feat(data-process): 优化上传切分流程与PDF高亮预览

2026-07-24 11:28:02 +08:00
parent 33d0ed2e01
commit ad64e44860
18 changed files with 1940 additions and 395 deletions
--- a/frontend/scripts/regression-data-process-wizard.mjs
+++ b/frontend/scripts/regression-data-process-wizard.mjs
@@ -12,17 +12,19 @@ const confirmDialogPath = path.resolve(scriptDir, '../src/components/AppConfirmD
 const layoutPath = path.resolve(scriptDir, '../src/layouts/MainLayout.vue')
 const apiPath = path.resolve(scriptDir, '../src/api/modules/dataProcess.ts')
 const contractTypesPath = path.resolve(scriptDir, '../src/types/dataProcess.ts')
+const sourceUploadWorkerPath = path.join(createDir, 'useDataProcessSourceUpload.ts')
 const viewSource = await readFile(viewPath, 'utf8')
 const layoutSource = await readFile(layoutPath, 'utf8')
-const [draftSource, stateSource, generationSource, viewStyleSource, apiSource, contractTypesSource] = await Promise.all([
-  readFile(path.join(createDir, 'useDataProcessDraft.ts'), 'utf8'),
+const [stateSource, generationSource, previewBuildSource, sourceUploadWorkerSource, viewStyleSource, apiSource, contractTypesSource] = await Promise.all([
  readFile(path.join(createDir, 'dataProcessCreateState.ts'), 'utf8'),
  readFile(path.join(createDir, 'useDataProcessGeneration.ts'), 'utf8'),
+  readFile(path.join(createDir, 'useDataProcessPreviewBuild.ts'), 'utf8'),
+  readFile(sourceUploadWorkerPath, 'utf8'),
  readFile(path.join(createDir, 'data-process-create.scss'), 'utf8'),
  readFile(apiPath, 'utf8'),
  readFile(contractTypesPath, 'utf8'),
 ])
-const implementationSource = [viewSource, draftSource, stateSource, generationSource].join('\n')
+const implementationSource = [viewSource, stateSource, generationSource, previewBuildSource, sourceUploadWorkerSource].join('\n')

 assert.ok(existsSync(confirmDialogPath), '缺少公共确认弹窗组件 AppConfirmDialog')
 const confirmDialogSource = await readFile(confirmDialogPath, 'utf8')
@@ -58,9 +60,13 @@ assert.match(
  /@media \(max-width: 1100px\)[\s\S]*?\.step-title\s*\{[\s\S]*?display:\s*none[\s\S]*?\.step-item\.is-active \.step-title\s*\{[\s\S]*?display:\s*block/,
  '六步向导在中等宽度下没有收起非当前步骤标题',
 )
-assert.match(draftSource, /localStorage\.setItem\(DATA_PROCESS_DRAFT_STORAGE_KEY/, '草稿没有持久化')
-assert.match(draftSource, /localStorage\.getItem\(DATA_PROCESS_DRAFT_STORAGE_KEY\)/, '草稿没有恢复读取')
-assert.match(viewSource, /restoreDraft\(\)/, '页面没有恢复草稿')
+assert.equal(existsSync(path.join(createDir, 'useDataProcessDraft.ts')), false, '不应保留新建任务草稿模块')
+assert.doesNotMatch(
+  viewSource,
+  /useDataProcessDraft|persistDraft|restoreDraft|restoringDraft|localStorage\.(?:setItem|getItem)/,
+  '新建任务不应保存或恢复草稿',
+)
+assert.match(viewSource, /localStorage\.removeItem\('yg-data-process-create-draft'\)/, '进入新建页时应清理遗留草稿')
 assert.ok(viewSource.split('\n').length < 1000, 'DataProcessCreateView 拆分后仍超过 1000 行')
 assert.match(viewSource, /useDataProcessGeneration\(\{/, '生成流程没有拆分到独立 composable')

@@ -79,13 +85,16 @@ for (const component of expectedComponents) {

 const typesPath = path.join(createDir, 'types.ts')
 const modelPath = path.join(createDir, 'previewModel.ts')
+const pdfViewerPath = path.join(createDir, 'PdfSourceViewer.vue')
 assert.ok(existsSync(typesPath), '缺少向导类型定义')
 assert.ok(existsSync(modelPath), '缺少来源映射模型')
+assert.ok(existsSync(pdfViewerPath), '缺少 PDF 原文件预览组件')

-const [typesSource, modelSource, previewSource] = await Promise.all([
+const [typesSource, modelSource, previewSource, pdfViewerSource] = await Promise.all([
  readFile(typesPath, 'utf8'),
  readFile(modelPath, 'utf8'),
  readFile(path.join(createDir, 'PreviewCompareStep.vue'), 'utf8'),
+  readFile(pdfViewerPath, 'utf8'),
 ])

 for (const field of ['sourceStart', 'sourceEnd', 'originalContent', 'editedContent']) {
@@ -98,8 +107,8 @@ assert.doesNotMatch(modelSource, /buildPreviewItems/, '前端不应保留与后
 assert.match(viewSource, /selectedPreviewFileId/, '父页面缺少当前预览文件状态')
 assert.match(
  viewSource,
-  /buildDataProcessPreview\(taskId\.value,\s*\{[\s\S]*?source_file_ids:\s*uploadedFiles\.value\.map/,
-  '预览没有通过后端按已上传源文件构建',
+  /const \{ buildPreviewsByFile \} = useDataProcessPreviewBuild\(\)/,
+  '父页面没有通过独立 composable 构建逐文件预览',
 )
 assert.match(viewSource, /getDataProcessPreview\(taskId\.value,\s*\{ page:\s*1, page_size:\s*500 \}\)/, '预览构建后没有分页读取后端数据')
 assert.doesNotMatch(viewSource, /buildPreviewItems\(/, '创建向导仍在本地构建集成预览数据')
@@ -142,6 +151,21 @@ assert.match(previewSource, /\.editor-actions\s*\{[\s\S]*?justify-content:\s*fle
 assert.doesNotMatch(previewSource, /item-token|item-status|modifiedOnly|仅看已修改/, '切片列表不应再显示 Token 或修改状态')
 assert.match(previewSource, /\.preview-editor\s*\{[\s\S]*?flex:\s*1 1 auto[\s\S]*?overflow-y:\s*auto/, '编辑模式必须占据右侧剩余区域并可滚动')
 assert.match(previewSource, /@media \(max-width: 900px\)/, '第四步缺少窄屏上下布局')
+assert.match(previewSource, /<PdfSourceViewer[\s\S]*v-if="isPdfSource"/, 'PDF 文件没有切换到原文件查看组件')
+assert.match(previewSource, /:selected-item="selectedItem \?\? null"/, 'PDF 查看组件没有接收当前选中切片')
+assert.match(previewSource, /:data-preview-id="item\.id"/, '切片行缺少稳定的交互定位标识')
+assert.match(previewSource, /<div v-else ref="sourceViewerRef" class="source-viewer"/, '非 PDF 文件没有保留文本预览')
+assert.match(pdfViewerSource, /getDataProcessSourceRawUrl/, 'PDF 查看组件没有使用受控原文件地址')
+assert.match(pdfViewerSource, /getDataProcessPdfPages/, 'PDF 查看组件没有读取页码与全文偏移映射')
+assert.match(pdfViewerSource, /pdfjs-dist/, 'PDF 查看组件没有使用可控的 PDF.js 渲染器')
+assert.match(pdfViewerSource, /TextLayer/, 'PDF 查看组件没有渲染可定位的 PDF 文字层')
+assert.match(pdfViewerSource, /is-slice-highlighted/, 'PDF 查看组件没有实现选中切片高亮')
+assert.match(pdfViewerSource, /:data-page-number="currentPage"/, 'PDF 查看组件缺少当前物理页标识')
+assert.doesNotMatch(pdfViewerSource, /<iframe/, 'PDF 查看组件不应继续使用无法控制高亮的浏览器 iframe')
+assert.match(pdfViewerSource, /:aria-label="`PDF 预览：\$\{fileName\}`"/, 'PDF 查看器缺少可访问标题')
+assert.match(apiSource, /getDataProcessSourceRawUrl/, '前端 API 缺少 PDF 原文件预览地址')
+assert.match(apiSource, /getDataProcessPdfPages/, '前端 API 缺少 PDF 页码映射接口')
+assert.match(apiSource, /source-files\/\$\{encodeURIComponent\(fileId\)\}\/pdf-pages/, 'PDF 页码映射接口地址不正确')

 const taskSetupPath = path.join(createDir, 'TaskSetupStep.vue')
 const structuredOptionsPath = path.join(createDir, 'StructuredOptionsPanel.vue')
@@ -190,11 +214,11 @@ assert.match(viewSource, /<ModelSelectionStep\s+[\s\S]*?v-else-if="currentStepId
 assert.match(viewSource, /<SourceUploadStep\s+[\s\S]*?v-else-if="currentStepId === 'upload'"/, '第三步没有挂载独立上传组件')
 assert.match(viewSource, /if \(currentStepId\.value === 'create'\) return '继续：选择大模型'/, '第一步主按钮没有指向大模型选择')
 assert.match(viewSource, /if \(currentStepId\.value === 'model'\) return '继续：上传文件'/, '第二步主按钮没有指向上传文件')
-assert.match(viewSource, /if \(currentStepId\.value === 'upload'\) return '继续：数据预览'/, '第三步主按钮没有指向数据预览')
-const draftVersionMatch = draftSource.match(/DATA_PROCESS_DRAFT_SCHEMA_VERSION\s*=\s*(\d+)/)
-assert.ok(draftVersionMatch, '草稿缺少数字版本标识')
-assert.ok(Number(draftVersionMatch[1]) >= 7, '安全草稿格式版本不得低于 v7')
-
+assert.match(
+  viewSource,
+  /if \(currentStepId\.value === 'upload'\) \{[\s\S]*?sourceUploading\.value[\s\S]*?'正在上传'[\s\S]*?previewBuilding\.value \? '正在切分' : '继续：数据预览'/,
+  '第三步主按钮没有依次反映上传、切分状态并指向数据预览',
+)
 const nextFromCreateStart = viewSource.indexOf('async function nextFromCreate()')
 const nextFromModelStart = viewSource.indexOf('async function nextFromModel()', nextFromCreateStart)
 const nextFromUploadStart = viewSource.indexOf('function nextFromUpload()', nextFromModelStart)
@@ -213,36 +237,95 @@ assert.match(nextFromModelSource, /modelSelectionRef\.value\?\.validate\(\)/, '
 assert.match(nextFromModelSource, /createDataProcessTask\(taskPayload\(\)\)/, '大模型选择完成后没有通过真实 API 创建任务')
 assert.match(nextFromModelSource, /goToStep\('upload'\)/, '大模型选择完成后没有进入上传文件')
 assert.match(nextFromUploadSource, /uploadedFiles\.value\.length === 0/, '上传步骤继续前没有校验源数据')
-assert.match(nextFromUploadSource, /buildDataProcessPreview\(/, '上传步骤没有调用后端构建预览')
+assert.match(
+  nextFromUploadSource,
+  /const pendingFileIds = uploadedFiles\.value[\s\S]*?\.filter\(\(file\) => file\.previewStatus !== 'success' \|\| file\.previewConfigSignature !== configSignature\)[\s\S]*?\.map\(\(file\) => file\.sourceFileId\)[\s\S]*?\.filter\(\(fileId\): fileId is string => Boolean\(fileId\)\)/,
+  '上传步骤没有仅选择待处理或失败文件，无法跳过成功文件并重试失败文件',
+)
+assert.match(
+  nextFromUploadSource,
+  /await buildPreviewsByFile\(taskId\.value, pendingFileIds, \(progress\) => \{/,
+  '上传步骤没有通过独立 composable 串行构建待处理文件',
+)
+assert.match(
+  nextFromUploadSource,
+  /file\.previewStatus = progress\.status[\s\S]*?file\.previewProgress = progress\.progress[\s\S]*?file\.previewError = progress\.error/,
+  '上传步骤没有把单文件构建进度与错误回写到对应文件',
+)
+assert.doesNotMatch(
+  nextFromUploadSource,
+  /source_file_ids:\s*uploadedFiles\.value\.map/,
+  '上传步骤仍一次性批量构建全部文件预览',
+)
+assert.match(
+  previewBuildSource,
+  /for \(const sourceFileId of sourceFileIds\)[\s\S]*?onProgress\(\{ source_file_id: sourceFileId, status: 'processing', progress: 0 \}\)/,
+  '逐文件构建 composable 没有串行处理并先上报 processing 进度',
+)
+assert.match(
+  previewBuildSource,
+  /buildDataProcessPreview\(taskId,\s*\{[\s\S]*?replace_existing:\s*true[\s\S]*?source_file_ids:\s*\[sourceFileId\]/,
+  '逐文件构建 composable 没有按单个源文件 ID 替换预览',
+)
+assert.match(
+  previewBuildSource,
+  /status: 'success', progress: 100, preview_count: previewCount/,
+  '单文件构建成功后没有上报 100% 和预览数量',
+)
+assert.match(
+  previewBuildSource,
+  /catch \(error\)[\s\S]*?status: 'failed'[\s\S]*?progress: 0[\s\S]*?error:/,
+  '单文件构建失败后没有保留可重试错误',
+)
 assert.match(nextFromUploadSource, /getDataProcessPreview\(/, '上传步骤没有读取后端预览结果')
-assert.match(nextFromUploadSource, /goToStep\('preview'\)/, '上传步骤完成后没有进入数据预览')
+assert.match(
+  nextFromUploadSource,
+  /const allFilesSucceeded = \(\) => uploadedFiles\.value\.every\(\(file\) => \([\s\S]*?file\.previewStatus === 'success'[\s\S]*?file\.previewConfigSignature === configSignature/,
+  '上传步骤缺少全部文件 success 且配置签名一致的完成判定',
+)
+const previewSuccessGateIndex = nextFromUploadSource.indexOf('if (failedCount || !allFilesSucceeded())')
+const previewStepIndex = nextFromUploadSource.lastIndexOf("goToStep('preview')")
+assert.ok(previewSuccessGateIndex >= 0, '上传步骤没有在任一文件未成功时停留当前步骤')
+assert.match(
+  nextFromUploadSource,
+  /if \(signature === previewSignature\.value && previewItems\.value\.length && allFilesSucceeded\(\)\) \{\s*goToStep\('preview'\)/,
+  '缓存预览快速路径没有要求全部文件 success',
+)
+assert.equal(
+  (nextFromUploadSource.match(/goToStep\('preview'\)/g) || []).length,
+  2,
+  '上传步骤只能通过全成功缓存路径或本轮全成功路径进入预览',
+)
+assert.ok(
+  previewStepIndex > previewSuccessGateIndex,
+  '上传步骤必须在全部文件 success 后才能进入数据预览',
+)
 assert.match(viewSource, /function goToStep\(stepId: StepId\)[\s\S]*?WIZARD_STEPS\.findIndex/, '向导跳转没有使用稳定步骤标识')
 assert.match(viewSource, /currentStepId\.value === 'preview'[\s\S]*?goToStep\('generate'\)/, '数据预览步骤没有进入开始生成')
 assert.match(viewSource, /generation\.status === 'success'[\s\S]*?goToStep\('results'\)/, '生成成功后没有进入结果编辑与保存')
-assert.match(draftSource, /currentStepId:\s*bindings\.currentStepId\.value/, '草稿没有保存稳定步骤标识')
-const draftSnapshotSource = draftSource.slice(
-  draftSource.indexOf('function draftSnapshot()'),
-  draftSource.indexOf('function writeDraft'),
-)
-for (const forbiddenField of ['uploadedFiles', 'previewItems', 'results', 'password', 'token']) {
-  assert.ok(!draftSnapshotSource.includes(forbiddenField), `安全草稿不应持久化：${forbiddenField}`)
-}
-assert.match(draftSource, /writeDraft\(false\)/, '恢复旧草稿后没有立即覆盖潜在敏感数据')
 assert.match(viewSource, /<ResultEditorStep\s+[\s\S]*?v-else-if="currentStepId === 'results'"/, '结果编辑器必须只在结果步骤渲染')
 assert.match(viewSource, /watch\(processType,[\s\S]*?resetSourceDataForProcessTypeChange\(\)/, '切换处理类型后没有失效旧源数据')
 assert.match(viewSource, /function resetSourceDataForProcessTypeChange\(\)[\s\S]*?uploadedFiles\.value = \[\][\s\S]*?selectedPreviewFileId\.value = null/, '旧源数据失效没有同步清理文件与预览选择')

 assert.match(taskSetupSource, /v-if="processType === 'structured'"/, '结构化配置必须仅在结构化数据类型下显示')
-for (const option of [
-  '清理无效数据',
-  '识别表格结构',
-  '重复数据去重',
-  '数据格式标准化',
-  '异常数据过滤',
-  '敏感信息脱敏',
-]) {
-  assert.ok(structuredOptionsSource.includes(option), `结构化预处理缺少选项：${option}`)
+const expectedStructuredOptions = [
+  ['clean_invalid', '清理无效数据', '清理全空列，并剔除关键字段残缺的数据行'],
+  ['detect_structure', '识别表格结构', '识别多级表头与合并单元格，并将嵌套字段展平'],
+  ['deduplicate', '重复数据去重', '基于整行精确匹配和关键字段组合删除重复记录'],
+  ['normalize_format', '数据格式标准化', '按所选规则统一编码、空白、字段名及 JSON 序列化格式'],
+  ['filter_anomaly', '异常数据过滤', '使用 IQR 识别数值离群值，并过滤乱码等异常记录'],
+  ['desensitize', '敏感信息脱敏', '识别并脱敏姓名、手机号、邮箱和身份证号'],
+]
+for (const [value, label, description] of expectedStructuredOptions) {
+  assert.ok(structuredOptionsSource.includes(`value: '${value}'`), `结构化预处理缺少值：${value}`)
+  assert.ok(structuredOptionsSource.includes(`label: '${label}'`), `结构化预处理缺少标签：${label}`)
+  assert.ok(structuredOptionsSource.includes(`description: '${description}'`), `结构化预处理语义不准确：${value}`)
 }
+const structuredOptionValues = [...structuredOptionsSource.matchAll(/\{ value: '([^']+)', label:/g)]
+  .map((match) => match[1])
+assert.deepEqual(structuredOptionValues, expectedStructuredOptions.map(([value]) => value), '结构化预处理值集合不准确')
+assert.equal(new Set(structuredOptionValues).size, structuredOptionValues.length, '结构化预处理 value 必须唯一')
+assert.match(structuredOptionsSource, /Array\.from\(new Set\(value\.filter\(/, '结构化预处理选中值没有去重')
 assert.ok(structuredOptionsSource.includes('生成选项'), '结构化配置缺少生成选项分类')
 for (const splitName of ['训练集', '验证集', '测试集']) {
  assert.ok(datasetSplitEditorSource.includes(splitName), `生成选项缺少数据集划分：${splitName}`)
@@ -260,8 +343,6 @@ for (const splitField of ['train', 'validation', 'test']) {
 assert.match(structuredOptionsSource, /<el-input-number[\s\S]*options\.qaPairsPerRow[\s\S]*:min="1"[\s\S]*:max="5"/, '每行生成数量必须限制在 1 到 5')
 assert.match(viewSource, /const structuredOptions = ref<StructuredProcessOptions>/, '父页面缺少结构化配置状态')
 assert.match(stateSource, /datasetSplit:\s*\{ train: 80, validation: 10, test: 10 \}/, '数据集划分默认值必须为 80/10/10')
-assert.match(draftSource, /structuredOptions:\s*\{[\s\S]*\.\.\.bindings\.structuredOptions\.value/, '结构化配置没有写入草稿')
-assert.match(draftSource, /bindings\.structuredOptions\.value = \{[\s\S]*\.\.\.snapshot\.structuredOptions/, '结构化配置没有从草稿恢复')
 assert.match(viewSource, /v-model:structured-options="structuredOptions"/, '父页面没有双向绑定结构化配置')
 assert.match(generationSource, /generateDataProcess\(taskId\)/, '开始生成没有调用真实 API')
 assert.match(generationSource, /getDataProcessProgress\(taskId\)/, '生成状态没有通过真实 API 轮询')
@@ -284,13 +365,39 @@ for (const apiName of [
 ]) {
  assert.match(apiSource, new RegExp(`export (?:const|async function|function) ${apiName}\\b`), `API 模块缺少 ${apiName}`)
 }
-assert.match(viewSource, /uploadDataProcessSourceFiles\(taskId\.value,\s*\[raw\]\)/, '文件上传没有调用真实 API')
+assert.match(sourceUploadWorkerSource, /uploadDataProcessSourceFiles\(currentTaskId,\s*\[job\.file\]/, '文件上传没有逐文件调用真实 API')
+assert.match(sourceUploadWorkerSource, /STRUCTURED_FILE_EXTENSIONS = new Set\(\['json', 'jsonl', 'ndjson', 'csv', 'tsv', 'xlsx'\]\)/, '结构化文件扩展名白名单不完整')
+for (const extension of ['txt', 'md', 'markdown', 'pdf', 'docx', 'pptx', 'json', 'jsonl', 'ndjson']) {
+  assert.ok(sourceUploadWorkerSource.includes(`'${extension}'`), `非结构化文件扩展名白名单缺少 ${extension}`)
+}
+assert.match(sourceUploadWorkerSource, /LEGACY_OFFICE_EXTENSIONS = new Set\(\['doc', 'xls', 'ppt'\]\)/, '缺少旧版 Office 格式识别')
+assert.ok(sourceUploadWorkerSource.includes('请分别转换为 DOCX、XLSX、PPTX 后上传'), '旧版 Office 文件缺少转换提示')
+assert.match(sourceUploadWorkerSource, /if \(!BINARY_FILE_EXTENSIONS\.has\(job\.extension\)\) \{[\s\S]*?TextDecoder/, '文本格式没有执行 UTF-8 客户端校验')
+assert.match(sourceUploadWorkerSource, /if \(BINARY_FILE_EXTENSIONS\.has\(job\.extension\)\) \{[\s\S]*?getDataProcessSourceContent\(currentTaskId, source\.id,[\s\S]*?start_line:\s*1,[\s\S]*?line_count:\s*10_000/, '二进制文档上传后没有读取后端解析文本')
 assert.match(apiSource, /formData\.append\('files', file\)/, '上传 API 没有使用 files 多文件表单字段')
+assert.match(apiSource, /onUploadProgress:[\s\S]*?event\.loaded \/ event\.total[\s\S]*?Math\.min\(99,/, '上传 API 没有接入真实字节进度或响应前未限制在 99%')
+assert.match(apiSource, /source-files`[\s\S]*?timeout: 5 \* 60 \* 1000/, '源文件上传缺少 5 分钟超时')
 assert.match(apiSource, /\/preview\/build/, 'API 模块缺少后端预览构建路径')
+assert.match(apiSource, /\/preview\/build`[\s\S]*?\{ timeout: 5 \* 60 \* 1000 \}/, '单文件切分请求缺少 5 分钟超时')
 assert.match(apiSource, /\/progress`/, 'API 模块缺少生成进度路径')
 assert.match(apiSource, /\/results`/, 'API 模块缺少结果分页路径')
 assert.match(apiSource, /\/publish`/, 'API 模块缺少数据集发布路径')
 assert.match(contractTypesSource, /source_file_ids\?: Array<string \| number>/, '预览构建契约缺少源文件 ID 列表')
+assert.match(
+  contractTypesSource,
+  /export type DataProcessPreviewFileStatus = 'waiting' \| 'processing' \| 'success' \| 'failed'/,
+  '文件预览状态契约不完整',
+)
+for (const field of ['rawFile', 'status', 'uploadProgress', 'uploadError', 'previewStatus', 'previewProgress', 'previewError', 'previewConfigSignature']) {
+  assert.ok(typesSource.includes(field), `上传文件缺少逐文件预览字段：${field}`)
+}
+assert.match(typesSource, /status: 'queued' \| 'uploading' \| 'ready' \| 'failed'/, '上传文件状态机不完整')
+assert.match(viewSource, /uploadedFiles\.value\.push\([\s\S]*?status: 'queued'[\s\S]*?enqueueSourceUpload/, '文件选择后没有先进入列表再加入上传队列')
+assert.match(sourceUploadWorkerSource, /while \(queue\.length\) \{[\s\S]*?await uploadOne\(job\)/, '多文件上传没有由单一队列逐个等待')
+assert.doesNotMatch(sourceUploadWorkerSource, /Promise\.(?:all|allSettled)/, '上传队列不得并发消费文件')
+assert.match(sourceUploadWorkerSource, /pending\.status = 'ready'[\s\S]*?pending\.uploadProgress = 100/, '服务端响应成功后没有将文件置为上传完成')
+assert.match(viewSource, /failedUploads[\s\S]*?hasUnfinishedUploads[\s\S]*?buildPreviewsByFile/, '上传失败或未完成时没有阻断切分')
+assert.doesNotMatch(viewSource, /file\.sourceFileId \|\| file\.uid/, '切分或删除仍可能把本地临时 UID 当成后端文件 ID')
 assert.match(contractTypesSource, /expected_updated_at\?: string/, '编辑契约缺少乐观并发版本字段')

 for (const field of [
@@ -302,7 +409,7 @@ for (const field of [
  'minOutputLength',
 ]) {
  assert.ok(typesSource.includes(field), `生成控制配置缺少字段：${field}`)
-  assert.ok(implementationSource.includes(field), `父页面默认值或草稿状态缺少字段：${field}`)
+  assert.ok(implementationSource.includes(field), `父页面默认值缺少字段：${field}`)
 }
 assert.match(structuredOptionsSource, /GenerationOptionsPanel/, '结构化生成选项没有复用统一的质量筛选组件')
 assert.match(unstructuredOptionsSource, /GenerationOptionsPanel/, '非结构化生成选项没有复用统一的质量筛选组件')
@@ -336,10 +443,23 @@ assert.match(generationControlSource, /v-if="options\.filterShortContent"/, '最
 assert.match(generationControlSource, /:min="1"[\s\S]*:max="1000"/, '最少字数缺少 1 到 1000 的边界限制')
 assert.match(taskSetupSource, /qualityValidationMessage/, '质量规则缺少继续前校验')
 assert.match(viewSource, /useModelsStore/, '创建页没有加载模型列表')
-assert.match(viewSource, /model\.type === 'LLM'/, '数据生成模型列表没有排除非大模型')
+const generationModelsStart = viewSource.indexOf('const generationModels')
+const generationModelsEnd = viewSource.indexOf('const taskSetupRef', generationModelsStart)
+const generationModelsSource = viewSource.slice(generationModelsStart, generationModelsEnd)
+assert.match(generationModelsSource, /computed\(\(\) => modelList\.value\)/, '数据生成模型候选没有直接使用模型管理完整列表')
+assert.doesNotMatch(generationModelsSource, /\.filter\(|model\.type|model\.model_source|model\.status|model\.purpose/, '数据生成模型候选仍按类型、来源、状态或用途静默过滤')
+assert.match(viewSource, /modelsStore\.load\(true\)/, '进入创建向导时没有强制刷新模型管理列表')
 assert.match(viewSource, /<ModelSelectionStep[\s\S]*?:models="generationModels"/, '创建页没有向独立大模型选择步骤传递模型列表')
+assert.match(generationControlSource, /v-for="model in models \|\| \[\]"/, '模型下拉没有遍历完整候选列表')
+assert.match(generationControlSource, /:value="model\.id"/, '模型下拉没有使用模型管理 ID 作为选中值')
+assert.match(generationControlSource, /modelMeta\(model\)/, '模型下拉缺少来源和类型说明')

 assert.match(typesSource, /export interface UnstructuredProcessOptions/, '缺少非结构化处理选项类型')
+assert.match(
+  typesSource,
+  /export type ChunkMethod = 'structure' \| 'fixed' \| 'custom'/,
+  '非结构化切分方式类型必须只保留 structure、fixed 和 custom',
+)
 for (const field of [
  'preprocessOptions',
  'chunkMethod',
@@ -363,14 +483,54 @@ for (const removedField of ['contextScope', 'generationTypes', 'skipUnanswerable
 assert.match(taskSetupSource, /v-if="processType === 'unstructured'"/, '非结构化配置必须仅在非结构化数据类型下显示')
 assert.ok(unstructuredOptionsSource.includes('智能预处理'), '简化后缺少智能预处理总开关')
 assert.ok(unstructuredOptionsSource.includes('敏感信息脱敏'), '简化后缺少脱敏开关')
+const expectedSmartPreprocessOptions = [
+  'clean_invalid_content',
+  'detect_document_structure',
+  'merge_short_content',
+  'filter_low_quality',
+  'deduplicate_content',
+  'preserve_context',
+]
+const smartOptionsStart = unstructuredOptionsSource.indexOf('const SMART_PREPROCESS_OPTIONS')
+const smartOptionsEnd = unstructuredOptionsSource.indexOf('const CHUNK_METHODS', smartOptionsStart)
+const smartOptionsSource = unstructuredOptionsSource.slice(smartOptionsStart, smartOptionsEnd)
+const smartOptionValues = [...smartOptionsSource.matchAll(/^\s*'([^']+)',?$/gm)].map((match) => match[1])
+assert.deepEqual(smartOptionValues, expectedSmartPreprocessOptions, '智能预处理内部值与后端语义不一致')
+assert.equal(new Set(smartOptionValues).size, smartOptionValues.length, '非结构化智能预处理 value 必须唯一')
+for (const descriptionPart of [
+  '清理无效内容',
+  '感知文档结构',
+  '合并短块',
+  '预过滤低质量内容',
+  '近重复去重',
+  '重叠保护上下文',
+]) {
+  assert.ok(unstructuredOptionsSource.includes(descriptionPart), `智能预处理说明缺少语义：${descriptionPart}`)
+}
+assert.ok(unstructuredOptionsSource.includes('姓名、手机号、邮箱和身份证号'), '非结构化脱敏说明缺少完整字段范围')
+assert.match(unstructuredOptionsSource, /Array\.from\(new Set\(props\.options\.preprocessOptions\.filter\(/, '非结构化预处理选中值没有去重')
 assert.match(unstructuredOptionsSource, /const smartPreprocessEnabled = computed/, '智能预处理没有映射到内部处理项')
 assert.match(unstructuredOptionsSource, /function updateSmartPreprocess/, '智能预处理开关缺少更新逻辑')
 assert.match(unstructuredOptionsSource, /function updateDesensitize/, '脱敏开关缺少更新逻辑')

 assert.ok(unstructuredOptionsSource.includes('切分选项'), '非结构化配置缺少切分选项分类')
-for (const method of ['自动语义切分', '按标题和段落', '按固定长度', '自定义分隔符']) {
-  assert.ok(unstructuredOptionsSource.includes(method), `切分方式缺少选项：${method}`)
+const expectedChunkMethods = [
+  ['structure', '文档结构'],
+  ['fixed', '固定 Token'],
+  ['custom', '自定义分隔符'],
+]
+const chunkMethodsStart = unstructuredOptionsSource.indexOf('const CHUNK_METHODS')
+const chunkMethodsEnd = unstructuredOptionsSource.indexOf('const UNSTRUCTURED_NUMBER_LIMITS', chunkMethodsStart)
+const chunkMethodsSource = unstructuredOptionsSource.slice(chunkMethodsStart, chunkMethodsEnd)
+const chunkMethodValues = [...chunkMethodsSource.matchAll(/value: '([^']+)'/g)].map((match) => match[1])
+assert.deepEqual(chunkMethodValues, expectedChunkMethods.map(([value]) => value), '切分方式值集合不准确')
+for (const [value, label] of expectedChunkMethods) {
+  assert.ok(chunkMethodsSource.includes(`value: '${value}', label: '${label}'`), `切分方式缺少选项：${label}`)
 }
+for (const removedMethod of ['semantic', 'heading']) {
+  assert.ok(!chunkMethodsSource.includes(`value: '${removedMethod}'`), `切分方式仍保留已移除值：${removedMethod}`)
+}
+assert.ok(unstructuredOptionsSource.includes('推荐使用文档结构'), '切分方式缺少文档结构默认推荐说明')
 for (const label of ['切片长度', '重叠长度', '最小切片长度', '保护表格、代码和列表']) {
  assert.ok(unstructuredOptionsSource.includes(label), `切分选项缺少配置：${label}`)
 }
@@ -395,19 +555,87 @@ assert.ok(unstructuredOptionsSource.includes('Token 数为轻量估算值'), '
 assert.match(unstructuredOptionsSource, /\.chunk-settings-grid\s*\{[\s\S]*?grid-template-columns:\s*repeat\(3,\s*minmax\(0,\s*1fr\)\)/, '核心切分参数没有收紧为三列布局')

 assert.match(viewSource, /const unstructuredOptions = ref<UnstructuredProcessOptions>/, '父页面缺少非结构化配置状态')
-assert.match(stateSource, /chunkMethod:\s*'semantic'/, '非结构化默认切分方式必须为自动语义切分')
+assert.match(stateSource, /chunkMethod:\s*'structure'/, '非结构化默认切分方式必须为文档结构')
 assert.match(stateSource, /chunkSize:\s*800/, '默认切片长度必须为 800 Token')
 assert.match(stateSource, /chunkOverlap:\s*100/, '默认重叠长度必须为 100 Token')
 assert.match(stateSource, /minChunkSize:\s*100/, '默认最小切片长度必须为 100 Token')
 assert.match(stateSource, /qaPairsPerChunk:\s*1/, '默认每个切片必须生成 1 个问答对')
-assert.match(draftSource, /unstructuredOptions:\s*\{[\s\S]*\.\.\.bindings\.unstructuredOptions\.value/, '非结构化配置没有写入草稿')
-assert.match(draftSource, /bindings\.unstructuredOptions\.value = \{[\s\S]*\.\.\.snapshot\.unstructuredOptions/, '非结构化配置没有从草稿恢复')
 assert.match(viewSource, /v-model:unstructured-options="unstructuredOptions"/, '父页面没有双向绑定非结构化配置')
-assert.match(draftSource, /DATA_PROCESS_DRAFT_SCHEMA_VERSION = \d+/, '草稿缺少版本标识')
-assert.match(draftSource, /schemaVersion:\s*DATA_PROCESS_DRAFT_SCHEMA_VERSION/, '草稿快照没有写入版本标识')
-assert.match(draftSource, /bindings\.goToStep\('create'\)/, '恢复配置后应回到安全的创建步骤')
 assert.match(viewSource, /JSON\.stringify\(previewAffectingOptions\(\)\)/, '影响切分的非结构化配置没有纳入预览失效判断')

+function defaultPreprocessValues(functionName, nextFunctionName) {
+  const start = stateSource.indexOf(`export function ${functionName}`)
+  const end = nextFunctionName ? stateSource.indexOf(`export function ${nextFunctionName}`, start) : stateSource.length
+  const functionSource = stateSource.slice(start, end)
+  const match = functionSource.match(/preprocessOptions:\s*\[([\s\S]*?)\]/)
+  assert.ok(match, `${functionName} 缺少 preprocessOptions 默认值`)
+  return [...match[1].matchAll(/'([^']+)'/g)].map((item) => item[1])
+}
+
+const defaultStructuredPreprocess = defaultPreprocessValues(
+  'createDefaultStructuredOptions',
+  'createDefaultUnstructuredOptions',
+)
+assert.deepEqual(
+  defaultStructuredPreprocess,
+  ['clean_invalid', 'detect_structure', 'deduplicate', 'normalize_format'],
+  '结构化默认预处理配置不准确',
+)
+assert.equal(new Set(defaultStructuredPreprocess).size, defaultStructuredPreprocess.length, '结构化默认预处理值重复')
+const defaultUnstructuredPreprocess = defaultPreprocessValues('createDefaultUnstructuredOptions')
+assert.deepEqual(defaultUnstructuredPreprocess, expectedSmartPreprocessOptions, '智能预处理默认值不完整')
+assert.equal(new Set(defaultUnstructuredPreprocess).size, defaultUnstructuredPreprocess.length, '非结构化默认预处理值重复')
+
+const backendConfigStart = viewSource.indexOf('function toBackendConfig()')
+const backendConfigEnd = viewSource.indexOf('function taskPayload()', backendConfigStart)
+assert.ok(backendConfigStart >= 0 && backendConfigEnd > backendConfigStart, '缺少任务后端配置映射')
+const backendConfigSource = viewSource.slice(backendConfigStart, backendConfigEnd)
+assert.ok(backendConfigSource.includes('preprocess_options: [...options.preprocessOptions]'), '预处理选项没有完整传入任务配置')
+assert.ok(backendConfigSource.includes('dataset_split: { ...options.datasetSplit }'), '数据集划分没有完整传入任务配置')
+for (const [backendField, frontendField] of [
+  ['semantic_enrichment', 'semanticEnrichment'],
+  ['generation_model_id', 'generationModelId'],
+  ['generation_prompt', 'generationPrompt'],
+  ['temperature', 'temperature'],
+  ['max_tokens', 'maxTokens'],
+  ['json_mode', 'jsonMode'],
+  ['quality_filter_enabled', 'qualityFilterEnabled'],
+  ['filter_low_quality', 'filterLowQuality'],
+  ['filter_short_content', 'filterShortContent'],
+  ['min_output_length', 'minOutputLength'],
+]) {
+  assert.ok(
+    backendConfigSource.includes(`${backendField}: options.${frontendField}`),
+    `公共配置 ${frontendField} 没有传入 task payload`,
+  )
+}
+for (const [backendField, frontendField] of [
+  ['chunk_method', 'chunkMethod'],
+  ['chunk_size', 'chunkSize'],
+  ['chunk_overlap', 'chunkOverlap'],
+  ['min_chunk_size', 'minChunkSize'],
+  ['custom_delimiter', 'customDelimiter'],
+  ['preserve_tables', 'preserveTables'],
+  ['preserve_code_blocks', 'preserveCodeBlocks'],
+  ['preserve_lists', 'preserveLists'],
+  ['qa_pairs_per_chunk', 'qaPairsPerChunk'],
+]) {
+  assert.ok(
+    backendConfigSource.includes(`${backendField}: unstructuredOptions.value.${frontendField}`),
+    `非结构化配置 ${frontendField} 没有传入 task payload`,
+  )
+}
+assert.ok(
+  backendConfigSource.includes('qa_pairs_per_row: structuredOptions.value.qaPairsPerRow'),
+  '结构化每行生成数量没有传入 task payload',
+)
+const taskPayloadStart = viewSource.indexOf('function taskPayload()')
+const taskPayloadEnd = viewSource.indexOf('function externalPayload()', taskPayloadStart)
+const taskPayloadSource = viewSource.slice(taskPayloadStart, taskPayloadEnd)
+for (const marker of ['name: task.name.trim()', 'description: task.description.trim()', 'process_type: processType.value', 'config: toBackendConfig()']) {
+  assert.ok(taskPayloadSource.includes(marker), `任务创建 payload 缺少：${marker}`)
+}
+
 const previewOptionsStart = viewSource.indexOf('function previewAffectingOptions()')
 const previewOptionsEnd = viewSource.indexOf('function generationAffectingOptions()', previewOptionsStart)
 assert.ok(previewOptionsStart >= 0 && previewOptionsEnd > previewOptionsStart, '缺少预览影响配置签名函数')
@@ -667,7 +895,7 @@ for (const property of ['max-height', 'overflow', 'overflow-x', 'overflow-y']) {
 for (const marker of [
  'uploaded-file-list-header',
  'uploaded-file-items',
-  '已添加 {{ uploadedFiles.length }} 个文件',
+  '已选择 {{ uploadedFiles.length }} 个文件',
  ':title="file.name"',
 ]) {
  assert.ok(sourceUploadSource.includes(marker), `源数据文件列表缺少：${marker}`)
@@ -720,7 +948,7 @@ assert.match(
 )
 assert.match(
  sourceUploadSource,
-  /<div\s+class="uploaded-file-list-header">\s*<span>已添加 \{\{ uploadedFiles\.length \}\} 个文件<\/span>/,
+  /<div\s+class="uploaded-file-list-header">[\s\S]*?已选择 \{\{ uploadedFiles\.length \}\} 个文件[\s\S]*?正在逐个上传/,
  '文件列表标题结构或文件数量文案缺失',
 )
 assert.match(
@@ -730,17 +958,27 @@ assert.match(
 )
 assert.match(
  sourceUploadSource,
-  /<el-upload\s+v-if="uploadedFiles\.length === 0"\s+drag\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>/,
+  /<el-upload\s+v-if="uploadedFiles\.length === 0"\s+drag\s+multiple\s+:accept="uploadAccept"\s+:disabled="previewBuilding"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>/,
  '无文件时未保留原有大拖拽上传区或上传配置',
 )
 assert.match(
  sourceUploadSource,
-  /<el-upload\s+v-if="uploadedFiles\.length === 0"\s+drag\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>[\s\S]*?<template\s+#tip>\s*<div\s+class="el-upload__tip">\s*\{\{\s*processType === 'unstructured'\s*\? '支持 TXT、Markdown、PDF、Word、JSON、JSONL，单文件不超过 200MB'\s*:\s*'支持 JSON、JSONL、CSV、Excel，单文件不超过 200MB'\s*\}\}/,
-  '无文件时大拖拽上传区缺少格式提示槽、处理类型分支或完整格式提示',
+  /<el-upload\s+v-if="uploadedFiles\.length === 0"[\s\S]*?<template\s+#tip>[\s\S]*?processType === 'unstructured'/,
+  '无文件时大拖拽上传区缺少按处理类型展示的格式提示',
 )
+assert.ok(
+  sourceUploadSource.includes("? '.txt,.md,.markdown,.pdf,.docx,.pptx,.json,.jsonl,.ndjson'"),
+  '非结构化上传 accept 不完整',
+)
+assert.ok(
+  sourceUploadSource.includes(": '.json,.jsonl,.ndjson,.csv,.tsv,.xlsx'"),
+  '结构化上传 accept 不完整',
+)
+assert.ok(sourceUploadSource.includes('旧版 DOC/PPT 请先转换'), '非结构化格式提示没有说明旧版 DOC/PPT 需转换')
+assert.ok(sourceUploadSource.includes('旧版 XLS 请先转换'), '结构化格式提示没有说明旧版 XLS 需转换')
 assert.match(
  sourceUploadSource,
-  /<div\s+class="uploaded-file-list-header">\s*<span>已添加 \{\{ uploadedFiles\.length \}\} 个文件<\/span>\s*<div\s+class="continue-upload">\s*<el-upload\s+multiple\s+:accept="uploadAccept"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>\s*<el-button\s+size="small"\s+type="primary">继续上传<\/el-button>\s*<\/el-upload>\s*<\/div>\s*<\/div>/,
+  /<div\s+class="uploaded-file-list-header">[\s\S]*?已选择 \{\{ uploadedFiles\.length \}\} 个文件[\s\S]*?<div\s+class="continue-upload">\s*<el-upload\s+multiple\s+:accept="uploadAccept"\s+:disabled="previewBuilding"\s+:auto-upload="false"\s+:show-file-list="false"\s+:on-change="\(file: UploadFile\) => emit\('file-change', file\)"[^>]*>\s*<el-button\s+size="small"\s+type="primary"\s+:disabled="previewBuilding">继续上传<\/el-button>\s*<\/el-upload>\s*<\/div>\s*<\/div>/,
  '有文件时缺少标题右侧的继续上传触发器或上传配置',
 )
 assert.match(
@@ -754,7 +992,20 @@ assert.match(
  '继续上传未覆盖内层上传节点的宽度和顶部间距',
 )
 assert.match(sourceUploadSource, /\.uploaded-file\s*\{[^}]*min-height:\s*48px/, '文件行没有保持 48px 最小高度')
-assert.match(sourceUploadSource, /<span class="file-status"><i class="fa fa-check-circle"[^>]*\/> 校验通过<\/span>/, '文件行缺少校验成功状态')
+assert.match(sourceUploadSource, /success:\s*\{ label: '切分完成', icon: 'fa-check-circle' \}/, '文件行缺少切分成功状态')
+assert.match(sourceUploadSource, /previewBuilding/, '上传组件没有接收逐文件预览构建状态')
+assert.match(sourceUploadSource, /sourceUploading/, '上传组件没有接收串行上传状态')
+assert.match(
+  sourceUploadSource,
+  /<el-progress[\s\S]*?:percentage="getFileBarPercentage\(file\)"[\s\S]*?:indeterminate="isFileProcessing\(file\)"/,
+  '文件行缺少上传与切分阶段的独立进度',
+)
+for (const label of ['等待上传', '正在上传', '上传失败', '等待切分', '正在切分', '切分完成', '切分失败']) {
+  assert.ok(sourceUploadSource.includes(`label: '${label}'`), `文件行缺少状态：${label}`)
+}
+assert.match(sourceUploadSource, /file\.uploadProgress/, '文件行没有使用真实上传进度')
+assert.match(sourceUploadSource, /file\.uploadError \|\| file\.previewError/, '文件行没有按阶段展示上传或切分失败原因')
+assert.match(sourceUploadSource, /file\.previewError/, '文件行没有展示逐文件失败原因以支持重试')
 assert.match(sourceUploadSource, /@click="emit\('remove-file', file\.uid\)"/, '文件行缺少 remove-file 删除动作')

 const { descriptor } = parseSfc(viewSource, { filename: viewPath })