fix: 完善数据预处理与 JSON 上传链路

frontend/src/views/data-process/create/OfficeSourceViewer.vue

@@ -46,6 +46,13 @@ const sourceUrl = computed(() => (
     ? getDataProcessSourceRawUrl(props.taskId, props.sourceFileId)
     : ''
 ))
+const selectedXlsxLocator = computed(() => {
+  const locator = props.selectedItem?.sourceLocator
+  if (!locator) return null
+  const hasSheet = locator.sheet_index != null || Boolean(locator.sheet_name)
+  const hasRow = locator.row_number != null || locator.sheet_record_index != null
+  return hasSheet && hasRow ? locator : null
+})
 const visibleRowRange = computed(() => {
   const sheet = xlsxPreview.value?.active_sheet
   if (!sheet || !sheet.rows.length) return '当前工作表没有可预览记录'
@@ -95,6 +102,16 @@ const selectedRecordKey = computed(() => {
 })
 
 function xlsxRowHighlighted(row: DataProcessXlsxPreviewRow) {
+  const locator = selectedXlsxLocator.value
+  const sheet = xlsxPreview.value?.active_sheet
+  if (locator && sheet) {
+    const sheetMatches = locator.sheet_index != null
+      ? sheet.index === locator.sheet_index
+      : sheet.name === locator.sheet_name
+    if (!sheetMatches) return false
+    if (locator.row_number != null) return row.row_number === locator.row_number
+    return row.record_index === locator.sheet_record_index
+  }
   return Boolean(selectedRecordKey.value && recordKey(row.record) === selectedRecordKey.value)
 }
 
@@ -119,8 +136,11 @@ async function locateSelectedItem() {
 async function loadPreview(options: { reset?: boolean } = {}) {
   const sequence = ++loadSequence
   if (options.reset) {
-    activeSheetIndex.value = 0
-    pageOffset.value = 0
+    const locator = selectedXlsxLocator.value
+    activeSheetIndex.value = locator?.sheet_index ?? 0
+    pageOffset.value = locator?.sheet_record_index == null
+      ? 0
+      : Math.floor(locator.sheet_record_index / XLSX_PAGE_SIZE) * XLSX_PAGE_SIZE
     preview.value = null
   }
   errorMessage.value = ''
@@ -178,8 +198,34 @@ watch(
 )
 
 watch(
-  () => props.selectedItem?.id,
-  () => void locateSelectedItem(),
+  () => [
+    props.selectedItem?.id,
+    props.selectedItem?.sourceLocator?.sheet_index,
+    props.selectedItem?.sourceLocator?.sheet_record_index,
+    props.selectedItem?.sourceLocator?.row_number,
+  ],
+  () => {
+    const locator = selectedXlsxLocator.value
+    if (!locator || isDocx.value) {
+      void locateSelectedItem()
+      return
+    }
+    const targetSheet = locator.sheet_index ?? activeSheetIndex.value
+    const targetOffset = locator.sheet_record_index == null
+      ? pageOffset.value
+      : Math.floor(locator.sheet_record_index / XLSX_PAGE_SIZE) * XLSX_PAGE_SIZE
+    const activeSheet = xlsxPreview.value?.active_sheet
+    if (
+      activeSheet?.index === targetSheet
+      && activeSheet.offset === targetOffset
+    ) {
+      void locateSelectedItem()
+      return
+    }
+    activeSheetIndex.value = targetSheet
+    pageOffset.value = targetOffset
+    void loadPreview()
+  },
 )
 </script>
 
@@ -301,6 +347,8 @@ watch(
               :key="row.row_number"
               class="xlsx-row"
               :class="{ 'is-highlighted': xlsxRowHighlighted(row) }"
+              :data-row-number="row.row_number"
+              :data-record-index="row.record_index"
             >
               <th class="row-number-cell">{{ row.row_number }}</th>
               <td

frontend/src/views/data-process/create/PreviewCompareStep.vue

@@ -2,7 +2,11 @@
 import { computed, nextTick, ref, watch } from 'vue'
 import OfficeSourceViewer from './OfficeSourceViewer.vue'
 import PdfSourceViewer from './PdfSourceViewer.vue'
-import { sourceLines } from './previewModel'
+import {
+  isManualPreviewItem,
+  sourceLineNumberAtOffset,
+  sourceLineWindow,
+} from './previewModel'
 import type { PreviewItem, ProcessType } from './types'
 
 const props = defineProps<{
@@ -31,9 +35,11 @@ const sourceViewerRef = ref<HTMLElement | null>(null)
 const search = ref('')
 const currentPage = ref(1)
 const PREVIEW_PAGE_SIZE = 10
+const SOURCE_LINE_RENDER_LIMIT = 240
+const SOURCE_LINE_CHARACTER_LIMIT = 4_000
+const sourceWindowStartLine = ref(1)
 const editingItemId = ref<string | null>(null)
 const editorDraft = ref('')
-const lines = computed(() => sourceLines(props.sourceText))
 const selectedItem = computed(() => props.items.find((item) => item.id === props.selectedId) ?? props.items[0])
 const editingItem = computed(() => props.items.find((item) => item.id === editingItemId.value))
 const normalizedFileFormat = computed(() => (
@@ -43,6 +49,27 @@ const normalizedFileFormat = computed(() => (
 ))
 const isPdfSource = computed(() => normalizedFileFormat.value === 'pdf')
 const isOfficeSource = computed(() => ['docx', 'xlsx'].includes(normalizedFileFormat.value))
+const selectedSourceOffset = computed(() => {
+  const item = selectedItem.value
+  return item ? sourceOffsetRange(item)?.start ?? null : null
+})
+const selectedSourceLine = computed(() => {
+  const item = selectedItem.value
+  if (!item) return null
+  return sourceLineRange(item)?.start
+    ?? (selectedSourceOffset.value == null
+      ? null
+      : sourceLineNumberAtOffset(props.sourceText, selectedSourceOffset.value))
+})
+const visibleSourceWindow = computed(() => sourceLineWindow(
+  props.sourceText,
+  sourceWindowStartLine.value,
+  SOURCE_LINE_RENDER_LIMIT,
+  SOURCE_LINE_CHARACTER_LIMIT,
+  selectedSourceLine.value,
+  selectedSourceOffset.value,
+))
+const lines = computed(() => visibleSourceWindow.value.lines)
 
 const filteredItems = computed(() => props.items.filter((item, index) => {
   const matchesSearch = !search.value.trim()
@@ -58,10 +85,27 @@ const pagedItems = computed(() => {
 
 const selectedIndex = computed(() => props.items.findIndex((item) => item.id === selectedItem.value?.id))
 
-function isLineHighlighted(lineStart: number, lineEnd: number) {
+function sourceLineRange(item: PreviewItem) {
+  const start = item.sourceLocator?.start_line ?? item.sourceStartLine
+  const end = item.sourceLocator?.end_line ?? item.sourceEndLine ?? start
+  return start == null ? null : { start, end: end ?? start }
+}
+
+function sourceOffsetRange(item: PreviewItem) {
+  const start = item.sourceLocator?.source_start ?? item.sourceStart
+  const end = item.sourceLocator?.source_end ?? item.sourceEnd ?? start
+  return start == null ? null : { start, end: Math.max(start, end ?? start) }
+}
+
+function isLineHighlighted(lineNumber: number, lineStart: number, lineEnd: number) {
   const item = selectedItem.value
-  if (!item || item.sourceStart == null || item.sourceEnd == null) return false
-  return lineEnd >= item.sourceStart && lineStart <= item.sourceEnd
+  if (!item) return false
+  const lineRange = sourceLineRange(item)
+  if (lineRange) return lineNumber >= lineRange.start && lineNumber <= lineRange.end
+  const offsetRange = sourceOffsetRange(item)
+  if (!offsetRange) return false
+  const effectiveEnd = Math.max(offsetRange.start + 1, offsetRange.end)
+  return lineEnd >= offsetRange.start && lineStart < effectiveEnd
 }
 
 function selectItem(id: string) {
@@ -107,34 +151,88 @@ watch(search, () => {
 
 watch(() => props.selectedFileId, closeEditor)
 
-watch(selectedItem, async (item) => {
+watch([selectedItem, () => props.sourceText], async ([item]) => {
   if (!item) return
   const visibleIndex = filteredItems.value.findIndex((entry) => entry.id === item.id)
   if (visibleIndex >= 0) {
     currentPage.value = Math.floor(visibleIndex / PREVIEW_PAGE_SIZE) + 1
   }
 
-  if (isPdfSource.value || isOfficeSource.value || item.sourceStart == null) return
+  if (isPdfSource.value || isOfficeSource.value) return
+  const itemLineRange = sourceLineRange(item)
+  const itemOffsetRange = sourceOffsetRange(item)
+  if (!itemLineRange && !itemOffsetRange) {
+    sourceWindowStartLine.value = 1
+    return
+  }
+  const targetLine = selectedSourceLine.value
+    ?? sourceLineNumberAtOffset(props.sourceText, itemOffsetRange?.start ?? 0)
+  sourceWindowStartLine.value = Math.max(1, targetLine - Math.floor(SOURCE_LINE_RENDER_LIMIT / 3))
   await nextTick()
-  const target = sourceViewerRef.value?.querySelector<HTMLElement>(`[data-source-start="${item.sourceStart}"]`)
+  const exactTarget = sourceViewerRef.value
+    ?.querySelector<HTMLElement>(`[data-line-number="${targetLine}"]`)
+  const target = exactTarget
     ?? sourceViewerRef.value?.querySelector<HTMLElement>('.source-line.is-highlighted')
   target?.scrollIntoView({ block: 'center', behavior: 'smooth' })
 }, { immediate: true })
 
+async function showPreviousSourceWindow() {
+  sourceWindowStartLine.value = Math.max(1, sourceWindowStartLine.value - SOURCE_LINE_RENDER_LIMIT)
+  await nextTick()
+  if (sourceViewerRef.value) sourceViewerRef.value.scrollTop = 0
+}
+
+async function showNextSourceWindow() {
+  if (!visibleSourceWindow.value.hasMore) return
+  sourceWindowStartLine.value = visibleSourceWindow.value.endLine + 1
+  await nextTick()
+  if (sourceViewerRef.value) sourceViewerRef.value.scrollTop = 0
+}
+
 function itemNumber(item: PreviewItem) {
   return props.items.findIndex((entry) => entry.id === item.id) + 1
 }
 
 function lineRange(item: PreviewItem) {
-  if (item.sourcePages?.length) {
-    const first = item.sourcePages[0]
-    const last = item.sourcePages[item.sourcePages.length - 1]
-    return first === last ? `来源：第 ${first} 页` : `来源：第 ${first}–${last} 页`
+  if (isManualPreviewItem(item)) return '手动新增，无源文件定位'
+
+  const locator = item.sourceLocator
+  const locatedLines = sourceLineRange(item)
+  if (props.processType === 'unstructured') {
+    const parts: string[] = []
+    if (item.sourcePages?.length) {
+      const first = item.sourcePages[0]
+      const last = item.sourcePages[item.sourcePages.length - 1]
+      parts.push(first === last ? `第 ${first} 页` : `第 ${first}–${last} 页`)
+    }
+    if (locatedLines) {
+      parts.push(
+        locatedLines.start === locatedLines.end
+          ? `第 ${locatedLines.start} 行`
+          : `第 ${locatedLines.start}–${locatedLines.end} 行`,
+      )
+    }
+    if (item.headingPath?.length) parts.push(`章节：${item.headingPath.join(' / ')}`)
+    return parts.length ? `来源：${parts.join(' · ')}` : '来源：源文件内容（无精确定位）'
   }
-  if (item.sourceStartLine == null || item.sourceEndLine == null) return '手动新增，无源文件定位'
-  return item.sourceStartLine === item.sourceEndLine
-    ? `来源：第 ${item.sourceStartLine} 行`
-    : `来源：第 ${item.sourceStartLine}–${item.sourceEndLine} 行`
+
+  if (locator?.kind === 'xlsx') {
+    const sheet = locator.sheet_name || `工作表 ${Number(locator.sheet_index ?? 0) + 1}`
+    return locator.row_number != null
+      ? `来源：${sheet} · 第 ${locator.row_number} 行`
+      : `来源：${sheet}`
+  }
+  if (locator?.kind === 'json') {
+    return locator.json_pointer
+      ? `来源：JSON 路径 ${locator.json_pointer}`
+      : '来源：JSON 根对象'
+  }
+  if (locatedLines) {
+    return locatedLines.start === locatedLines.end
+      ? `来源：第 ${locatedLines.start} 行`
+      : `来源：第 ${locatedLines.start}–${locatedLines.end} 行`
+  }
+  return '来源：源文件记录'
 }
 </script>
 
@@ -175,6 +273,31 @@ function lineRange(item: PreviewItem) {
           <div>
             <strong>源文件 · {{ fileName }}</strong>
           </div>
+          <div
+            v-if="!isPdfSource && !isOfficeSource && lines.length"
+            class="source-window-controls"
+            aria-label="源文件行窗口"
+          >
+            <span>第 {{ visibleSourceWindow.startLine }}–{{ visibleSourceWindow.endLine }} 行</span>
+            <el-button
+              link
+              size="small"
+              aria-label="查看上一段源文件"
+              :disabled="!visibleSourceWindow.hasPrevious"
+              @click="showPreviousSourceWindow"
+            >
+              上一段
+            </el-button>
+            <el-button
+              link
+              size="small"
+              aria-label="查看下一段源文件"
+              :disabled="!visibleSourceWindow.hasMore"
+              @click="showNextSourceWindow"
+            >
+              下一段
+            </el-button>
+          </div>
         </div>
 
         <PdfSourceViewer
@@ -197,8 +320,9 @@ function lineRange(item: PreviewItem) {
             v-for="line in lines"
             :key="line.number"
             class="source-line"
-            :class="{ 'is-highlighted': isLineHighlighted(line.start, line.end) }"
+            :class="{ 'is-highlighted': isLineHighlighted(line.number, line.start, line.end) }"
             :data-source-start="line.start"
+            :data-line-number="line.number"
           >
             <span class="line-number">{{ line.number }}</span>
             <span class="line-content">{{ line.content || ' ' }}</span>
@@ -282,7 +406,7 @@ function lineRange(item: PreviewItem) {
           />
           <div class="editor-actions">
             <el-button
-              v-if="editingItem.sourceStart != null"
+              v-if="!isManualPreviewItem(editingItem)"
               link
               @click="restoreItem"
             >
@@ -431,6 +555,22 @@ function lineRange(item: PreviewItem) {
   }
 }
 
+.source-window-controls {
+  flex: none;
+  gap: 2px !important;
+
+  > span {
+    margin-right: 4px;
+    color: #8a93a3;
+    font-size: 11px;
+    white-space: nowrap;
+  }
+
+  :deep(.el-button) {
+    margin-left: 0;
+  }
+}
+
 .source-viewer {
   flex: 1;
   height: 538px;

frontend/src/views/data-process/create/StructuredOptionsPanel.vue

@@ -1,4 +1,5 @@
 <script setup lang="ts">
+import { computed } from 'vue'
 import type {
   GenerationControlOptions,
   PreprocessOption,
@@ -21,27 +22,32 @@ const emit = defineEmits<{
   'update:options': [value: StructuredProcessOptions]
 }>()
 
-const PREPROCESS_OPTIONS: Array<{
-  value: PreprocessOption
+const PREPROCESS_GROUPS: Array<{
+  values: PreprocessOption[]
   label: string
   description: string
 }> = [
-  { value: 'clean_invalid', label: '清理无效数据', description: '清理全空列，并剔除关键字段残缺的数据行' },
   {
-    value: 'detect_structure',
-    label: '嵌套结构展平',
-    description: '展平嵌套对象和可解析的 JSON 字段；Excel 表头与合并单元格在上传时自动解析',
+    values: ['clean_invalid', 'deduplicate'],
+    label: '数据清洗',
+    description: '清理全空列和空记录，并删除内容完全相同的记录；不会猜测可空字段是否必填',
   },
   {
-    value: 'deduplicate',
-    label: '重复记录去重',
-    description: '按整行内容或 id、uuid、key、code、*_id 等身份字段去重，暂不支持自定义组合字段',
+    values: ['detect_structure', 'normalize_format'],
+    label: '结构标准化',
+    description: '展平嵌套对象和可解析的 JSON 字段，并统一编码、空白、字段名和 JSON 序列化格式',
+  },
+  {
+    values: ['desensitize'],
+    label: '敏感信息脱敏',
+    description: '识别并脱敏姓名、手机号、邮箱和身份证号',
   },
-  { value: 'normalize_format', label: '数据格式标准化', description: '按所选规则统一编码、空白、字段名及 JSON 序列化格式' },
-  { value: 'filter_anomaly', label: '异常数据过滤', description: '使用 IQR 识别数值离群值，并过滤乱码等异常记录' },
-  { value: 'desensitize', label: '敏感信息脱敏', description: '识别并脱敏姓名、手机号、邮箱和身份证号' },
 ]
 
+const legacyAnomalyFilterEnabled = computed(() => (
+  props.options.preprocessOptions.includes('filter_anomaly')
+))
+
 function updateField<K extends keyof StructuredProcessOptions>(
   field: K,
   value: StructuredProcessOptions[K],
@@ -57,14 +63,26 @@ function updateQaPairsPerRow(value: number | undefined) {
   updateField('qaPairsPerRow', normalizeQaPairsGenerationCount(value))
 }
 
-function updatePreprocessOptions(value: Array<string | number | boolean>) {
-  const allowedValues = new Set(PREPROCESS_OPTIONS.map((option) => option.value))
-  const preprocessOptions = Array.from(new Set(value.filter(
-    (option): option is PreprocessOption => (
-      typeof option === 'string' && allowedValues.has(option as PreprocessOption)
-    ),
-  )))
-  updateField('preprocessOptions', preprocessOptions)
+function selectedCount(values: PreprocessOption[]) {
+  return values.filter((value) => props.options.preprocessOptions.includes(value)).length
+}
+
+function groupSelected(values: PreprocessOption[]) {
+  return selectedCount(values) === values.length
+}
+
+function groupIndeterminate(values: PreprocessOption[]) {
+  const count = selectedCount(values)
+  return count > 0 && count < values.length
+}
+
+function updatePreprocessGroup(values: PreprocessOption[], checked: string | number | boolean) {
+  const next = new Set(props.options.preprocessOptions)
+  values.forEach((value) => {
+    if (Boolean(checked)) next.add(value)
+    else next.delete(value)
+  })
+  updateField('preprocessOptions', [...next])
 }
 </script>
 
@@ -73,26 +91,34 @@ function updatePreprocessOptions(value: Array<string | number | boolean>) {
     <div class="section-title-row">
       <div>
         <h3>预处理选项</h3>
-        <p>选择在生成问答对之前需要执行的数据处理方式</p>
+        <p>默认不执行预处理，请按数据情况自行选择</p>
       </div>
     </div>
-    <el-checkbox-group
-      :model-value="options.preprocessOptions"
-      class="preprocess-option-grid"
-      @update:model-value="updatePreprocessOptions"
-    >
-      <el-checkbox
-        v-for="option in PREPROCESS_OPTIONS"
-        :key="option.value"
-        :value="option.value"
+    <div class="preprocess-option-grid">
+      <label
+        v-for="group in PREPROCESS_GROUPS"
+        :key="group.label"
         class="preprocess-option"
+        :class="{ 'is-checked': groupSelected(group.values) }"
       >
+        <el-checkbox
+          :model-value="groupSelected(group.values)"
+          :indeterminate="groupIndeterminate(group.values)"
+          @update:model-value="updatePreprocessGroup(group.values, $event)"
+        />
         <span class="preprocess-option-copy">
-          <strong>{{ option.label }}</strong>
-          <small>{{ option.description }}</small>
+          <strong>{{ group.label }}</strong>
+          <small>{{ group.description }}</small>
         </span>
-      </el-checkbox>
-    </el-checkbox-group>
+      </label>
+    </div>
+    <el-alert
+      v-if="legacyAnomalyFilterEnabled"
+      class="legacy-preprocess-alert"
+      type="warning"
+      :closable="false"
+      title="该历史任务仍启用了已停用的“异常数据过滤”；为保证结果可复现，本次继续保留"
+    />
   </div>
 
   <div class="form-section generation-options-section">

frontend/src/views/data-process/create/UnstructuredOptionsPanel.vue

@@ -119,7 +119,7 @@ defineExpose({ revealValidation })
     <div class="section-title-row">
       <div>
         <h3>预处理选项</h3>
-        <p>默认启用结构感知的推荐策略，只需决定是否需要脱敏</p>
+        <p>默认不执行预处理，请按文档情况自行选择</p>
       </div>
     </div>
     <div class="preprocess-option-grid">

frontend/src/views/data-process/create/dataProcessCreateState.ts

@@ -97,7 +97,7 @@ export function isBuiltInGenerationPrompt(value: string) {
 
 export function createDefaultStructuredOptions(): StructuredProcessOptions {
   return {
-    preprocessOptions: ['clean_invalid', 'detect_structure', 'deduplicate', 'normalize_format'],
+    preprocessOptions: [],
     semanticEnrichment: false,
     qaPairsPerRow: 1,
     datasetSplit: { train: 80, validation: 10, test: 10 },
@@ -117,22 +117,15 @@ export function createDefaultStructuredOptions(): StructuredProcessOptions {
 
 export function createDefaultUnstructuredOptions(): UnstructuredProcessOptions {
   return {
-    preprocessOptions: [
-      'clean_invalid_content',
-      'detect_document_structure',
-      'merge_short_content',
-      'filter_low_quality',
-      'deduplicate_content',
-      'preserve_context',
-    ],
+    preprocessOptions: [],
     chunkMethod: 'layout_hybrid',
     chunkSize: 800,
     chunkOverlap: 100,
     minChunkSize: 100,
     semanticBreakpointPercentile: 95,
-    preserveTables: true,
-    preserveCodeBlocks: true,
-    preserveLists: true,
+    preserveTables: false,
+    preserveCodeBlocks: false,
+    preserveLists: false,
     semanticEnrichment: false,
     qaPairsPerChunk: 1,
     datasetSplit: { train: 80, validation: 10, test: 10 },
@@ -218,11 +211,21 @@ function generationOptionsFromConfig(
 export function createStructuredOptionsFromConfig(config: DataProcessConfig): StructuredProcessOptions {
   const defaults = createDefaultStructuredOptions()
   const preprocessOptions = configValue<unknown>(config, 'preprocess_options', [])
+  const supportedPreprocessOptions = new Set<PreprocessOption>([
+    'clean_invalid',
+    'deduplicate',
+    'detect_structure',
+    'normalize_format',
+    'desensitize',
+    'filter_anomaly',
+  ])
   return {
     ...defaults,
     ...generationOptionsFromConfig(config, defaults),
     preprocessOptions: Array.isArray(preprocessOptions)
-      ? preprocessOptions.map(String) as PreprocessOption[]
+      ? Array.from(new Set(preprocessOptions.map(String).filter(
+        (option): option is PreprocessOption => supportedPreprocessOptions.has(option as PreprocessOption),
+      )))
       : defaults.preprocessOptions,
     semanticEnrichment: Boolean(configValue(
       config,

frontend/src/views/data-process/create/previewModel.ts

@@ -1,4 +1,4 @@
-import type { SourceLine } from './types'
+import type { PreviewItem, SourceLine } from './types'
 
 /** 仅用于“使用示例”上传；正式预览和切片全部由后端生成。 */
 export const DEFAULT_SOURCE_TEXT = [
@@ -12,19 +12,141 @@ export const DEFAULT_SOURCE_TEXT = [
   '答：复利是将上一期利息加入本金，再计算下一期利息。',
 ].join('\n')
 
-/**
- * 把后端返回的字符偏移映射为源文件行，仅负责界面高亮，不参与切片。
- */
-export function sourceLines(sourceText: string): SourceLine[] {
-  const rawLines = sourceText.split('\n')
-  let cursor = 0
+export interface SourceLineWindow {
+  lines: SourceLine[]
+  startLine: number
+  endLine: number
+  hasPrevious: boolean
+  hasMore: boolean
+}
 
-  return rawLines.map((content, index) => {
-    const start = cursor
-    const end = start + content.length
-    cursor = end + (index < rawLines.length - 1 ? 1 : 0)
-    return { number: index + 1, content, start, end }
-  })
+function unicodeCodePointLength(value: string, start = 0, end = value.length) {
+  let length = 0
+  let index = start
+  while (index < end) {
+    const codePoint = value.codePointAt(index)
+    index += codePoint != null && codePoint > 0xffff ? 2 : 1
+    length += 1
+  }
+  return length
+}
+
+function advanceCodePoints(value: string, start: number, end: number, count: number) {
+  let index = start
+  let remaining = Math.max(0, count)
+  while (index < end && remaining > 0) {
+    const codePoint = value.codePointAt(index)
+    index += codePoint != null && codePoint > 0xffff ? 2 : 1
+    remaining -= 1
+  }
+  return index
+}
+
+/**
+ * 只扫描并返回当前可见行窗口，不对全文 split，避免大文件生成巨量字符串数组。
+ * 字符定位场景可开启 code point 偏移，以与后端 Python 的字符计数保持一致。
+ */
+export function sourceLineWindow(
+  sourceText: string,
+  requestedStartLine: number,
+  maxLines: number,
+  maxCharactersPerLine: number,
+  focusLine: number | null = null,
+  focusOffset: number | null = null,
+): SourceLineWindow {
+  const startLine = Math.max(1, Math.trunc(requestedStartLine) || 1)
+  const limit = Math.max(1, Math.trunc(maxLines) || 1)
+  const characterLimit = Math.max(1, Math.trunc(maxCharactersPerLine) || 1)
+  const trackUnicodeOffsets = focusOffset != null
+  const lines: SourceLine[] = []
+  let lineNumber = 1
+  let jsCursor = 0
+  let sourceCursor = 0
+
+  while (jsCursor <= sourceText.length && lineNumber < startLine) {
+    const newlineIndex = sourceText.indexOf('\n', jsCursor)
+    const jsEnd = newlineIndex >= 0 ? newlineIndex : sourceText.length
+    sourceCursor = trackUnicodeOffsets
+      ? sourceCursor + unicodeCodePointLength(sourceText, jsCursor, jsEnd) + (newlineIndex >= 0 ? 1 : 0)
+      : (newlineIndex >= 0 ? newlineIndex + 1 : sourceText.length + 1)
+    jsCursor = newlineIndex >= 0 ? newlineIndex + 1 : sourceText.length + 1
+    lineNumber += 1
+  }
+
+  while (jsCursor <= sourceText.length && lines.length < limit) {
+    const newlineIndex = sourceText.indexOf('\n', jsCursor)
+    const jsEnd = newlineIndex >= 0 ? newlineIndex : sourceText.length
+    const fullSourceEnd = trackUnicodeOffsets
+      ? sourceCursor + unicodeCodePointLength(sourceText, jsCursor, jsEnd)
+      : jsEnd
+    const focusedStart = focusLine === lineNumber && focusOffset != null
+      ? Math.max(sourceCursor, focusOffset - Math.floor(characterLimit / 3))
+      : sourceCursor
+    const segmentSourceStart = Math.min(
+      focusedStart,
+      Math.max(sourceCursor, fullSourceEnd - characterLimit),
+    )
+    const relativeSegmentStart = trackUnicodeOffsets
+      ? segmentSourceStart - sourceCursor
+      : Math.max(0, segmentSourceStart - jsCursor)
+    const segmentJsStart = advanceCodePoints(
+      sourceText,
+      jsCursor,
+      jsEnd,
+      relativeSegmentStart,
+    )
+    const segmentJsEnd = advanceCodePoints(
+      sourceText,
+      segmentJsStart,
+      jsEnd,
+      characterLimit,
+    )
+    const segmentLength = trackUnicodeOffsets
+      ? unicodeCodePointLength(sourceText, segmentJsStart, segmentJsEnd)
+      : segmentJsEnd - segmentJsStart
+    const start = trackUnicodeOffsets ? segmentSourceStart : segmentJsStart
+    const end = start + segmentLength
+    const content = `${segmentJsStart > jsCursor ? '… ' : ''}${sourceText.slice(segmentJsStart, segmentJsEnd)}${segmentJsEnd < jsEnd ? ' …' : ''}`
+    lines.push({ number: lineNumber, content, start, end })
+    sourceCursor = fullSourceEnd + (newlineIndex >= 0 ? 1 : 0)
+    jsCursor = newlineIndex >= 0 ? newlineIndex + 1 : sourceText.length + 1
+    lineNumber += 1
+  }
+
+  return {
+    lines,
+    startLine: lines[0]?.number ?? startLine,
+    endLine: lines[lines.length - 1]?.number ?? startLine,
+    hasPrevious: startLine > 1,
+    hasMore: jsCursor <= sourceText.length,
+  }
+}
+
+/** 根据后端 code point 偏移查找物理行号，不构建全文行数组。 */
+export function sourceLineNumberAtOffset(sourceText: string, targetOffset: number) {
+  const normalizedOffset = Math.max(0, Math.trunc(targetOffset) || 0)
+  let offset = 0
+  let lineNumber = 1
+  for (const character of sourceText) {
+    if (offset >= normalizedOffset) break
+    if (character === '\n') lineNumber += 1
+    offset += 1
+  }
+  return lineNumber
+}
+
+/**
+ * 手动新增项可能先以空内容保存为 invalid，编辑后又由后端标记为 modified，
+ * 因此不能只依赖可变的 status；空原文且完全没有来源定位才是稳定兜底。
+ */
+export function isManualPreviewItem(item: PreviewItem): boolean {
+  const hasSourceLocation = item.sourceStart != null
+    || item.sourceEnd != null
+    || item.sourceStartLine != null
+    || item.sourceEndLine != null
+    || Boolean(item.sourcePages?.length)
+    || Boolean(item.sourceLocator)
+  return item.status === 'manual' || (!item.originalContent && !hasSourceLocation)
 }
 
 /** 与后端预览 token 估算规则一致，仅用于编辑中的即时计数。 */

frontend/src/views/data-process/create/types.ts

@@ -24,6 +24,7 @@ export type PreprocessOption =
   | 'detect_structure'
   | 'deduplicate'
   | 'normalize_format'
+  /** 仅用于恢复历史任务，新任务界面不再提供。 */
   | 'filter_anomaly'
   | 'desensitize'
 
@@ -94,7 +95,6 @@ export interface ExternalDataSource {
 export interface UploadedDataFile {
   uid: string | number
   sourceFileId?: string
-  rawFile?: File
   name: string
   size: number
   count: number
@@ -118,6 +118,22 @@ export interface SourceLine {
   end: number
 }
 
+export type PreviewSourceLocatorKind = 'json' | 'jsonl' | 'csv' | 'xlsx'
+
+export interface PreviewSourceLocator {
+  kind: PreviewSourceLocatorKind
+  record_index?: number | null
+  start_line?: number | null
+  end_line?: number | null
+  source_start?: number | null
+  source_end?: number | null
+  json_pointer?: string | null
+  sheet_index?: number | null
+  sheet_name?: string | null
+  row_number?: number | null
+  sheet_record_index?: number | null
+}
+
 export interface PreviewItem {
   id: string
   sourceFileId: string
@@ -129,6 +145,8 @@ export interface PreviewItem {
   sourceStartLine: number | null
   sourceEndLine: number | null
   sourcePages?: number[]
+  sourceLocator?: PreviewSourceLocator
+  headingPath?: string[]
   tokenCount: number
   status: 'original' | 'modified' | 'manual' | 'invalid'
   qualityScore?: number

frontend/src/views/data-process/create/useDataProcessGeneration.ts

@@ -71,7 +71,11 @@ export function useDataProcessGeneration(bindings: GenerationBindings) {
   let generationTimer: ReturnType<typeof setTimeout> | null = null
   let generationRun = 0
   let pollFailureCount = 0
-  let generationStarting = false
+  const generationStarting = ref(false)
+  const generationRestoring = ref(false)
+  const canReturnFromGeneration = computed(() => (
+    generation.status === 'idle' && !generationStarting.value && !generationRestoring.value
+  ))
 
   function stopGenerationTimer() {
     generationRun += 1
@@ -170,14 +174,14 @@ export function useDataProcessGeneration(bindings: GenerationBindings) {
   }
 
   async function startGeneration() {
-    if (generationStarting || generation.status === 'running') return false
+    if (generationStarting.value || generation.status === 'running') return false
     const taskId = bindings.taskId.value
     if (!taskId) {
       ElMessage.error('任务尚未创建，请返回上一步重试')
       return false
     }
 
-    generationStarting = true
+    generationStarting.value = true
     let runId: number | null = null
     try {
       const canStart = await bindings.beforeGenerate?.()
@@ -204,17 +208,18 @@ export function useDataProcessGeneration(bindings: GenerationBindings) {
       generation.message = error instanceof Error ? error.message : '启动数据处理失败，请重试。'
       return false
     } finally {
-      generationStarting = false
+      generationStarting.value = false
     }
   }
 
   async function resumeGeneration() {
     const taskId = bindings.taskId.value
     if (!taskId) return
-    stopGenerationTimer()
-    const activeRunId = generationRun
-    pollFailureCount = 0
+    generationRestoring.value = true
     try {
+      stopGenerationTimer()
+      const activeRunId = generationRun
+      pollFailureCount = 0
       const progress = await getDataProcessProgress(taskId)
       if (activeRunId !== generationRun) return
       if (progress.status === 'running') {
@@ -236,6 +241,8 @@ export function useDataProcessGeneration(bindings: GenerationBindings) {
     } catch (error) {
       generation.status = 'failed'
       generation.message = error instanceof Error ? error.message : '查询任务进度失败，请重试。'
+    } finally {
+      generationRestoring.value = false
     }
   }
 
@@ -430,7 +437,9 @@ export function useDataProcessGeneration(bindings: GenerationBindings) {
 
   return {
     bulkRegeneration,
+    canReturnFromGeneration,
     generation,
+    generationStarting,
     regeneratingResultId,
     resultRegenerationBusy,
     results,

frontend/src/views/data-process/create/useDataProcessRegeneration.ts

@@ -2,7 +2,6 @@ import { computed, nextTick, ref, type Reactive, type Ref } from 'vue'
 import { useRoute } from 'vue-router'
 import {
   getDataProcessPreview,
-  getDataProcessSourceContent,
   getDataProcessTask,
   regenerateDataProcessTask,
 } from '@/api/modules/dataProcess'
@@ -15,7 +14,10 @@ import {
   createStructuredOptionsFromConfig,
   createUnstructuredOptionsFromConfig,
 } from './dataProcessCreateState'
-import { mapDataProcessSourceFile } from './useDataProcessSourceUpload'
+import {
+  loadCanonicalSourceContent,
+  mapDataProcessSourceFile,
+} from './useDataProcessSourceUpload'
 import type {
   PreviewItem,
   ProcessType,
@@ -50,24 +52,6 @@ interface RegenerationBindings {
   resetDownstream: () => void
 }
 
-async function loadSourceContent(taskId: string, fileId: string | number) {
-  const chunks: string[] = []
-  let startLine = 1
-  while (true) {
-    const source = await getDataProcessSourceContent(taskId, fileId, {
-      start_line: startLine,
-      line_count: 10_000,
-    })
-    chunks.push(source.content || '')
-    if (!source.has_more) break
-    const nextLine = Number(source.end_line || startLine) + 1
-    if (nextLine <= startLine) break
-    startLine = nextLine
-  }
-  // source_content_lines 已保留原始换行；分页之间直接拼接，避免凭空增加空行并破坏偏移。
-  return chunks.join('')
-}
-
 async function loadAllPreviews(taskId: string, mapPreviewItem: RegenerationBindings['mapPreviewItem']) {
   const first = await getDataProcessPreview(taskId, { page: 1, page_size: 500 })
   const items = [...first.items]
@@ -97,7 +81,7 @@ export function useDataProcessRegeneration(bindings: RegenerationBindings) {
   async function hydrateWorkspace(task: DataProcessTask, preservePreviews: boolean) {
     const taskId = String(task.id)
     bindings.uploadedFiles.value = await Promise.all((task.source_files || []).map(async (file) => (
-      mapDataProcessSourceFile(file, await loadSourceContent(taskId, file.id))
+      mapDataProcessSourceFile(file, await loadCanonicalSourceContent(taskId, file.id))
     )))
     bindings.previewItems.value = preservePreviews
       ? await loadAllPreviews(taskId, bindings.mapPreviewItem)

frontend/src/views/data-process/create/useDataProcessSourceUpload.ts

@@ -6,7 +6,6 @@ import {
 } from '@/api/modules/dataProcess'
 import type { ProcessType, UploadedDataFile } from './types'
 
-const BINARY_FILE_EXTENSIONS = new Set(['xlsx', 'pdf', 'docx', 'pptx'])
 const STRUCTURED_FILE_EXTENSIONS = new Set(['json', 'jsonl', 'ndjson', 'csv', 'tsv', 'xlsx'])
 const UNSTRUCTURED_FILE_EXTENSIONS = new Set([
   'txt', 'md', 'markdown', 'pdf', 'docx', 'pptx', 'json', 'jsonl', 'ndjson',
@@ -15,11 +14,11 @@ const LEGACY_OFFICE_EXTENSIONS = new Set(['doc', 'xls', 'ppt'])
 const MAX_SOURCE_FILE_BYTES = 200 * 1024 * 1024
 const MAX_SOURCE_FILE_COUNT = 20
 const MAX_SOURCE_BATCH_BYTES = 500 * 1024 * 1024
+const SOURCE_CONTENT_PAGE_CHARS = 1_000_000
 
 interface SourceUploadJob {
   uid: string
   file: File
-  extension: string
 }
 
 interface SourceUploadOptions {
@@ -60,9 +59,6 @@ export function validateSourceFileSelection(
         : '结构化数据支持 JSON、JSONL、NDJSON、CSV、TSV、XLSX',
     }
   }
-  if (selectedFiles.some((file) => file.name === raw.name && file.size === raw.size)) {
-    return { valid: false, severity: 'warning', message: '同名且同大小的文件已经选择' }
-  }
   if (selectedFiles.length >= MAX_SOURCE_FILE_COUNT) {
     return { valid: false, severity: 'warning', message: `每个任务最多选择 ${MAX_SOURCE_FILE_COUNT} 个文件` }
   }
@@ -73,6 +69,34 @@ export function validateSourceFileSelection(
   return { valid: true, extension }
 }
 
+function unicodeCodePointLength(value: string) {
+  let length = 0
+  for (const _character of value) length += 1
+  return length
+}
+
+/** 分页读取服务端保存的规范化正文，避免重新使用浏览器本地解码结果。 */
+export async function loadCanonicalSourceContent(
+  taskId: string | number,
+  fileId: string | number,
+) {
+  const chunks: string[] = []
+  let offset = 0
+  while (true) {
+    const source = await getDataProcessSourceContent(taskId, fileId, {
+      offset,
+      limit: SOURCE_CONTENT_PAGE_CHARS,
+    })
+    const content = source.content || ''
+    chunks.push(content)
+    if (!source.has_more) break
+    const nextOffset = Number(source.offset ?? offset) + unicodeCodePointLength(content)
+    if (nextOffset <= offset) throw new Error('服务端规范化内容分页异常，请删除文件后重试')
+    offset = nextOffset
+  }
+  return chunks.join('')
+}
+
 export function mapDataProcessSourceFile(
   file: DataProcessSourceFile,
   content = '',
@@ -126,16 +150,6 @@ export function useDataProcessSourceUpload(options: SourceUploadOptions) {
     pending.uploadError = undefined
 
     try {
-      let content = ''
-      if (!BINARY_FILE_EXTENSIONS.has(job.extension)) {
-        try {
-          content = new TextDecoder('utf-8', { fatal: true }).decode(await job.file.arrayBuffer())
-        } catch {
-          throw new Error('文本文件不是有效的 UTF-8 编码，请转换编码后重试')
-        }
-        if (!content.trim()) throw new Error('不能上传空文件')
-      }
-
       const uploaded = await uploadDataProcessSourceFiles(currentTaskId, [job.file], (progress) => {
         pending.uploadProgress = progress
       })
@@ -144,22 +158,13 @@ export function useDataProcessSourceUpload(options: SourceUploadOptions) {
 
       // 先登记后端 ID，确保正文读取失败时仍可正确删除已落库的文件。
       Object.assign(pending, mapDataProcessSourceFile(source), {
-        rawFile: job.file,
         status: 'uploading',
         uploadProgress: 99,
       })
-      if (BINARY_FILE_EXTENSIONS.has(job.extension)) {
-        try {
-          const parsed = await getDataProcessSourceContent(currentTaskId, source.id, {
-            start_line: 1,
-            line_count: 10_000,
-          })
-          pending.content = parsed.content
-        } catch {
-          // 原文件已经成功落库，正文稍后仍可由预览构建接口读取，不重复上传。
-        }
-      } else {
-        pending.content = content
+      try {
+        pending.content = await loadCanonicalSourceContent(currentTaskId, source.id)
+      } catch {
+        throw new Error('文件已上传，但服务端规范化内容读取失败，请删除文件后重试')
       }
 
       pending.status = 'ready'