executionPlan = Object.fromEntries(
-    preparedCases.flatMap(({ result, plannedError }) =>
-      plannedError ? [[result.id, plannedError]] : [],
-    ),
-  );
-
-  return {
-    snapshot: {
-      title: input.title,
-      assistantName: input.assistantName,
-      config: {
-        suiteCount: new Set(input.cases.map((item) => item.suiteId)).size,
-        concurrency: input.concurrency,
-        timeoutSecs: input.timeoutSecs,
-        failureStrategy: input.failureStrategy,
-        errorRetryCount: input.errorRetryCount,
-        errorStrategy: input.errorStrategy,
-      },
-      startedAt: new Date().toISOString(),
-      finishedAt: null,
-      stopReason: null,
-      cases: preparedCases.map(({ result }) => result),
-    },
-    executionPlan,
-  };
-}
-
-export function hasEvaluationFailure(item: BatchRunCase): boolean {
-  const evaluations = [
-    ...item.turns.flatMap((turn) => turn.evaluations),
-    ...item.overallCriteria,
-  ];
-  return evaluations.some((evaluation) => evaluation.status === "fail");
-}
-
 export function firstEvaluationFailureReason(item: BatchRunCase): string {
  const evaluations = [
    ...item.turns.flatMap((turn) => turn.evaluations),
@@ -358,15 +114,13 @@ export function countByStatus(cases: BatchRunCase[]) {
    waiting: 0,
    skipped: 0,
  };
-  for (const item of cases) {
-    counts[item.status] += 1;
-  }
+  for (const item of cases) counts[item.status] += 1;
  return counts;
 }

 export function formatRunTime(iso: string): string {
  const date = new Date(iso);
-  const pad = (n: number) => String(n).padStart(2, "0");
+  const pad = (value: number) => String(value).padStart(2, "0");
  return `${date.getFullYear()}-${pad(date.getMonth() + 1)}-${pad(date.getDate())} ${pad(date.getHours())}:${pad(date.getMinutes())}:${pad(date.getSeconds())}`;
 }

--- a/frontend/src/data/test-suites.ts
+++ b/frontend/src/data/test-suites.ts
@@ -1,5 +1,5 @@
 /**
- * 测试集 / 测试用例 — 管理页用的本地 mock。
+ * 测试集 / 测试用例的前后端共享契约与编辑器辅助函数。
 * 两层：Test Suite → Test Case。
 * MVP 仅运行固定文字脚本；其它输入模式保留为产品路线提示。
 */
@@ -12,7 +12,7 @@ export type TestCaseInputMode =
  | "user_sim_text"
  | "user_sim_voice";

-type TestCaseResult = "pass" | "fail" | "not_run";
+export type TestCaseResult = "pass" | "fail" | "not_run";

 export type AssertionType = "keyword" | "llm";

@@ -105,8 +105,9 @@ export type TestSuite = {
  id: string;
  name: string;
  description: string;
-  /** 关联助手展示名；MVP 直接存文案，接 API 后可改成 assistantId */
-  assistantName: string;
+  caseCount: number;
+  passedCount: number;
+  runCount: number;
  updatedAt: string;
 };

@@ -160,12 +161,60 @@ export function getTestCaseValidationMessage(
    }
    if (context.role === "tool_call" || context.role === "tool_result") {
      try {
-        JSON.parse(context.content);
+        const payload = JSON.parse(context.content);
+        if (
+          context.role === "tool_call" &&
+          (payload === null || Array.isArray(payload) || typeof payload !== "object")
+        ) {
+          return `上下文第 ${index + 1} 条 Tool Call 参数必须是 JSON 对象`;
+        }
      } catch {
        return `上下文第 ${index + 1} 条工具数据必须是有效 JSON`;
      }
    }
  }
+  const pendingToolCalls: Array<{ id: string | null; toolName: string }> = [];
+  const seenToolCallIds = new Set<string>();
+  for (let index = 0; index < item.contextTurns.length; index += 1) {
+    const context = item.contextTurns[index];
+    if (context.role === "tool_call") {
+      const id = context.toolCallId?.trim() || null;
+      if (id && seenToolCallIds.has(id)) {
+        return `上下文第 ${index + 1} 条 Tool Call ID 重复`;
+      }
+      if (id) seenToolCallIds.add(id);
+      pendingToolCalls.push({ id, toolName: context.toolName?.trim() ?? "" });
+      continue;
+    }
+    if (context.role !== "tool_result") continue;
+    const resultId = context.toolCallId?.trim() || null;
+    let matchedIndex = -1;
+    for (
+      let pendingIndex = pendingToolCalls.length - 1;
+      pendingIndex >= 0;
+      pendingIndex -= 1
+    ) {
+      const pending = pendingToolCalls[pendingIndex];
+      if (
+        resultId ? pending.id === resultId : pending.toolName === context.toolName?.trim()
+      ) {
+        matchedIndex = pendingIndex;
+        break;
+      }
+    }
+    if (matchedIndex < 0) {
+      return `上下文第 ${index + 1} 条 Tool Result 没有匹配的 Tool Call`;
+    }
+    const [matched] = pendingToolCalls.splice(matchedIndex, 1);
+    if (matched.toolName !== context.toolName?.trim()) {
+      return `上下文第 ${index + 1} 条 Tool Result 的工具名称不匹配`;
+    }
+  }
+  if (pendingToolCalls.length > 0) {
+    return `上下文 Tool Call 缺少对应的 Tool Result：${pendingToolCalls
+      .map((item) => item.toolName)
+      .join("、")}`;
+  }
  const emptyTurnIndex = item.turns.findIndex(
    (turn) => !turn.userInput.trim(),
  );
@@ -311,26 +360,19 @@ export function getExpectedBehaviorValidationMessage(
  return null;
 }

-let turnSeq = 1;
-let behaviorSeq = 1;
-let criterionSeq = 1;
-
-function nextTurnId() {
-  const id = `turn_${String(turnSeq).padStart(4, "0")}`;
-  turnSeq += 1;
-  return id;
+function newEditorId(prefix: string) {
+  const randomPart = globalThis.crypto?.randomUUID
+    ? globalThis.crypto.randomUUID().replaceAll("-", "")
+    : `${Date.now().toString(36)}${Math.random().toString(36).slice(2)}`;
+  return `${prefix}_${randomPart}`;
 }

 function nextBehaviorId() {
-  const id = `beh_${String(behaviorSeq).padStart(4, "0")}`;
-  behaviorSeq += 1;
-  return id;
+  return newEditorId("beh");
 }

 function nextCriterionId() {
-  const id = `criterion_${String(criterionSeq).padStart(4, "0")}`;
-  criterionSeq += 1;
-  return id;
+  return newEditorId("criterion");
 }

 export function createOverallCriterion(
@@ -365,7 +407,7 @@ export function createEmptyFixedInputTurn(
  userInput = "",
 ): FixedInputTurn {
  return {
-    id: nextTurnId(),
+    id: newEditorId("turn"),
    userInput,
    behaviors: [],
  };
@@ -435,10 +477,6 @@ function cloneBehavior(behavior: ExpectedBehavior): ExpectedBehavior {
  };
 }

-function cloneBehaviorWithNewId(behavior: ExpectedBehavior): ExpectedBehavior {
-  return { ...cloneBehavior(behavior), id: nextBehaviorId() };
-}
-
 export function cloneTurns(turns: FixedInputTurn[]): FixedInputTurn[] {
  return turns.map((turn) => ({
    id: turn.id,
@@ -447,480 +485,6 @@ export function cloneTurns(turns: FixedInputTurn[]): FixedInputTurn[] {
  }));
 }

-/** 新建用例时复制轮次并分配新 id */
-function cloneTurnsWithNewIds(turns: FixedInputTurn[]): FixedInputTurn[] {
-  return turns.map((turn) => ({
-    id: nextTurnId(),
-    userInput: turn.userInput,
-    behaviors: turn.behaviors.map(cloneBehaviorWithNewId),
-  }));
-}
-
-const INITIAL_SUITES: TestSuite[] = [
-  {
-    id: "suite_001",
-    name: "事故基础流程",
-    description: "核心业务流程和正常事故处理",
-    assistantName: "视频快处助手",
-    updatedAt: "2026-08-05T10:20:00+08:00",
-  },
-  {
-    id: "suite_002",
-    name: "异常输入与澄清",
-    description: "模糊表达、误打断、主动唤醒等",
-    assistantName: "视频快处助手",
-    updatedAt: "2026-08-04T16:40:00+08:00",
-  },
-  {
-    id: "suite_003",
-    name: "实时语音交互",
-    description: "打断、延迟、VAD、语音链路等",
-    assistantName: "视频快处助手",
-    updatedAt: "2026-08-03T09:15:00+08:00",
-  },
-];
-
-type RawCaseSeed = Omit<TestCase, "sortOrder" | "inputMode" | "overallCriteria"> & {
-  inputMode?: TestCaseInputMode;
-  overallCriteria?: OverallCriterion[];
-};
-
-function seedReplyTurn(
-  seedId: string,
-  userInput: string,
-  expectation: Omit<ReplyExpectedBehavior, "id" | "type">,
-): FixedInputTurn {
-  return {
-    id: `turn_seed_${seedId}`,
-    userInput,
-    behaviors: [
-      {
-        id: `beh_seed_${seedId}`,
-        type: "reply",
-        ...expectation,
-      },
-    ],
-  };
-}
-
-const RAW_CASE_SEEDS: RawCaseSeed[] = [
-  {
-    id: "tc_001",
-    suiteId: "suite_001",
-    name: "正常双车事故开场",
-    description: "开场后用户补充事故经过",
-    lastResult: "pass",
-    contextTurns: [],
-    turns: [
-      seedReplyTurn("001", "我这里刚刚撞了一下。", {
-        assertionType: "keyword",
-        keywords: ["经过", "描述", "受伤"],
-        keywordMatchMode: "any",
-        negateKeywords: false,
-        llmCriteria: "",
-      }),
-    ],
-    updatedAt: "2026-08-05T10:18:00+08:00",
-  },
-  {
-    id: "tc_002",
-    suiteId: "suite_001",
-    name: "有人伤转人工",
-    description: "用户提到人伤时应引导转人工",
-    lastResult: "pass",
-    contextTurns: [],
-    turns: [
-      seedReplyTurn("002", "有人受伤了，流血不止。", {
-        assertionType: "llm",
-        keywords: [],
-        keywordMatchMode: "any",
-        negateKeywords: false,
-        llmCriteria:
-          "Agent 应识别人员受伤风险，明确告知将转接人工，并避免继续推进普通快处流程。",
-      }),
-    ],
-    overallCriteria: [
-      {
-        id: "criterion_seed_001",
-        type: "llm",
-        name: "人伤流程正确",
-        criteria:
-          "整段对话应正确识别人员受伤场景，及时并准确转接人工处理。",
-      },
-      {
-        id: "criterion_seed_002",
-        type: "llm",
-        name: "不推进普通快处",
-        criteria:
-          "确认存在人员受伤后，不应继续引导用户进入普通事故快处流程。",
-      },
-    ],
-    updatedAt: "2026-08-05T10:12:00+08:00",
-  },
-  {
-    id: "tc_101",
-    suiteId: "suite_002",
-    name: "用户说“喂”",
-    description: "验证主动唤醒回复且业务状态不推进",
-    lastResult: "pass",
-    contextTurns: [],
-    turns: [
-      seedReplyTurn("101", "喂", {
-        assertionType: "keyword",
-        keywords: ["我在", "请说", "继续"],
-        keywordMatchMode: "any",
-        negateKeywords: false,
-        llmCriteria: "",
-      }),
-    ],
-    updatedAt: "2026-08-04T16:35:00+08:00",
-  },
-  {
-    id: "tc_102",
-    suiteId: "suite_002",
-    name: "用户只说“嗯”",
-    description: "短促确认不应误推进流程",
-    lastResult: "fail",
-    contextTurns: [],
-    turns: [
-      seedReplyTurn("102", "嗯", {
-        assertionType: "llm",
-        keywords: [],
-        keywordMatchMode: "any",
-        negateKeywords: false,
-        llmCriteria:
-          "Agent 应当回应用户的主动唤醒，同时继续引导用户描述事故情况，不应该无响应。",
-      }),
-    ],
-    updatedAt: "2026-08-04T16:20:00+08:00",
-  },
-  {
-    id: "tc_103",
-    suiteId: "suite_002",
-    name: "模糊事故描述",
-    description: "地点含糊时应主动澄清",
-    lastResult: "not_run",
-    contextTurns: [],
-    turns: [
-      seedReplyTurn("103", "就在那边……撞了一下。", {
-        assertionType: "keyword",
-        keywords: ["哪里", "路口", "路名", "再说"],
-        keywordMatchMode: "any",
-        negateKeywords: false,
-        llmCriteria: "",
-      }),
-    ],
-    updatedAt: "2026-08-04T15:30:00+08:00",
-  },
-  {
-    id: "tc_201",
-    suiteId: "suite_003",
-    name: "用户打断播报",
-    description: "播报中打断后正确切换聆听并承接",
-    lastResult: "pass",
-    contextTurns: [],
-    turns: [
-      seedReplyTurn("201", "等一下，对方走了。", {
-        assertionType: "llm",
-        keywords: [],
-        keywordMatchMode: "any",
-        negateKeywords: false,
-        llmCriteria:
-          "Agent 应立即停止原播报思路，确认已听到用户新信息，并围绕「对方离开」继续询问。",
-      }),
-    ],
-    updatedAt: "2026-08-03T09:10:00+08:00",
-  },
-];
-
-/** 按套件出现顺序写入 sortOrder，并补齐可选字段。 */
-const INITIAL_CASES: TestCase[] = (() => {
-  const counters = new Map<string, number>();
-  return RAW_CASE_SEEDS.map((item) => {
-    const order = counters.get(item.suiteId) ?? 0;
-    counters.set(item.suiteId, order + 1);
-    return {
-      ...item,
-      inputMode: item.inputMode ?? DEFAULT_INPUT_MODE,
-      turns: cloneTurns(item.turns),
-      overallCriteria: cloneOverallCriteria(item.overallCriteria),
-      sortOrder: order,
-    };
-  });
-})();
-
-/** 会话内可变的 mock 仓库（刷新页面会重置） */
-let suites = [...INITIAL_SUITES];
-let cases = structuredClone(INITIAL_CASES);
-let suiteSeq = 4;
-let caseSeq = 300;
-
-function nowIso() {
-  return new Date().toISOString();
-}
-
-function nextSuiteId() {
-  const id = `suite_${String(suiteSeq).padStart(3, "0")}`;
-  suiteSeq += 1;
-  return id;
-}
-
-function nextCaseId() {
-  const id = `tc_${String(caseSeq).padStart(3, "0")}`;
-  caseSeq += 1;
-  return id;
-}
-
-export function listTestSuites(): TestSuite[] {
-  return [...suites].sort(
-    (a, b) => new Date(b.updatedAt).getTime() - new Date(a.updatedAt).getTime(),
-  );
-}
-
-export function getTestSuite(id: string): TestSuite | null {
-  return suites.find((item) => item.id === id) ?? null;
-}
-
-function getTestCase(id: string): TestCase | null {
-  return cases.find((item) => item.id === id) ?? null;
-}
-
-export function listTestCases(suiteId: string): TestCase[] {
-  return cases
-    .filter((item) => item.suiteId === suiteId)
-    .sort(
-      (a, b) =>
-        a.sortOrder - b.sortOrder || a.name.localeCompare(b.name, "zh-CN"),
-    );
-}
-
-export function suiteCaseStats(suiteId: string): {
-  total: number;
-  passed: number;
-  run: number;
-} {
-  const items = cases.filter((item) => item.suiteId === suiteId);
-  const run = items.filter((item) => item.lastResult !== "not_run");
-  const passed = run.filter((item) => item.lastResult === "pass");
-  return { total: items.length, passed: passed.length, run: run.length };
-}
-
-export function createTestSuite(input: {
-  name: string;
-  description: string;
-  assistantName: string;
-}): TestSuite {
-  const suite: TestSuite = {
-    id: nextSuiteId(),
-    name: input.name.trim(),
-    description: input.description.trim(),
-    assistantName: input.assistantName.trim() || "未关联助手",
-    updatedAt: nowIso(),
-  };
-  suites = [suite, ...suites];
-  return suite;
-}
-
-export function updateTestSuite(
-  id: string,
-  patch: Partial<Pick<TestSuite, "name" | "description" | "assistantName">>,
-): TestSuite | null {
-  const index = suites.findIndex((item) => item.id === id);
-  if (index < 0) return null;
-  const next = {
-    ...suites[index],
-    ...patch,
-    updatedAt: nowIso(),
-  };
-  suites = [...suites.slice(0, index), next, ...suites.slice(index + 1)];
-  return next;
-}
-
-export function removeTestSuite(id: string): boolean {
-  const before = suites.length;
-  suites = suites.filter((item) => item.id !== id);
-  cases = cases.filter((item) => item.suiteId !== id);
-  return suites.length < before;
-}
-
-/** 复制测试集及其全部用例；名称加「（副本）」 */
-export function duplicateTestSuite(id: string): TestSuite | null {
-  const source = getTestSuite(id);
-  if (!source) return null;
-
-  const copied = createTestSuite({
-    name: `${source.name}（副本）`,
-    description: source.description,
-    assistantName: source.assistantName,
-  });
-
-  const sourceCases = listTestCases(id);
-  const clonedCases: TestCase[] = sourceCases.map((item, index) => {
-    const turns = cloneTurnsWithNewIds(item.turns);
-    return {
-      id: nextCaseId(),
-      suiteId: copied.id,
-      name: item.name,
-      description: item.description,
-      inputMode: item.inputMode,
-      lastResult: "not_run" as const,
-      contextTurns: item.contextTurns.map((turn) => ({ ...turn })),
-      turns,
-      overallCriteria: cloneOverallCriteria(item.overallCriteria),
-      sortOrder: index,
-      updatedAt: nowIso(),
-    };
-  });
-  cases = [...cases, ...clonedCases];
-  return copied;
-}
-
-export function createTestCase(input: {
-  suiteId: string;
-  name?: string;
-  description?: string;
-}): TestCase | null {
-  if (!getTestSuite(input.suiteId)) return null;
-  const maxOrder = cases
-    .filter((item) => item.suiteId === input.suiteId)
-    .reduce((max, item) => Math.max(max, item.sortOrder), -1);
-  const turns = [createEmptyFixedInputTurn()];
-  const item: TestCase = {
-    id: nextCaseId(),
-    suiteId: input.suiteId,
-    name: (input.name ?? "未命名用例").trim() || "未命名用例",
-    description: (input.description ?? "").trim(),
-    inputMode: DEFAULT_INPUT_MODE,
-    lastResult: "not_run",
-    contextTurns: [],
-    turns,
-    overallCriteria: [],
-    sortOrder: maxOrder + 1,
-    updatedAt: nowIso(),
-  };
-  cases = [...cases, item];
-  updateTestSuite(input.suiteId, {});
-  return item;
-}
-
-/** 复制单条用例；名称加「（副本）」，排在同套件末尾 */
-export function duplicateTestCase(id: string): TestCase | null {
-  const source = getTestCase(id);
-  if (!source || !getTestSuite(source.suiteId)) return null;
-
-  const maxOrder = cases
-    .filter((item) => item.suiteId === source.suiteId)
-    .reduce((max, item) => Math.max(max, item.sortOrder), -1);
-
-  const turns = cloneTurnsWithNewIds(source.turns);
-  const copied: TestCase = {
-    id: nextCaseId(),
-    suiteId: source.suiteId,
-    name: `${source.name}（副本）`,
-    description: source.description,
-    inputMode: source.inputMode,
-    lastResult: "not_run",
-    contextTurns: source.contextTurns.map((turn) => ({ ...turn })),
-    turns,
-    overallCriteria: cloneOverallCriteria(source.overallCriteria),
-    sortOrder: maxOrder + 1,
-    updatedAt: nowIso(),
-  };
-  cases = [...cases, copied];
-  updateTestSuite(source.suiteId, {});
-  return copied;
-}
-
-type TestCasePatch = Partial<
-  Pick<
-    TestCase,
-    | "name"
-    | "description"
-    | "inputMode"
-    | "contextTurns"
-    | "turns"
-    | "overallCriteria"
-    | "lastResult"
-  >
->;
-
-export function updateTestCase(
-  id: string,
-  patch: TestCasePatch,
-): TestCase | null {
-  const index = cases.findIndex((item) => item.id === id);
-  if (index < 0) return null;
-  const current = cases[index];
-  const nextTurns =
-    patch.turns !== undefined
-      ? cloneTurns(
-          patch.turns.length > 0 ? patch.turns : [createEmptyFixedInputTurn()],
-        )
-      : current.turns;
-  const nextOverallCriteria =
-    patch.overallCriteria !== undefined
-      ? cloneOverallCriteria(patch.overallCriteria)
-      : cloneOverallCriteria(current.overallCriteria);
-  const next: TestCase = {
-    ...current,
-    ...patch,
-    turns: nextTurns,
-    overallCriteria: nextOverallCriteria,
-    updatedAt: nowIso(),
-  };
-  cases = [...cases.slice(0, index), next, ...cases.slice(index + 1)];
-  updateTestSuite(next.suiteId, {});
-  return next;
-}
-
-export function removeTestCase(id: string): boolean {
-  const existing = cases.find((item) => item.id === id);
-  if (!existing) return false;
-  cases = cases.filter((item) => item.id !== id);
-  renumberSortOrder(existing.suiteId);
-  updateTestSuite(existing.suiteId, {});
-  return true;
-}
-
-/** 批量删除；返回受影响的 suiteId（若有） */
-export function removeTestCases(ids: string[]): string | null {
-  const idSet = new Set(ids);
-  const affected = cases.find((item) => idSet.has(item.id));
-  if (!affected) return null;
-  const suiteId = affected.suiteId;
-  cases = cases.filter((item) => !idSet.has(item.id));
-  renumberSortOrder(suiteId);
-  updateTestSuite(suiteId, {});
-  return suiteId;
-}
-
-/**
- * 按给定 id 顺序写回 sortOrder（应包含该套件全部用例 id）。
- * 用于拖拽结束后持久化顺序。
- */
-export function reorderTestCases(suiteId: string, orderedIds: string[]): void {
-  const orderMap = new Map(orderedIds.map((id, index) => [id, index]));
-  cases = cases.map((item) => {
-    if (item.suiteId !== suiteId) return item;
-    const nextOrder = orderMap.get(item.id);
-    if (nextOrder === undefined) return item;
-    return { ...item, sortOrder: nextOrder };
-  });
-  updateTestSuite(suiteId, {});
-}
-
-function renumberSortOrder(suiteId: string) {
-  const ordered = cases
-    .filter((item) => item.suiteId === suiteId)
-    .sort((a, b) => a.sortOrder - b.sortOrder);
-  const orderMap = new Map(ordered.map((item, index) => [item.id, index]));
-  cases = cases.map((item) => {
-    if (item.suiteId !== suiteId) return item;
-    const nextOrder = orderMap.get(item.id);
-    return nextOrder === undefined ? item : { ...item, sortOrder: nextOrder };
-  });
-}
-
 export function formatUpdatedAt(value?: string | null) {
  if (!value) return "—";
  const date = new Date(value);
--- a/frontend/src/lib/api.ts
+++ b/frontend/src/lib/api.ts
@@ -6,6 +6,15 @@
 */

 import { loginPathWithReturnTo } from "@/lib/auth-redirect";
+import type { BatchRunSnapshot } from "@/data/batch-run";
+import type {
+  ContextTurn,
+  FixedInputTurn,
+  OverallCriterion,
+  TestCase,
+  TestCaseInputMode,
+  TestSuite,
+} from "@/data/test-suites";

 const LOOPBACK_HOSTS = new Set(["localhost", "127.0.0.1", "[::1]"]);

@@ -769,3 +778,94 @@ export const webrtcApi = {
  iceServers: () =>
    request<{ iceServers: IceServerConfig[] }>("/api/webrtc/ice-servers"),
 };
+
+// ---------- 测试用例与批量运行 ----------
+export type TestSuiteWrite = {
+  name: string;
+  description: string;
+};
+
+export type TestCaseWrite = {
+  name: string;
+  description: string;
+  inputMode: TestCaseInputMode;
+  contextTurns: ContextTurn[];
+  turns: FixedInputTurn[];
+  overallCriteria: OverallCriterion[];
+};
+
+export const testSuitesApi = {
+  list: () => request<TestSuite[]>("/api/test-suites"),
+  get: (id: string) => request<TestSuite>(`/api/test-suites/${id}`),
+  create: (body: TestSuiteWrite) =>
+    request<TestSuite>("/api/test-suites", {
+      method: "POST",
+      body: JSON.stringify(body),
+    }),
+  update: (id: string, body: TestSuiteWrite) =>
+    request<TestSuite>(`/api/test-suites/${id}`, {
+      method: "PUT",
+      body: JSON.stringify(body),
+    }),
+  duplicate: (id: string) =>
+    request<TestSuite>(`/api/test-suites/${id}/duplicate`, { method: "POST" }),
+  remove: (id: string) =>
+    request<{ ok: boolean }>(`/api/test-suites/${id}`, { method: "DELETE" }),
+};
+
+export const testCasesApi = {
+  list: (suiteId?: string) =>
+    request<TestCase[]>(
+      `/api/test-cases${suiteId ? `?suiteId=${encodeURIComponent(suiteId)}` : ""}`,
+    ),
+  get: (id: string) => request<TestCase>(`/api/test-cases/${id}`),
+  create: (suiteId: string, body: TestCaseWrite) =>
+    request<TestCase>(`/api/test-suites/${suiteId}/cases`, {
+      method: "POST",
+      body: JSON.stringify(body),
+    }),
+  update: (id: string, body: TestCaseWrite) =>
+    request<TestCase>(`/api/test-cases/${id}`, {
+      method: "PUT",
+      body: JSON.stringify(body),
+    }),
+  duplicate: (id: string) =>
+    request<TestCase>(`/api/test-cases/${id}/duplicate`, { method: "POST" }),
+  remove: (id: string) =>
+    request<{ ok: boolean }>(`/api/test-cases/${id}`, { method: "DELETE" }),
+  bulkRemove: (caseIds: string[]) =>
+    request<{ ok: boolean; deleted: number }>("/api/test-cases/bulk-delete", {
+      method: "POST",
+      body: JSON.stringify({ caseIds }),
+    }),
+  reorder: (suiteId: string, caseIds: string[]) =>
+    request<{ ok: boolean }>(`/api/test-suites/${suiteId}/case-order`, {
+      method: "PUT",
+      body: JSON.stringify({ caseIds }),
+    }),
+};
+
+export const batchRunsApi = {
+  create: (body: {
+    assistantId: string;
+    evaluatorModelResourceId: string;
+    caseIds: string[];
+    title?: string;
+    config: {
+      concurrency: number;
+      timeoutSecs: number;
+      failureStrategy: "continue" | "stop_on_fail";
+      errorRetryCount: number;
+      errorStrategy: "continue" | "stop_on_error";
+    };
+  }) =>
+    request<BatchRunSnapshot>("/api/test-runs", {
+      method: "POST",
+      body: JSON.stringify(body),
+    }),
+  get: (id: string) => request<BatchRunSnapshot>(`/api/test-runs/${id}`),
+  cancel: (id: string) =>
+    request<BatchRunSnapshot>(`/api/test-runs/${id}/cancel`, {
+      method: "POST",
+    }),
+};