feature: 增加功能导出详细报告

2026-07-19 00:29:06 +08:00
parent 789d4fc4e9
commit 12d8293250
12 changed files with 92 additions and 15 deletions
@@ -23,6 +23,8 @@ docs/               # UI、交互、架构、API、测试与部署文档

 认证支持 access/refresh token 原子轮换、活跃用户静默续约、标签页内/跨标签页 single-flight，以及业务请求 401 后的一次续约与重放。跨标签页协调使用浏览器 Web Locks API。

+终态测试运行支持按执行状态和仲裁结论筛选并导出 Markdown 详细报告。
+
 ## 本地开发

 需要 Node.js 20–24 和 Corepack。
@@ -9,7 +9,7 @@
 | 提供商 | 列表、详情、创建、修改、删除 | `GET/POST /providers`, `GET/PATCH/DELETE /providers/{id}` |
 | 提供商 | 连通检查、模型发现 | `POST /providers/{id}/check`, `GET /providers/{id}/models` |
 | 运行 | 列表、创建、详情、取消、删除 | `GET/POST /runs`, `GET/PATCH/DELETE /runs/{id}` |
-| 运行 | 恢复、重试错误、单条重试、逐条结果 | `POST /runs/{id}/resume`, `POST /runs/{id}/retry-errors`, `POST /runs/{id}/results/{execution_id}/retry`, `GET /runs/{id}/results` |
+| 运行 | 恢复、重试错误、单条重试、逐条结果、导出详细报告 | `POST /runs/{id}/resume`, `POST /runs/{id}/retry-errors`, `POST /runs/{id}/results/{execution_id}/retry`, `GET /runs/{id}/results`, `GET /runs/{id}/export` |
 | 报告 | 列表、详情 | `GET /reports`, `GET /reports/{run_id}` |
 | 系统 | 健康状态 | `GET /health` |

@@ -19,6 +19,7 @@
 - 运行详情必须满足 `processed_count = completed_count + error_count`；`error_count` 仅统计执行异常，不包含 `verdict = fail`；终态必须有 `poll_after_seconds = 0`。
 - 逐条结果使用 `execution_status`（`completed | error`）表示调用是否完成，使用可空 `verdict`（`pass | fail | needs_human_review | judge_format_error | null`）表示仲裁结论；前端不读取旧 `status`。
 - 逐条结果的执行 ID 搜索直接使用 `GET /runs/{id}/results` 已返回的 `execution_id` 在 UI 过滤，不新增请求参数或后端端点。
+- `GET /runs/{id}/export` 仅允许终态 Run，返回 `text/markdown` 文件而非 JSON；可组合传入 `execution_status=completed|error` 与 `verdict=pass|fail|needs_human_review|judge_format_error|none`，其中 `none` 表示未评价。前端复用结果筛选器，但执行 ID 搜索不参与导出。
 - 报告的 `summary.execution_statuses` 与 `summary.verdicts` 分开统计；`by_mode` 下也分别包含 `execution_statuses` 和 `verdicts`。
 - `POST /runs/{id}/retry-errors` 只重试 `execution_status = error`，不重试任何仲裁结论。
 - `POST /runs/{id}/results/{execution_id}/retry` 无请求体，沿用原 Run 的测试范围和自动仲裁配置，并以相同 `execution_id` 替换旧结果；仅 `completed / completed_with_errors` 状态允许提交。
@@ -29,4 +30,4 @@
 - access token 剩余不超过 30 分钟且用户最近 30 分钟有真实操作时，`POST /auth/refresh` 携带同一共享生成器产生的 UUID v4 `Idempotency-Key`。
 - 网络错误和 5xx 最多尝试 3 次，且一直复用原 refresh token 和原 key；409 保留凭据并等待 `refresh_after`。
 - 普通业务请求首次 401 共享 single-flight 续约，成功后仅重放一次；只有 refresh 本身返回 401 才清除会话。
- HTTP 204 按无响应体处理；403/404/409/422/502 映射为可操作的用户消息。
+- HTTP 204 按无响应体处理；Markdown 成功响应按 Blob 读取；错误响应仍按 JSON 归一化。403/404/409/422/502 映射为可操作的用户消息。
@@ -4,6 +4,8 @@

 UI 页面只调用 `services/`；service 先用 `schemas/` 中的 Zod schema 校验请求，再由 `api/client.ts` 发起请求并用 Zod 校验响应。跨字段业务不变量也在 schema 层完成。错误统一进入 `errors/`，日志进入 `logging/` 并在输出前递归脱敏。

+API 客户端默认读取并校验 JSON；文件端点显式选择 Blob 响应类型，同时复用 Bearer Token、401 单次续约重放和 JSON 错误归一化。下载由页面使用浏览器原生 Object URL 完成，不引入额外下载依赖。
+
 创建运行和 Token 续约统一通过 `idempotencyKey.ts` 生成 UUID v4 幂等键：安全上下文优先使用浏览器原生 `crypto.randomUUID()`，HTTP 或旧浏览器回退到 `crypto.getRandomValues()`；不使用可预测的 `Math.random()`。

 ```text
@@ -8,10 +8,11 @@
 3. 管理员在模型配置页建立 target/judge 配置，可查询模型并执行最小连通性检查。
 4. 用户选择 smoke/all 与是否自动裁判，确认后创建运行。客户端自动携带 UUID 幂等键，避免重复提交。
 5. 详情页展示探测、执行、裁判和完成阶段，按后端 `poll_after_seconds` 轮询。
-6. 运行中可取消；中断后可恢复；局部错误可只重试错误样例；终态后可查看逐条结果、汇总报告或确认删除。
+6. 运行中可取消；中断后可恢复；局部错误可只重试错误样例；终态后可查看逐条结果、汇总报告、导出 Markdown 详细报告或确认删除。
   `cancelled / failed / pending / probing / running` 可确认后“恢复运行”；`completed_with_errors` 且 `error_count > 0` 可确认后“重试错误（N）”。提交期间按钮禁用，成功提示跳过或重试数量；409 会刷新详情以校正操作状态。
 7. 逐条结果可按执行状态、仲裁结论筛选，并可按执行 ID 搜索；三项条件组合生效，执行 ID 支持忽略大小写的片段匹配。
 8. `completed / completed_with_errors` 的每条结果可确认后单独“重新执行”。提交中仅该按钮显示“提交中…”，受理后该 Run 的全部单条重试按钮锁定并显示目标执行 ID“重新执行中…”，直到 Run 再次进入终态。目标结果会进入跟踪状态并置顶，暂时绕过执行状态、仲裁结论和执行 ID 筛选，用户可手动停止跟踪；原筛选值始终保留。
+9. “导出详细报告”位于逐条结果筛选栏，复用当前执行状态和仲裁结论筛选，不占用页头运行控制区；“全部”不传参数，“未仲裁”映射为 `verdict=none`，执行 ID 搜索不参与导出。非终态按钮禁用，导出期间显示加载状态并阻止重复提交，文件名为 `run_{run_id}_results.md`；409 会刷新运行详情。

 ## 模型等待交互

@@ -2,7 +2,7 @@

 ## 测试

-`tests/unit` 覆盖脱敏、错误归一化、主题持久化、API 响应校验、认证原子存储、single-flight 续约、HTTP/HTTPS UUID v4 幂等键生成、幂等重试、401 单次重放、运行数据不变量、结果语义标识和运行进入终态时的最终结果同步。`tests/integration` 验证主题切换、逐条结果筛选与搜索、单条重试结果跨筛选置顶跟踪等页面交互，以及所有 service 到后端路由的映射。
+`tests/unit` 覆盖脱敏、错误归一化、主题持久化、JSON/Blob API 响应、认证原子存储、single-flight 续约、HTTP/HTTPS UUID v4 幂等键生成、幂等重试、401 单次重放、运行数据不变量、结果语义标识和运行进入终态时的最终结果同步。`tests/integration` 验证主题切换、逐条结果筛选与搜索、终态 Markdown 导出、单条重试结果跨筛选置顶跟踪等页面交互，以及所有 service 到后端路由的映射。

 ```bash
 pnpm test
@@ -49,6 +49,7 @@
 - 运行详情：阶段、进度、计数和当前样例位于首屏，结果和报告位于下方标签页。
 - 结果卡片：顶部使用统一 Tag 明确标识“类型 / 模式 / 执行状态 / 仲裁结论”，将后端原始值转为中文语义。执行状态和仲裁结论使用两个独立筛选器，不混合 `completed/error` 与 `pass/fail`；执行 ID 搜索与两个筛选器组合生效，并支持忽略大小写的片段匹配。所有标签使用相同的 `8px` 圆角；结论只增加字重和状态点。通过为绿色、未通过或错误为红色、待复核为橙色，文字始终保留。
 - 结果卡片在 Run 完成或部分错误时提供“重新执行”；使用 Ant Design 确认框说明结果将被替换。提交、运行中状态以按钮文字和信息提示表达，不只依赖颜色；Run 非终态时同 Run 的全部单条重试按钮禁用。重试目标使用信息提示说明跟踪状态并置顶显示，不修改用户已有筛选；提示提供“停止跟踪”按钮。
+- Run 详情页在逐条结果筛选栏提供 Markdown 详细报告导出，复用执行状态和仲裁结论筛选，不复制第二组筛选器，也不占用页头运行控制区；非终态保留禁用按钮以说明能力，加载状态使用按钮文字和图标表达。执行 ID 搜索只影响页面结果，不影响导出。

 ## 响应式与可访问性

@@ -5,6 +5,7 @@ import { logger } from '../logging'
 export type RequestOptions<T> = Omit<RequestInit, 'body'> & {
  body?: unknown
  schema?: ZodType<T>
+  responseType?: 'json' | 'blob'
 }

 export class ApiClient {
@@ -22,26 +23,27 @@ export class ApiClient {
  private async execute<T>(path: string, options: RequestOptions<T>, replayed: boolean): Promise<T> {
    const token = this.getToken()
    const headers = new Headers(options.headers)
-    headers.set('Accept', 'application/json')
+    headers.set('Accept', options.responseType === 'blob' ? 'text/markdown' : 'application/json')
    if (options.body !== undefined) headers.set('Content-Type', 'application/json')
    if (token) headers.set('Authorization', `Bearer ${token}`)

    try {
+      const { responseType, schema, ...requestOptions } = options
      const response = await fetch(`${this.baseUrl}${path}`, {
-        ...options,
+        ...requestOptions,
        headers,
        body: options.body === undefined ? undefined : JSON.stringify(options.body),
      })
      if (response.status === 204) return undefined as T
-      const body = await response.json().catch(() => null)
+      const body = response.ok && responseType === 'blob' ? await response.blob() : await response.json().catch(() => null)
      if (!response.ok) {
        if (response.status === 401 && !replayed && await this.refreshAuth?.(true)) {
          return this.execute(path, options, true)
        }
        throw { status: response.status, body }
      }
-      if (!options.schema) return body as T
-      const parsed = options.schema.safeParse(body)
+      if (!schema) return body as T
+      const parsed = schema.safeParse(body)
      if (!parsed.success) {
        throw new AppError('服务响应格式异常', 'validation', response.status, 'INVALID_RESPONSE', undefined, {
          issues: parsed.error.issues,
@@ -43,6 +43,24 @@ export function RunDetailPage() {
      message.error(status === undefined || (typeof status === 'number' && status >= 500) ? '重新执行提交失败，请稍后重试' : error.message)
    },
  })
+  const exportReport = useMutation({
+    mutationFn: () => services.runs.export(id, {
+      executionStatus: executionFilter === 'all' ? undefined : executionFilter as 'completed' | 'error',
+      verdict: verdictFilter === 'all' ? undefined : verdictFilter === 'unjudged' ? 'none' : verdictFilter as 'pass' | 'fail' | 'needs_human_review' | 'judge_format_error',
+    }),
+    onSuccess: (blob) => {
+      const url = URL.createObjectURL(blob)
+      const link = document.createElement('a')
+      link.href = url
+      link.download = `run_${id}_results.md`
+      link.click()
+      URL.revokeObjectURL(url)
+    },
+    onError: (error) => {
+      message.error(error.message)
+      if ('status' in error && error.status === 409) refresh()
+    },
+  })
  const action = useMutation({
    mutationFn: async (name: 'cancel' | 'resume' | 'retry' | 'delete') => {
      if (name === 'cancel') return services.runs.cancel(id)
@@ -83,16 +101,16 @@ export function RunDetailPage() {
        <div className="stat-strip"><span><strong>{data.processed_count}</strong>已处理</span><span><strong>{data.completed_count}</strong>已完成</span><span><strong>{data.error_count}</strong>执行错误</span><span><strong>{data.selected_count}</strong>总样例</span></div>
      </Card>
      <Tabs items={[
-        { key: 'results', label: `逐条结果 (${filtered.length} / ${results.data?.length || 0})`, children: <ResultsPanel rows={filtered} executionFilter={executionFilter} verdictFilter={verdictFilter} executionIdQuery={executionIdQuery} onExecutionFilter={setExecutionFilter} onVerdictFilter={setVerdictFilter} onExecutionIdQuery={setExecutionIdQuery} loading={results.isLoading} canRetry={['completed', 'completed_with_errors'].includes(data.status)} retryingExecutionId={activeRetryingExecutionId} followedExecutionId={followedExecutionId} submittingExecutionId={retryResult.isPending ? retryResult.variables : null} onStopFollowing={() => setFollowedExecutionId(null)} onRetry={(executionId) => Modal.confirm({ title: `确定重新执行 ${executionId} 吗？`, content: '当前测试结果将被新结果替换。', onOk: () => retryResult.mutate(executionId) })} /> },
+        { key: 'results', label: `逐条结果 (${filtered.length} / ${results.data?.length || 0})`, children: <ResultsPanel rows={filtered} executionFilter={executionFilter} verdictFilter={verdictFilter} executionIdQuery={executionIdQuery} onExecutionFilter={setExecutionFilter} onVerdictFilter={setVerdictFilter} onExecutionIdQuery={setExecutionIdQuery} loading={results.isLoading} canRetry={['completed', 'completed_with_errors'].includes(data.status)} canExport={data.terminal} exporting={exportReport.isPending} retryingExecutionId={activeRetryingExecutionId} followedExecutionId={followedExecutionId} submittingExecutionId={retryResult.isPending ? retryResult.variables : null} onExport={() => exportReport.mutate()} onStopFollowing={() => setFollowedExecutionId(null)} onRetry={(executionId) => Modal.confirm({ title: `确定重新执行 ${executionId} 吗？`, content: '当前测试结果将被新结果替换。', onOk: () => retryResult.mutate(executionId) })} /> },
        { key: 'report', label: '汇总报告', children: report.data ? <ReportSummary summary={report.data.summary} /> : <Empty description={data.terminal ? '暂无报告' : '运行结束后生成报告'} /> },
      ]} />
    </>
  )
 }

-function ResultsPanel({ rows, executionFilter, verdictFilter, executionIdQuery, onExecutionFilter, onVerdictFilter, onExecutionIdQuery, loading, canRetry, retryingExecutionId, followedExecutionId, submittingExecutionId, onStopFollowing, onRetry }: { rows: ResultItem[]; executionFilter: string; verdictFilter: string; executionIdQuery: string; onExecutionFilter: (value: string) => void; onVerdictFilter: (value: string) => void; onExecutionIdQuery: (value: string) => void; loading: boolean; canRetry: boolean; retryingExecutionId: string | null; followedExecutionId: string | null; submittingExecutionId: string | null; onStopFollowing: () => void; onRetry: (executionId: string) => void }) {
+function ResultsPanel({ rows, executionFilter, verdictFilter, executionIdQuery, onExecutionFilter, onVerdictFilter, onExecutionIdQuery, loading, canRetry, canExport, exporting, retryingExecutionId, followedExecutionId, submittingExecutionId, onExport, onStopFollowing, onRetry }: { rows: ResultItem[]; executionFilter: string; verdictFilter: string; executionIdQuery: string; onExecutionFilter: (value: string) => void; onVerdictFilter: (value: string) => void; onExecutionIdQuery: (value: string) => void; loading: boolean; canRetry: boolean; canExport: boolean; exporting: boolean; retryingExecutionId: string | null; followedExecutionId: string | null; submittingExecutionId: string | null; onExport: () => void; onStopFollowing: () => void; onRetry: (executionId: string) => void }) {
  if (loading) return <LoadingState />
-  return <div className="results-panel"><div className="action-group"><Select aria-label="执行状态" value={executionFilter} onChange={onExecutionFilter} labelRender={({ label }) => `执行状态：${label}`} options={[{ value: 'all', label: '全部' }, { value: 'completed', label: '已完成' }, { value: 'error', label: '执行错误' }]} /><Select aria-label="仲裁结论" value={verdictFilter} onChange={onVerdictFilter} labelRender={({ label }) => `仲裁结论：${label}`} options={[{ value: 'all', label: '全部' }, { value: 'pass', label: '通过' }, { value: 'fail', label: '不通过' }, { value: 'needs_human_review', label: '人工复核' }, { value: 'judge_format_error', label: '裁判格式错误' }, { value: 'unjudged', label: '未仲裁' }]} /><Input type="search" allowClear aria-label="执行 ID" placeholder="根据执行 ID 搜索，如 R0029" value={executionIdQuery} onChange={(event) => onExecutionIdQuery(event.target.value)} /></div>
+  return <div className="results-panel"><div className="action-group"><Select aria-label="执行状态" value={executionFilter} onChange={onExecutionFilter} labelRender={({ label }) => `执行状态：${label}`} options={[{ value: 'all', label: '全部' }, { value: 'completed', label: '已完成' }, { value: 'error', label: '执行错误' }]} /><Select aria-label="仲裁结论" value={verdictFilter} onChange={onVerdictFilter} labelRender={({ label }) => `仲裁结论：${label}`} options={[{ value: 'all', label: '全部' }, { value: 'pass', label: '通过' }, { value: 'fail', label: '不通过' }, { value: 'needs_human_review', label: '人工复核' }, { value: 'judge_format_error', label: '裁判格式错误' }, { value: 'unjudged', label: '未仲裁' }]} /><Input type="search" allowClear aria-label="执行 ID" placeholder="根据执行 ID 搜索，如 R0029" value={executionIdQuery} onChange={(event) => onExecutionIdQuery(event.target.value)} /><Button type="primary" disabled={!canExport} loading={exporting} title={canExport ? undefined : '运行结束后可导出'} onClick={onExport}>导出详细报告</Button></div>
    {followedExecutionId && <Alert type="info" showIcon message={retryingExecutionId ? `${followedExecutionId} 重新执行中…` : `正在跟踪 ${followedExecutionId}；即使不符合当前筛选也会置顶显示。`} action={<Button size="small" onClick={onStopFollowing}>停止跟踪</Button>} />}
    {!rows.length ? <Empty description={followedExecutionId ? '正在等待跟踪结果…' : '暂无符合条件的结果'} /> : rows.map((row) => <Card key={row.execution_id} className="result-card"><div className="result-title"><strong>{row.execution_id}</strong><span className="result-tags" aria-label="结果标识">{resultTags({ caseKind: row.case_kind, mode: row.interaction_mode, executionStatus: row.execution_status, verdict: row.verdict }).map((tag) => <Tag key={tag.label} className={`result-tag-${tag.color}${tag.emphasis ? ' result-verdict' : ''}`}>{tag.label}</Tag>)}</span><Button disabled={!canRetry || Boolean(retryingExecutionId) || Boolean(submittingExecutionId)} loading={submittingExecutionId === row.execution_id} onClick={() => onRetry(row.execution_id)} aria-label={submittingExecutionId === row.execution_id ? '提交中…' : retryingExecutionId === row.execution_id ? '重新执行中…' : `重新执行 ${row.execution_id}`}>{submittingExecutionId === row.execution_id ? '提交中…' : retryingExecutionId === row.execution_id ? '重新执行中…' : '重新执行'}</Button></div><div className="result-columns"><section><h3>模型输入</h3><pre>{JSON.stringify(row.model_input, null, 2)}</pre></section><section><h3>模型回复</h3><pre>{row.model_response || '-'}</pre></section></div>{Object.keys(row.judge_result).length > 0 && <section className="judge-result"><h3>Judge 结果</h3><pre>{JSON.stringify(row.judge_result, null, 2)}</pre></section>}{row.error_message && <Alert type="error" message={row.error_message} />}</Card>)}</div>
 }
@@ -2,6 +2,8 @@ import type { ApiClient } from '../api/client'
 import { createIdempotencyKey } from '../idempotencyKey'
 import { resultListSchema, retryErrorsSchema, retryResultSchema, runDetailSchema, runListSchema, runResponseSchema, startRunSchema, resumeRunSchema, type StartRunInput } from '../schemas/runs'

+type ExportFilters = { executionStatus?: 'completed' | 'error'; verdict?: 'pass' | 'fail' | 'needs_human_review' | 'judge_format_error' | 'none' }
+
 export const runService = (api: ApiClient) => ({
  list: (limit = 50) => api.request(`/runs?limit=${limit}`, { schema: runListSchema }),
  get: (id: number) => api.request(`/runs/${id}`, { schema: runDetailSchema }),
@@ -13,4 +15,11 @@ export const runService = (api: ApiClient) => ({
  retryErrors: (id: number) => api.request(`/runs/${id}/retry-errors`, { method: 'POST', schema: retryErrorsSchema }),
  retryResult: (id: number, executionId: string) => api.request(`/runs/${id}/results/${encodeURIComponent(executionId)}/retry`, { method: 'POST', schema: retryResultSchema }),
  results: (id: number) => api.request(`/runs/${id}/results`, { schema: resultListSchema }),
+  export: (id: number, filters: ExportFilters = {}) => {
+    const params = new URLSearchParams()
+    if (filters.executionStatus) params.set('execution_status', filters.executionStatus)
+    if (filters.verdict) params.set('verdict', filters.verdict)
+    const query = params.toString()
+    return api.request<Blob>(`/runs/${id}/export${query ? `?${query}` : ''}`, { responseType: 'blob' })
+  },
 })
@@ -16,6 +16,7 @@ const mocks = vi.hoisted(() => ({
  retryErrors: vi.fn(),
  retryResult: vi.fn(),
  results: vi.fn(),
+  exportRun: vi.fn(),
  reports: vi.fn(),
  report: vi.fn(),
 }))
@@ -24,7 +25,7 @@ vi.mock('../../src/services', () => ({ services: {
  auth: { me: mocks.me, login: vi.fn(), logout: vi.fn(), changePassword: vi.fn(), listUsers: vi.fn().mockResolvedValue([]) },
  health: { get: mocks.health },
  providers: { list: mocks.providers },
-  runs: { list: mocks.runs, get: mocks.run, resume: mocks.resume, retryErrors: mocks.retryErrors, retryResult: mocks.retryResult, results: mocks.results },
+  runs: { list: mocks.runs, get: mocks.run, resume: mocks.resume, retryErrors: mocks.retryErrors, retryResult: mocks.retryResult, results: mocks.results, export: mocks.exportRun },
  reports: { list: mocks.reports, get: mocks.report },
 } }))

@@ -56,6 +57,7 @@ beforeEach(() => {
  mocks.retryErrors.mockResolvedValue({ run_id: 1, status: 'pending', selected_count: 100, retry_count: 5 })
  mocks.retryResult.mockResolvedValue({ run_id: 1, status: 'pending', selected_count: 1, execution_id: 'R0049' })
  mocks.results.mockResolvedValue([])
+  mocks.exportRun.mockResolvedValue(new Blob(['# report'], { type: 'text/markdown' }))
  mocks.reports.mockResolvedValue([])
  mocks.report.mockResolvedValue({ run_id: 1, summary: { test_result: {}, admission: { decision: 'pass', coverage: {} } } })
 })
@@ -64,6 +66,7 @@ afterEach(() => {
  cleanup()
  localStorage.clear()
  vi.clearAllMocks()
+  vi.unstubAllGlobals()
 })

 describe('page mounting and route security', () => {
@@ -122,6 +125,36 @@ describe('page mounting and route security', () => {
    expect(screen.getByText('逐条结果 (1 / 2)')).toBeInTheDocument()
  })

+  it('exports terminal run results with the active filters', async () => {
+    const createObjectURL = vi.fn(() => 'blob:report')
+    const revokeObjectURL = vi.fn()
+    class MockURL extends URL {}
+    MockURL.createObjectURL = createObjectURL
+    MockURL.revokeObjectURL = revokeObjectURL
+    vi.stubGlobal('URL', MockURL)
+    const click = vi.spyOn(HTMLAnchorElement.prototype, 'click').mockImplementation(() => undefined)
+    mount('/runs/1')
+
+    const exportButton = await screen.findByRole('button', { name: '导出详细报告' })
+    expect(exportButton.closest('.results-panel')).not.toBeNull()
+    expect(document.querySelector('.page-actions')).not.toContainElement(exportButton)
+    await userEvent.click(screen.getByRole('combobox', { name: '仲裁结论' }))
+    await userEvent.click(await screen.findByText('不通过'))
+    await userEvent.click(screen.getByRole('button', { name: '导出详细报告' }))
+
+    expect(mocks.exportRun).toHaveBeenCalledWith(1, { executionStatus: undefined, verdict: 'fail' })
+    expect(createObjectURL).toHaveBeenCalledWith(expect.any(Blob))
+    expect(click).toHaveBeenCalled()
+    expect(click.mock.contexts[0]).toMatchObject({ download: 'run_1_results.md', href: 'blob:report' })
+    expect(revokeObjectURL).toHaveBeenCalledWith('blob:report')
+  }, 20_000)
+
+  it('disables report export while the run is not terminal', async () => {
+    mocks.run.mockResolvedValueOnce({ ...run, status: 'running', terminal: false, phase: 'executing', progress_percent: 50, poll_after_seconds: 2 })
+    mount('/runs/1')
+    expect(await screen.findByRole('button', { name: '导出详细报告' })).toBeDisabled()
+  })
+
  it('searches run results by execution ID', async () => {
    mocks.results.mockResolvedValueOnce([
      { execution_id: 'R0001', case_kind: 'risk', interaction_mode: 'single_turn', execution_status: 'completed', verdict: 'fail', model_input: {}, model_response: 'response', judge_result: {}, error_message: '' },
@@ -37,13 +37,13 @@ describe('service endpoint contracts', () => {
    await providers.create({ provider_id: 'target', base_url: 'https://example.com', chat_path: '/chat', models_path: '/models', model_name: 'm', auth_type: 'none', api_key: '', auth_header: 'Authorization', auth_prefix: 'Bearer', verify_ssl: true })
    await providers.update('target', { model_name: 'm2' }); await providers.check('target'); await providers.models('target'); await providers.remove('target')
    const idempotencyKey = '00000000-0000-4000-8000-000000000001'
-    await runs.list(); await runs.get(1); await runs.start({ profile: 'smoke', auto_judge: true }, idempotencyKey); await runs.cancel(1); await runs.resume(1); await runs.retryErrors(1); await runs.retryResult(1, 'R0049'); await runs.results(1); await runs.remove(1)
+    await runs.list(); await runs.get(1); await runs.start({ profile: 'smoke', auto_judge: true }, idempotencyKey); await runs.cancel(1); await runs.resume(1); await runs.retryErrors(1); await runs.retryResult(1, 'R0049'); await runs.results(1); await runs.export(1, { executionStatus: 'error', verdict: 'none' }); await runs.remove(1)
    await reports.list(); await reports.get(1)
    expect(vi.mocked(api.request).mock.calls.map(([path]) => path)).toEqual([
      '/health', '/providers', '/providers/target', '/providers', '/providers/target',
      '/providers/target/check', '/providers/target/models', '/providers/target',
      '/runs?limit=50', '/runs/1', '/runs', '/runs/1', '/runs/1/resume',
-      '/runs/1/retry-errors', '/runs/1/results/R0049/retry', '/runs/1/results', '/runs/1', '/reports?limit=50', '/reports/1',
+      '/runs/1/retry-errors', '/runs/1/results/R0049/retry', '/runs/1/results', '/runs/1/export?execution_status=error&verdict=none', '/runs/1', '/reports?limit=50', '/reports/1',
    ])
    expect(vi.mocked(api.request).mock.calls[10][1]).toMatchObject({ headers: { 'Idempotency-Key': idempotencyKey } })
  })
@@ -17,6 +17,14 @@ describe('ApiClient', () => {
    ).rejects.toMatchObject({ kind: 'validation' })
  })

+  it('returns file responses without parsing them as JSON', async () => {
+    const blob = new Blob(['# report'], { type: 'text/markdown' })
+    const fetchMock = vi.spyOn(globalThis, 'fetch').mockResolvedValue(new Response(blob))
+
+    await expect(new ApiClient('/api', () => 'token').request('/runs/1/export', { responseType: 'blob' })).resolves.toEqual(expect.any(Blob))
+    expect(new Headers(fetchMock.mock.calls[0][1]?.headers).get('Accept')).toBe('text/markdown')
+  })
+
  it('refreshes and replays a business request once after 401', async () => {
    let token = 'old-token'
    const refresh = vi.fn(async (force?: boolean) => {