feat(data_process): 问答对数据评测体系与质量分雷达图

- 三层评测：规则层沿用原五维规则分，语义层用本地 BGE 向量算问答/来源相关性，评审层复用生成模型按 rubric 打分（忠实度/正确性/清晰度等，区分 standard/reasoning/dpo 输出类型），任一层失败自动降级 - 组合分 = 规则 35% + 语义 20% + 评审 45%，缺层自动重归一 - 新增 results/evaluate-batch 批量评测接口，镜像批量重生成的并发、乐观锁与部分成功语义；生成阶段不再展示质量分 - 详情页与结果编辑页新增"数据评测"按钮和批量进度；质量分列悬停弹出雷达图浮窗（评审 5 维 + 语义 2 维、三层分项、评审理由） - 手动编辑/恢复后重算规则与语义层并丢弃过期评审分，雷达图不再展示失效数据
2026-08-19 14:21:54 +08:00
parent f47611020a
commit 81c2f85c3a
16 changed files with 1917 additions and 33 deletions
--- a/frontend/src/api/modules/dataProcess.ts
+++ b/frontend/src/api/modules/dataProcess.ts
@@ -20,6 +20,8 @@ import type {
  DataProcessPublishResult,
  DataProcessQualityScore,
  DataProcessResult,
+  DataProcessResultBatchEvaluatePayload,
+  DataProcessResultBatchEvaluateResult,
  DataProcessResultBatchRegeneratePayload,
  DataProcessResultBatchRegenerateResult,
  DataProcessResultRegeneratePayload,
@@ -320,5 +322,14 @@ export const regenerateDataProcessResults = (
  { timeout: 240_000 },
 )

+export const evaluateDataProcessResults = (
+  taskId: string | number,
+  payload: DataProcessResultBatchEvaluatePayload,
+) => post<DataProcessResultBatchEvaluateResult>(
+  `/data-process/${encodeURIComponent(taskId)}/results/evaluate-batch`,
+  payload,
+  { timeout: 240_000 },
+)
+
 export const publishDataProcess = (taskId: string | number, payload: DataProcessPublishPayload) =>
  post<DataProcessPublishResult>(`/data-process/${encodeURIComponent(taskId)}/publish`, payload)
--- a/frontend/src/plugins/echarts.ts
+++ b/frontend/src/plugins/echarts.ts
@@ -3,18 +3,21 @@
 */
 import { use } from 'echarts/core'
 import { CanvasRenderer } from 'echarts/renderers'
-import { BarChart, PieChart } from 'echarts/charts'
+import { BarChart, PieChart, RadarChart } from 'echarts/charts'
 import {
  GridComponent,
  TooltipComponent,
  LegendComponent,
+  RadarComponent,
 } from 'echarts/components'

 use([
  CanvasRenderer,
  BarChart,
  PieChart,
+  RadarChart,
  GridComponent,
  TooltipComponent,
  LegendComponent,
+  RadarComponent,
 ])
--- a/frontend/src/types/dataProcess.ts
+++ b/frontend/src/types/dataProcess.ts
@@ -398,6 +398,52 @@ export interface DataProcessResultBatchRegenerateResult {
  failures: DataProcessResultBatchRegenerateFailure[]
 }

+export interface DataProcessResultBatchEvaluateItem {
+  result_id: string
+  expected_updated_at: string
+}
+
+export interface DataProcessResultBatchEvaluatePayload {
+  items: DataProcessResultBatchEvaluateItem[]
+}
+
+export interface DataProcessResultBatchEvaluateFailure {
+  result_id: string
+  code: 'conflict' | 'skipped' | 'evaluation_failed' | 'internal_error'
+  message: string
+}
+
+export interface DataProcessResultBatchEvaluateResult {
+  batch_id: string
+  total: number
+  succeeded: number
+  failed: number
+  duration_ms: number
+  items: DataProcessResult[]
+  failures: DataProcessResultBatchEvaluateFailure[]
+}
+
+export interface DataProcessQualitySemantic {
+  question_answer?: number
+  answer_source?: number
+  overall?: number
+}
+
+export interface DataProcessQualityJudge {
+  scores?: Record<string, number>
+  overall?: number
+  reason?: string
+  issues?: string[]
+  model?: string
+  output_type?: string
+}
+
+export interface DataProcessQualityLayers {
+  rule?: number | null
+  semantic?: number | null
+  judge?: number | null
+}
+
 export interface DataProcessQualityScore {
  overall?: number
  completeness?: number
@@ -408,6 +454,11 @@ export interface DataProcessQualityScore {
  is_valid?: boolean
  flags?: string[]
  fingerprint?: string
+  semantic?: DataProcessQualitySemantic | null
+  judge?: DataProcessQualityJudge | null
+  layers?: DataProcessQualityLayers | null
+  evaluated?: boolean
+  evaluated_at?: string | null
  source_pages?: number[]
  heading_path?: string[]
  source_locator?: DataProcessSourceLocator
--- a/frontend/src/views/data-process/DataProcessCreateView.vue
+++ b/frontend/src/views/data-process/DataProcessCreateView.vue
@@ -18,6 +18,7 @@ import {
  previewAffectingOptionsFor,
 } from './create/dataProcessCreateState'
 import { useDataProcessGeneration } from './create/useDataProcessGeneration'
+import { useDataProcessEvaluation } from './create/useDataProcessEvaluation'
 import { useDataProcessPreviewBuild } from './create/useDataProcessPreviewBuild'
 import { useDataProcessRegeneration } from './create/useDataProcessRegeneration'
 import { createDefaultExternalSource, externalSourcePayload, restoreExternalSourceConfig, sourceConfigForBackend } from './create/externalSourceConfig'
@@ -126,6 +127,19 @@ const {
  outputType: activeOutputType,
  beforeGenerate: beforeStartGeneration,
 })
+const {
+  evaluation,
+  evaluateAllResults,
+  resetEvaluation,
+} = useDataProcessEvaluation({
+  taskId,
+  results,
+  selectedResultId,
+})
+// 生成结果被重置（重新切分/上传/重新生成配置）时同步清空评测进度。
+watch(results, (items) => {
+  if (!items.length) resetEvaluation()
+})
 const { enqueueSourceUpload, sourceUploading } = useDataProcessSourceUpload({
  taskId,
  uploadedFiles,
@@ -1156,10 +1170,12 @@ onMounted(() => {
            :preview-items="previewItems"
            :regenerating-result-id="regeneratingResultId"
            :bulk-regeneration="bulkRegeneration"
+            :evaluation="evaluation"
            :output-type="activeOutputType"
            @update:field="updateResultField"
            @regenerate:all="regenerateAllResults"
            @regenerate:item="regenerateResult"
+            @evaluate:all="evaluateAllResults"
          />
        </div>
      </div>
--- a/frontend/src/views/data-process/DataProcessDetailView.vue
+++ b/frontend/src/views/data-process/DataProcessDetailView.vue
@@ -5,6 +5,7 @@ import { ElMessage, ElMessageBox } from 'element-plus'
 import PageCard from '@/components/PageCard.vue'
 import { usePolling } from '@/composables/usePolling'
 import {
+  evaluateDataProcessResults,
  getDataProcessProgress,
  getDataProcessResults,
  getDataProcessTask,
@@ -13,6 +14,7 @@ import {
  restoreDataProcessResult,
  updateDataProcessResult,
 } from '@/api/modules/dataProcess'
+import QualityRadarPopover from './create/QualityRadarPopover.vue'
 import type {
  DataProcessDatasetSplit,
  DataProcessPublishPayload,
@@ -456,13 +458,101 @@ function resultStatusType(status: DataProcessResultStatus) {
 }

 function qualityScoreLabel(value: DataProcessResult['quality_score']) {
-  if (value == null) return '-'
+  if (value == null || !value.evaluated) return '-'
  const score = value.overall
  return Number.isFinite(score) ? Number(score).toFixed(1) : '-'
 }

-function qualityFlagsLabel(value: DataProcessResult['quality_score']) {
-  return value?.flags?.length ? value.flags.join('、') : '未命中质量规则'
+function qualityScoreTone(value: DataProcessResult['quality_score']) {
+  const score = Number(value?.overall)
+  if (!value?.evaluated || !Number.isFinite(score)) return ''
+  return score >= 80 ? 'is-success' : score >= 60 ? 'is-warning' : 'is-danger'
+}
+
+function qualityScoreEvaluated(value: DataProcessResult['quality_score']) {
+  return Boolean(value?.evaluated && Number.isFinite(Number(value?.overall)))
+}
+
+const evaluationRunning = ref(false)
+const evaluationProgress = reactive({
+  visible: false,
+  total: 0,
+  completed: 0,
+  succeeded: 0,
+  failed: 0,
+})
+// 与批量重生成一致的分块大小，单批在接口 240 秒超时预算内。
+const EVALUATION_CHUNK_SIZE = 12
+const canEvaluate = computed(() => (
+  detail.value?.status === 'completed' && !hasCurrentPublishedDataset.value
+))
+const evaluationPercentage = computed(() => (
+  evaluationProgress.total
+    ? Math.round((evaluationProgress.completed / evaluationProgress.total) * 100)
+    : 0
+))
+
+async function loadAllResultIds() {
+  const first = await getDataProcessResults(taskId.value, { page: 1, page_size: 500 })
+  const items = [...first.items]
+  const pages = Math.ceil(first.total / first.page_size)
+  for (let page = 2; page <= pages; page += 1) {
+    const next = await getDataProcessResults(taskId.value, { page, page_size: 500 })
+    items.push(...next.items)
+  }
+  return items
+}
+
+async function runResultEvaluation() {
+  if (evaluationRunning.value || !canEvaluate.value) return
+  evaluationRunning.value = true
+  Object.assign(evaluationProgress, {
+    visible: true,
+    total: 0,
+    completed: 0,
+    succeeded: 0,
+    failed: 0,
+  })
+  try {
+    const candidates = (await loadAllResultIds()).filter((item) => item.updated_at)
+    if (!candidates.length) {
+      ElMessage.info('当前没有可评测的结果')
+      return
+    }
+    evaluationProgress.total = candidates.length
+    for (let offset = 0; offset < candidates.length; offset += EVALUATION_CHUNK_SIZE) {
+      const chunk = candidates.slice(offset, offset + EVALUATION_CHUNK_SIZE)
+      try {
+        const evaluated = await evaluateDataProcessResults(taskId.value, {
+          items: chunk.map((item) => ({
+            result_id: String(item.id),
+            expected_updated_at: item.updated_at as string,
+          })),
+        })
+        evaluationProgress.completed += evaluated.total
+        evaluationProgress.succeeded += evaluated.succeeded
+        evaluationProgress.failed += evaluated.failed
+      } catch {
+        evaluationProgress.completed = evaluationProgress.total
+        evaluationProgress.failed += candidates.length - offset
+        break
+      }
+    }
+    await loadResults()
+    if (evaluationProgress.failed === 0) {
+      ElMessage.success(`数据评测完成：成功 ${evaluationProgress.succeeded} 条`)
+    } else if (evaluationProgress.succeeded > 0) {
+      ElMessage.warning(
+        `数据评测完成：成功 ${evaluationProgress.succeeded} 条，失败 ${evaluationProgress.failed} 条`,
+      )
+    } else {
+      ElMessage.error(`数据评测失败：${evaluationProgress.failed} 条结果未完成评测`)
+    }
+  } catch {
+    ElMessage.error('数据评测中断，已完成的评分保持不变')
+  } finally {
+    evaluationRunning.value = false
+  }
 }

 function replaceResult(updated: DataProcessResult) {
@@ -824,10 +914,33 @@ onBeforeUnmount(() => {
              <el-option label="已修改" value="modified" />
              <el-option label="无效" value="invalid" />
            </el-select>
+            <el-button
+              v-if="canEvaluate"
+              type="primary"
+              plain
+              :loading="evaluationRunning"
+              :disabled="resultLoading"
+              @click="runResultEvaluation"
+            >
+              <i v-if="!evaluationRunning" class="fa fa-check-square-o" aria-hidden="true" /> 数据评测
+            </el-button>
            <el-button :loading="resultLoading" @click="loadResults"><i class="fa fa-refresh" /></el-button>
          </div>
        </div>

+        <div v-if="evaluationProgress.visible" class="evaluation-progress">
+          <span>
+            数据评测 {{ evaluationProgress.completed }} / {{ evaluationProgress.total }}
+            · 成功 {{ evaluationProgress.succeeded }} · 失败 {{ evaluationProgress.failed }}
+          </span>
+          <el-progress
+            :percentage="evaluationPercentage"
+            :show-text="false"
+            :stroke-width="5"
+            :color="evaluationProgress.failed > 0 ? '#d97706' : '#5b50f2'"
+          />
+        </div>
+
        <el-table
          v-if="results.length"
          :data="results"
@@ -845,9 +958,26 @@ onBeforeUnmount(() => {
          </template>
          <el-table-column label="质量分" width="88" align="center">
            <template #default="{ row }">
-              <el-tooltip :content="qualityFlagsLabel((row as DataProcessResult).quality_score)">
-                <span>{{ qualityScoreLabel((row as DataProcessResult).quality_score) }}</span>
-              </el-tooltip>
+              <el-popover
+                v-if="qualityScoreEvaluated((row as DataProcessResult).quality_score)"
+                placement="top"
+                :width="296"
+                trigger="hover"
+                :show-after="150"
+                popper-class="quality-radar-popper"
+              >
+                <template #reference>
+                  <span
+                    class="detail-quality-score"
+                    :class="qualityScoreTone((row as DataProcessResult).quality_score)"
+                  >{{ qualityScoreLabel((row as DataProcessResult).quality_score) }}</span>
+                </template>
+                <QualityRadarPopover
+                  :quality="(row as DataProcessResult).quality_score!"
+                  :score="Number((row as DataProcessResult).quality_score?.overall)"
+                />
+              </el-popover>
+              <span v-else class="detail-quality-empty">{{ qualityScoreLabel((row as DataProcessResult).quality_score) }}</span>
            </template>
          </el-table-column>
          <el-table-column label="状态" width="90" align="center">
@@ -1099,6 +1229,35 @@ onBeforeUnmount(() => {
 .result-filters :deep(.el-select) { width: 120px; }
 .result-section :deep(.el-table) { border-radius: 0; }

+.evaluation-progress {
+  display: grid;
+  grid-template-columns: 1fr 220px;
+  align-items: center;
+  gap: 14px;
+  padding: 10px 18px;
+  color: #667085;
+  background: #f8f9fc;
+  font-size: 12px;
+}
+
+.detail-quality-score {
+  display: inline-block;
+  min-width: 44px;
+  padding: 2px 8px;
+  border-radius: 10px;
+  color: #475467;
+  background: #f2f4f7;
+  font-weight: 700;
+  font-variant-numeric: tabular-nums;
+  cursor: default;
+
+  &.is-success { color: #067647; background: #e6f4ee; }
+  &.is-warning { color: #b54708; background: #fef0c7; }
+  &.is-danger { color: #b42318; background: #fee4e2; }
+}
+
+.detail-quality-empty { color: #98a2b3; }
+
 :global(.data-process-result-tooltip) {
  box-sizing: border-box;
  max-width: min(520px, calc(100vw - 32px));
--- a/frontend/src/views/data-process/create/QualityRadarPopover.vue
+++ b/frontend/src/views/data-process/create/QualityRadarPopover.vue
@@ -0,0 +1,252 @@
+<script setup lang="ts">
+import { computed } from 'vue'
+import VChart from 'vue-echarts'
+import '@/plugins/echarts'
+import type { EChartsOption } from 'echarts'
+import type { ResultQualityDetails } from './types'
+
+const props = defineProps<{
+  quality: ResultQualityDetails
+  score?: number
+}>()
+
+// 轴标签按词意预置断行，避免长中文标签把雷达网格挤偏或被机械切词。
+const JUDGE_DIMENSION_LABELS: Record<string, string> = {
+  faithfulness: '忠实度',
+  correctness: '正确性',
+  clarity: '问题\n清晰度',
+  completeness: '回答\n完整性',
+  alignment: '指令\n对齐',
+  reasoning_validity: '推理\n有效性',
+  chosen_quality: 'chosen\n质量',
+  rejected_quality: 'rejected\n质量',
+  preference_reasonableness: '偏好\n区分',
+}
+
+const SEMANTIC_DIMENSION_LABELS: Record<string, string> = {
+  question_answer: '问答\n相关',
+  answer_source: '来源\n覆盖',
+}
+
+interface RadarDimension {
+  name: string
+  value: number
+}
+
+const radarDimensions = computed<RadarDimension[]>(() => {
+  const dimensions: RadarDimension[] = []
+  for (const [key, value] of Object.entries(props.quality?.judge?.scores ?? {})) {
+    dimensions.push({
+      name: JUDGE_DIMENSION_LABELS[key] ?? key,
+      value: Math.round(value * 20),
+    })
+  }
+  for (const [key, value] of Object.entries(props.quality?.semantic ?? {})) {
+    if (key === 'overall' || typeof value !== 'number') continue
+    dimensions.push({
+      name: SEMANTIC_DIMENSION_LABELS[key] ?? key,
+      value: Math.round(value),
+    })
+  }
+  return dimensions
+})
+
+// 可用维度太少时雷达图失去意义，降级为分层分数展示。
+const showRadar = computed(() => radarDimensions.value.length >= 3)
+
+const radarOption = computed<EChartsOption>(() => ({
+  radar: {
+    indicator: radarDimensions.value.map((dimension) => ({
+      name: dimension.name,
+      max: 100,
+    })),
+    radius: '56%',
+    center: ['50%', '50%'],
+    splitNumber: 4,
+    axisName: {
+      color: '#667085',
+      fontSize: 10,
+      lineHeight: 13,
+    },
+    splitArea: { areaStyle: { color: ['#fbfbfd', '#f2f4f8'] } },
+    splitLine: { lineStyle: { color: '#e4e7ec' } },
+    axisLine: { lineStyle: { color: '#e4e7ec' } },
+  },
+  series: [{
+    type: 'radar',
+    symbol: 'circle',
+    symbolSize: 3,
+    data: [{
+      value: radarDimensions.value.map((dimension) => dimension.value),
+      name: '质量维度',
+      areaStyle: { color: 'rgba(91, 80, 242, 0.18)' },
+      lineStyle: { color: '#5b50f2', width: 1.5 },
+      itemStyle: { color: '#5b50f2' },
+    }],
+  }],
+}))
+
+const layerScores = computed(() => {
+  const layers = props.quality?.layers ?? {}
+  return [
+    { label: '规则层', value: layers.rule },
+    { label: '语义层', value: layers.semantic },
+    { label: '评审层', value: layers.judge },
+  ].filter((layer): layer is { label: string; value: number } => (
+    typeof layer.value === 'number'
+  ))
+})
+
+const displayScore = computed(() => {
+  if (typeof props.score === 'number' && !isNaN(props.score)) {
+    return props.score.toFixed(1)
+  }
+  return null
+})
+
+const scoreTone = computed(() => {
+  const numScore = props.score ?? 0
+  return numScore >= 80 ? 'is-success' : numScore >= 60 ? 'is-warning' : 'is-danger'
+})
+</script>
+
+<template>
+  <div class="quality-popover">
+    <div v-if="displayScore !== null" class="popover-header">
+      <strong>质量评测</strong>
+      <span class="popover-score" :class="scoreTone">{{ displayScore }}</span>
+    </div>
+
+    <VChart
+      v-if="showRadar"
+      class="quality-radar"
+      :option="radarOption"
+      autoresize
+    />
+    <div v-else class="radar-fallback">
+      维度数据不足，已评测维度少于 3 个时以分层分数为准。
+    </div>
+
+    <div class="layer-scores">
+      <div v-for="layer in layerScores" :key="layer.label" class="layer-item">
+        <span>{{ layer.label }}</span>
+        <el-progress
+          :percentage="Math.round(layer.value)"
+          :stroke-width="6"
+          :show-text="false"
+          :color="layer.value >= 80 ? '#12b76a' : layer.value >= 60 ? '#f0b429' : '#d92d20'"
+        />
+        <em>{{ layer.value.toFixed(0) }}</em>
+      </div>
+    </div>
+
+    <p v-if="quality?.judge?.reason" class="judge-reason">{{ quality.judge.reason }}</p>
+    <div v-if="quality?.judge?.issues?.length" class="judge-issues">
+      <span v-for="issue in quality.judge.issues" :key="issue" class="issue-tag">{{ issue }}</span>
+    </div>
+    <div v-if="quality?.judge?.model" class="judge-model">评审模型：{{ quality.judge.model }}</div>
+  </div>
+</template>
+
+<style scoped lang="scss">
+.quality-popover {
+  display: flex;
+  flex-direction: column;
+  gap: 10px;
+  width: 100%;
+  box-sizing: border-box;
+}
+
+.popover-header {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  padding-bottom: 6px;
+  border-bottom: 1px solid #f2f4f7;
+
+  strong {
+    color: #344054;
+    font-size: 13px;
+  }
+}
+
+.popover-score {
+  color: #344054;
+  font-size: 18px;
+  font-weight: 700;
+  font-variant-numeric: tabular-nums;
+
+  &.is-success { color: #12b76a; }
+  &.is-warning { color: #d99b0b; }
+  &.is-danger { color: #d92d20; }
+}
+
+.quality-radar {
+  width: 100%;
+  height: 220px;
+  box-sizing: border-box;
+}
+
+.radar-fallback {
+  padding: 18px 10px;
+  color: #98a2b3;
+  font-size: 12px;
+  text-align: center;
+}
+
+.layer-scores {
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+}
+
+.layer-item {
+  display: grid;
+  grid-template-columns: 44px 1fr 28px;
+  align-items: center;
+  gap: 8px;
+  color: #667085;
+  font-size: 11px;
+
+  em {
+    color: #344054;
+    font-style: normal;
+    font-weight: 600;
+    text-align: right;
+    font-variant-numeric: tabular-nums;
+  }
+}
+
+.judge-reason {
+  margin: 0;
+  color: #475467;
+  font-size: 12px;
+  line-height: 1.6;
+}
+
+.judge-issues {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 4px;
+}
+
+.issue-tag {
+  padding: 2px 8px;
+  color: #b54708;
+  background: #fef0c7;
+  border-radius: 3px;
+  font-size: 11px;
+}
+
+.judge-model {
+  color: #98a2b3;
+  font-size: 11px;
+}
+</style>
+
+<style lang="scss">
+.quality-radar-popper {
+  padding: 14px 16px !important;
+}
+</style>
+
--- a/frontend/src/views/data-process/create/ResultEditorStep.vue
+++ b/frontend/src/views/data-process/create/ResultEditorStep.vue
@@ -1,6 +1,7 @@
 <script setup lang="ts">
 import { computed, ref } from 'vue'
-import type { BulkResultRegenerationState, PreviewItem, ResultItem } from './types'
+import QualityRadarPopover from './QualityRadarPopover.vue'
+import type { BulkResultRegenerationState, PreviewItem, ResultEvaluationState, ResultItem } from './types'
 import type { DataProcessOutputType } from '@/types/dataProcess'

 const props = defineProps<{
@@ -9,6 +10,7 @@ const props = defineProps<{
  selectedId: string | null
  regeneratingResultId: string | null
  bulkRegeneration: BulkResultRegenerationState
+  evaluation: ResultEvaluationState
  outputType: DataProcessOutputType
 }>()

@@ -17,6 +19,7 @@ const emit = defineEmits<{
  'update:field': [id: string, field: 'instruction' | 'input' | 'output' | 'chosen' | 'rejected', value: string]
  'regenerate:item': [id: string]
  'regenerate:all': []
+  'evaluate:all': []
 }>()

 const search = ref('')
@@ -30,6 +33,15 @@ const bulkRegenerationActive = computed(() => props.bulkRegeneration.status ===
 const bulkRegenerationVisible = computed(() => (
  props.bulkRegeneration.status !== 'idle' && props.bulkRegeneration.total > 0
 ))
+const evaluationActive = computed(() => props.evaluation.status === 'running')
+const evaluationVisible = computed(() => (
+  props.evaluation.status !== 'idle' && props.evaluation.total > 0
+))
+const evaluationPercentage = computed(() => {
+  if (!props.evaluation.total) return 0
+  return Math.round((props.evaluation.completed / props.evaluation.total) * 100)
+})
+const evaluatedCount = computed(() => props.items.filter((item) => item.qualityDetails?.evaluated).length)
 const bulkRegenerationPercentage = computed(() => (
  props.bulkRegeneration.total > 0
    ? Math.round((props.bulkRegeneration.completed / props.bulkRegeneration.total) * 100)
@@ -98,21 +110,47 @@ function selectRelative(offset: number) {
 <template>
  <section class="result-step">
    <div class="result-workspace">
-      <aside class="result-list-pane" :class="{ 'has-bulk-progress': bulkRegenerationVisible }">
+      <aside class="result-list-pane" :class="{ 'has-bulk-progress': bulkRegenerationVisible || evaluationVisible }">
        <div class="pane-header result-list-header">
-          <div class="result-list-title"><strong>生成结果</strong><span>共 {{ items.length }} 条</span></div>
-          <el-button
-            v-if="invalidCount > 0"
-            size="small"
-            plain
-            type="primary"
-            :loading="bulkRegenerationActive"
-            :disabled="Boolean(regeneratingResultId) || bulkRegenerationActive"
-            @click="emit('regenerate:all')"
-          >
-            <i v-if="!bulkRegenerationActive" class="fa fa-refresh" style="margin-right: 4px;" />
-            {{ bulkRegenerationActive ? '重新生成中' : `全部重新生成（${invalidCount}）` }}
-          </el-button>
+          <div class="result-list-title">
+            <strong>生成结果</strong><span>共 {{ items.length }} 条<template v-if="evaluatedCount"> · 已评测 {{ evaluatedCount }}</template></span>
+          </div>
+          <div class="result-list-actions">
+            <el-button
+              size="small"
+              plain
+              :loading="evaluationActive"
+              :disabled="!items.length || bulkRegenerationActive || Boolean(regeneratingResultId)"
+              @click="emit('evaluate:all')"
+            >
+              <i v-if="!evaluationActive" class="fa fa-check-square-o" style="margin-right: 4px;" />
+              数据评测
+            </el-button>
+            <el-button
+              v-if="invalidCount > 0"
+              size="small"
+              plain
+              type="primary"
+              :loading="bulkRegenerationActive"
+              :disabled="Boolean(regeneratingResultId) || bulkRegenerationActive || evaluationActive"
+              @click="emit('regenerate:all')"
+            >
+              <i v-if="!bulkRegenerationActive" class="fa fa-refresh" style="margin-right: 4px;" />
+              {{ bulkRegenerationActive ? '重新生成中' : `全部重新生成（${invalidCount}）` }}
+            </el-button>
+          </div>
+        </div>
+        <div v-if="evaluationVisible" class="bulk-regeneration-progress">
+          <div>
+            <span>数据评测 {{ evaluation.completed }} / {{ evaluation.total }}</span>
+            <span>成功 {{ evaluation.succeeded }} · 失败 {{ evaluation.failed }}</span>
+          </div>
+          <el-progress
+            :percentage="evaluationPercentage"
+            :show-text="false"
+            :stroke-width="5"
+            :color="evaluation.failed > 0 ? '#d97706' : '#5b50f2'"
+          />
        </div>
        <div v-if="bulkRegenerationVisible" class="bulk-regeneration-progress">
          <div>
@@ -146,6 +184,23 @@ function selectRelative(offset: number) {
              <strong>{{ item.instruction || '未填写指令' }}</strong>
              <small>{{ outputType === 'dpo' ? (item.chosen || '未填写 Chosen') : (item.output || '未填写输出') }}</small>
            </span>
+            <el-popover
+              v-if="item.qualityScore != null && item.qualityDetails"
+              placement="right"
+              :width="296"
+              trigger="hover"
+              :show-after="150"
+              popper-class="quality-radar-popper"
+            >
+              <template #reference>
+                <span
+                  class="result-score"
+                  :class="item.qualityScore >= 80 ? 'is-success' : item.qualityScore >= 60 ? 'is-warning' : 'is-danger'"
+                  @click.stop
+                >{{ item.qualityScore.toFixed(0) }}</span>
+              </template>
+              <QualityRadarPopover :quality="item.qualityDetails" :score="item.qualityScore" />
+            </el-popover>
            <i v-if="itemRegenerating(item.id)" class="css-spinner" />
            <i
              v-else
@@ -303,6 +358,42 @@ function selectRelative(offset: number) {
  }
 }

+.result-list-actions {
+  display: flex;
+  flex: none;
+  align-items: center;
+  gap: 8px;
+}
+
+.result-score {
+  flex: none;
+  min-width: 34px;
+  padding: 2px 8px;
+  border-radius: 10px;
+  color: #475467;
+  background: #f2f4f7;
+  font-size: 12px;
+  font-weight: 700;
+  font-variant-numeric: tabular-nums;
+  text-align: center;
+  cursor: default;
+
+  &.is-success {
+    color: #067647;
+    background: #e6f4ee;
+  }
+
+  &.is-warning {
+    color: #b54708;
+    background: #fef0c7;
+  }
+
+  &.is-danger {
+    color: #b42318;
+    background: #fee4e2;
+  }
+}
+
 .pane-header {
  display: flex;
  align-items: center;
--- a/frontend/src/views/data-process/create/types.ts
+++ b/frontend/src/views/data-process/create/types.ts
@@ -1,6 +1,9 @@
 import type {
  DataProcessOutputType,
  DataProcessPreviewFileStatus,
+  DataProcessQualityJudge,
+  DataProcessQualityLayers,
+  DataProcessQualitySemantic,
  DataProcessReasoningDetail,
 } from '@/types/dataProcess'

@@ -165,6 +168,14 @@ export interface GenerationState {
  message: string
 }

+export interface ResultQualityDetails {
+  semantic?: DataProcessQualitySemantic | null
+  judge?: DataProcessQualityJudge | null
+  layers?: DataProcessQualityLayers | null
+  evaluated?: boolean
+  flags?: string[]
+}
+
 export interface ResultItem {
  id: string
  previewItemId: string | null
@@ -188,7 +199,7 @@ export interface ResultItem {
  error?: string
  split?: 'train' | 'validation' | 'test'
  qualityScore?: number
-  qualityDetails?: Record<string, number>
+  qualityDetails?: ResultQualityDetails
  updatedAt?: string
 }

@@ -201,3 +212,11 @@ export interface BulkResultRegenerationState {
  targetIds: string[]
  failedIds: string[]
 }
+
+export interface ResultEvaluationState {
+  status: 'idle' | 'running' | 'completed' | 'partial' | 'failed'
+  total: number
+  completed: number
+  succeeded: number
+  failed: number
+}
--- a/frontend/src/views/data-process/create/useDataProcessEvaluation.ts
+++ b/frontend/src/views/data-process/create/useDataProcessEvaluation.ts
@@ -0,0 +1,126 @@
+import { computed, reactive, type Ref } from 'vue'
+import { ElMessage } from 'element-plus'
+import { evaluateDataProcessResults } from '@/api/modules/dataProcess'
+import { mapResult } from './useDataProcessGeneration'
+import type { ResultEvaluationState, ResultItem } from './types'
+
+interface EvaluationBindings {
+  taskId: Ref<string | null>
+  results: Ref<ResultItem[]>
+  selectedResultId: Ref<string | null>
+}
+
+// 与批量重新生成一致的分块大小：4 个后端 worker 消费三轮，
+// 单条评测最长 60 秒，12 条在批量接口 240 秒超时预算内。
+const EVALUATION_CHUNK_SIZE = 12
+
+function hasUnsavedChanges(item: ResultItem) {
+  return item.instruction !== item.savedInstruction
+    || item.input !== item.savedInput
+    || item.output !== item.savedOutput
+    || item.chosen !== item.savedChosen
+    || item.rejected !== item.savedRejected
+}
+
+export function useDataProcessEvaluation(bindings: EvaluationBindings) {
+  const evaluation = reactive<ResultEvaluationState>({
+    status: 'idle',
+    total: 0,
+    completed: 0,
+    succeeded: 0,
+    failed: 0,
+  })
+  const evaluationBusy = computed(() => evaluation.status === 'running')
+
+  function resetEvaluation() {
+    Object.assign(evaluation, {
+      status: 'idle',
+      total: 0,
+      completed: 0,
+      succeeded: 0,
+      failed: 0,
+    })
+  }
+
+  async function evaluateAllResults() {
+    const taskId = bindings.taskId.value
+    if (!taskId) return false
+    if (evaluationBusy.value) {
+      ElMessage.warning('请等待当前数据评测完成')
+      return false
+    }
+
+    const candidates = bindings.results.value.filter((item) => item.updatedAt)
+    if (!candidates.length) {
+      ElMessage.info('当前没有可评测的结果')
+      return false
+    }
+    const unsaved = candidates.find(hasUnsavedChanges)
+    if (unsaved) {
+      bindings.selectedResultId.value = unsaved.id
+      ElMessage.warning('存在未保存的修改，请先保存后再进行数据评测')
+      return false
+    }
+
+    Object.assign(evaluation, {
+      status: 'running',
+      total: candidates.length,
+      completed: 0,
+      succeeded: 0,
+      failed: 0,
+    })
+
+    let interrupted = false
+    try {
+      for (let offset = 0; offset < candidates.length; offset += EVALUATION_CHUNK_SIZE) {
+        const chunk = candidates.slice(offset, offset + EVALUATION_CHUNK_SIZE)
+        try {
+          const evaluated = await evaluateDataProcessResults(taskId, {
+            items: chunk.map((item) => ({
+              result_id: item.id,
+              expected_updated_at: item.updatedAt as string,
+            })),
+          })
+          for (const item of evaluated.items) {
+            const index = bindings.results.value.findIndex((entry) => entry.id === String(item.id))
+            if (index >= 0) bindings.results.value[index] = mapResult(item)
+          }
+          evaluation.completed += evaluated.total
+          evaluation.succeeded += evaluated.succeeded
+          evaluation.failed += evaluated.failed
+        } catch {
+          const remaining = candidates.slice(offset)
+          evaluation.completed = evaluation.total
+          evaluation.failed += remaining.length
+          interrupted = true
+          break
+        }
+      }
+
+      if (!interrupted && evaluation.failed === 0) {
+        evaluation.status = 'completed'
+        ElMessage.success(`数据评测完成：成功 ${evaluation.succeeded} 条`)
+      } else if (evaluation.succeeded > 0) {
+        evaluation.status = 'partial'
+        ElMessage.warning(
+          `数据评测完成：成功 ${evaluation.succeeded} 条，失败 ${evaluation.failed} 条`,
+        )
+      } else {
+        evaluation.status = 'failed'
+        ElMessage.error(`数据评测失败：${evaluation.failed} 条结果未完成评测`)
+      }
+      return evaluation.failed === 0
+    } catch {
+      evaluation.status = 'failed'
+      ElMessage.error('批量数据评测意外中断，已完成的评分保持不变')
+      return false
+    }
+  }
+
+  return {
+    evaluation,
+    evaluationBusy,
+    evaluateAllResults,
+    resetEvaluation,
+  }
+}
--- a/frontend/src/views/data-process/create/useDataProcessGeneration.ts
+++ b/frontend/src/views/data-process/create/useDataProcessGeneration.ts
@@ -28,6 +28,7 @@ const POLL_INTERVAL_MS = 1500
 const BULK_REGENERATION_CHUNK_SIZE = 12

 function mapResult(item: DataProcessResult): ResultItem {
+  const quality = item.quality_score
  return {
    id: String(item.id),
    previewItemId: item.preview_item_id == null ? null : String(item.preview_item_id),
@@ -50,11 +51,21 @@ function mapResult(item: DataProcessResult): ResultItem {
    status: item.status,
    error: item.error || undefined,
    split: item.split || undefined,
-    qualityScore: item.quality_score?.overall,
+    // 生成阶段只有内部规则分，界面不展示；数据评测完成后才显示组合分。
+    qualityScore: quality?.evaluated ? quality.overall : undefined,
+    qualityDetails: {
+      semantic: quality?.semantic ?? null,
+      judge: quality?.judge ?? null,
+      layers: quality?.layers ?? null,
+      evaluated: Boolean(quality?.evaluated),
+      flags: quality?.flags ?? [],
+    },
    updatedAt: item.updated_at,
  }
 }

+export { mapResult }
+
 export function useDataProcessGeneration(bindings: GenerationBindings) {
  const results = ref<ResultItem[]>([])
  const selectedResultId = ref<string | null>(null)