diff --git a/packages/node-runtime/src/semantic-index/service.test.ts b/packages/node-runtime/src/semantic-index/service.test.ts
index 561be344..0589d394 100644
--- a/packages/node-runtime/src/semantic-index/service.test.ts
+++ b/packages/node-runtime/src/semantic-index/service.test.ts
@@ -14,6 +14,9 @@ import type { SessionRuntimeAdapter } from '../services/adapters'
 import type { LocalPipelineFactory } from './embedding/local'
 
 const SESSION_ID = 'sess1'
+// 真实秒级基准时间（2023-11-14），用于让证据 snippet 渲染出真实年份，
+// 暴露「毫秒被当作秒再 *1000」导致五位数年份（如 56150）的回归。
+const BASE_TS_SECONDS = 1_700_000_000
 
 function setup(opts?: { embedDelayMs?: number }) {
   const baseDir = fs.existsSync('/private/tmp') ? '/private/tmp' : os.tmpdir()
@@ -28,7 +31,7 @@ function setup(opts?: { embedDelayMs?: number }) {
   `)
   const insert = db.prepare('INSERT INTO message (id, sender_id, ts, type, content) VALUES (?, ?, ?, 0, ?)')
   for (let i = 1; i <= 40; i++) {
-    insert.run(i, (i % 2) + 1, i * 60, `第${i}条关于项目排期和需求讨论的消息内容`)
+    insert.run(i, (i % 2) + 1, BASE_TS_SECONDS + i * 60, `第${i}条关于项目排期和需求讨论的消息内容`)
   }
   buildFtsIndex(db)
 
@@ -343,6 +346,13 @@ test('searchForTool desensitizes + anonymizes via pipeline, never leaks raw', as
   }
   // 元数据不得夹带原始消息
   assert.equal(JSON.stringify(result.sources).includes('rawMessages'), false)
+
+  // 回归：snippet 时间口径正确——证据消息 ts 是毫秒，预处理管道按秒渲染（内部 *1000），
+  // 服务必须把毫秒转回秒，否则会渲染出五位数年份（如 2023 -> 56xxx）。
+  const fiveDigitYear = /\b\d{5}\/\d{1,2}\/\d{1,2}/
+  assert.equal(fiveDigitYear.test(result.text), false)
+  assert.ok(result.text.includes('2023/'))
+
   service.close()
   db.close()
 })
diff --git a/packages/node-runtime/src/semantic-index/service.ts b/packages/node-runtime/src/semantic-index/service.ts
index 8b235f43..155bb845 100644
--- a/packages/node-runtime/src/semantic-index/service.ts
+++ b/packages/node-runtime/src/semantic-index/service.ts
@@ -524,7 +524,8 @@ export class SemanticIndexService {
         senderName: m.senderName,
         senderPlatformId: m.senderPlatformId,
         content: m.content,
-        timestamp: m.ts,
+        // EvidenceMessage.ts 是毫秒；预处理管道按秒渲染时间（format.ts 内部 *1000），故此处回退为秒
+        timestamp: Math.floor(m.ts / 1000),
       }))
       for (const m of block.messages) {
         if (m.senderId != null && !nameMap.has(m.senderId)) {
diff --git a/packages/tools/src/definitions/retrieve-chat-evidence.test.ts b/packages/tools/src/definitions/retrieve-chat-evidence.test.ts
index ff62621e..f640ef0d 100644
--- a/packages/tools/src/definitions/retrieve-chat-evidence.test.ts
+++ b/packages/tools/src/definitions/retrieve-chat-evidence.test.ts
@@ -197,6 +197,35 @@ describe('retrieveChatEvidenceTool candidates & grouping', () => {
     assert.equal(source.timestamp, 1714521600 * 1000)
   })
 
+  it('strips per-line time prefixes from semantic snippets (display shows one group time)', async () => {
+    const service = makeService({
+      searchForTool: async () =>
+        semanticResult({
+          sources: [
+            {
+              startMessageId: 100,
+              endMessageId: 110,
+              score: 0.9,
+              chunkIds: ['c1'],
+              // 预处理管道渲染格式：每行「时间 发送者: 内容」
+              snippet: '2024/5/1 08:00:00 甲: 到了乐山\n2024/5/1 09:00:00 乙: 住了一晚',
+              startTime: '2024-05-01T00:00:00.000Z',
+              endTime: '2024-05-01T01:00:00.000Z',
+            },
+          ],
+        }),
+    })
+    const res = await retrieveChatEvidenceTool.handler(
+      { query: 'q', criteria: 'c', mode: 'semantic' },
+      makeContext({ semanticIndexService: service })
+    )
+    const snippet = getPayload(res.data).groups[0].sources[0].snippet
+    assert.equal(snippet.includes('2024/5/1'), false)
+    assert.equal(snippet.includes('08:00:00'), false)
+    assert.ok(snippet.includes('甲: 到了乐山'))
+    assert.ok(snippet.includes('乙: 住了一晚'))
+  })
+
   it('desensitizes keyword snippets and never persists raw secret content', async () => {
     const desensitize = (messages: RawMessage[]): RawMessage[] =>
       messages.map((m) => ({ ...m, content: (m.content ?? '').replace('13800000000', '***') }))
diff --git a/packages/tools/src/definitions/retrieve-chat-evidence.ts b/packages/tools/src/definitions/retrieve-chat-evidence.ts
index fd676d06..31fc2112 100644
--- a/packages/tools/src/definitions/retrieve-chat-evidence.ts
+++ b/packages/tools/src/definitions/retrieve-chat-evidence.ts
@@ -94,6 +94,20 @@ function truncateSnippet(text: string): string {
   return t.length > EVIDENCE_SNIPPET_MAX_CHARS ? `${t.slice(0, EVIDENCE_SNIPPET_MAX_CHARS)}…` : t
 }
 
+/**
+ * 语义 snippet 由预处理管道渲染为多行「时间 发送者: 内容」，时间对证据块展示是噪声
+ * （分组已给出总时间）。这里去掉每行行首的时间前缀，仅保留「发送者: 内容」。
+ * 覆盖 zh-CN（`2025/3/3 07:25:04`）与 en-US（`3/3/2025, 7:25:04 AM`）两种 toLocaleString 输出。
+ */
+const LINE_TIME_PREFIX = /^\s*\d{1,4}[/-]\d{1,2}[/-]\d{1,4},?\s+\d{1,2}:\d{2}(?::\d{2})?(?:\s?[APap][Mm])?\s+/
+function stripLineTimestamps(text: string): string {
+  return text
+    .split('\n')
+    .map((line) => line.replace(LINE_TIME_PREFIX, ''))
+    .join('\n')
+    .trim()
+}
+
 function parseDateMs(s?: unknown): number | undefined {
   if (typeof s !== 'string' || !s.trim()) return undefined
   const d = new Date(s.trim().replace(' ', 'T'))
@@ -367,7 +381,7 @@ async function handler(params: Record<string, unknown>, context: ToolExecutionCo
             timestamp: startMs ?? 0,
             rangeStartMs: startMs ?? 0,
             rangeEndMs: endMs ?? startMs ?? 0,
-            snippet: truncateSnippet(s.snippet),
+            snippet: truncateSnippet(stripLineTimestamps(s.snippet)),
             sourceKind: 'semantic',
             score: s.score,
           })
diff --git a/src/components/AIChat/chat/EvidenceBlock.vue b/src/components/AIChat/chat/EvidenceBlock.vue
index 841775f1..41557a13 100644
--- a/src/components/AIChat/chat/EvidenceBlock.vue
+++ b/src/components/AIChat/chat/EvidenceBlock.vue
@@ -1,8 +1,7 @@
 <script setup lang="ts">
 import { computed, ref } from 'vue'
 import { useI18n } from 'vue-i18n'
-import dayjs from 'dayjs'
-import type { ChatEvidencePayload, ChatEvidenceGroup, EvidenceStatus } from '@openchatlab/core'
+import type { ChatEvidencePayload, EvidenceStatus } from '@openchatlab/core'
 import { useLayoutStore } from '@/stores/layout'
 
 const props = defineProps<{
@@ -43,20 +42,6 @@ const statusDotClass: Record<EvidenceStatus, string> = {
   excluded: 'bg-gray-400',
 }
 
-function formatGroupTimeRange(group: ChatEvidenceGroup): string | null {
-  let startTs = group.timeRange?.startTs
-  let endTs = group.timeRange?.endTs
-  if (startTs == null || endTs == null) {
-    const timestamps = group.sources.map((source) => source.timestamp).filter((ts) => Number.isFinite(ts))
-    if (timestamps.length === 0) return null
-    startTs = Math.min(...timestamps)
-    endTs = Math.max(...timestamps)
-  }
-  const start = dayjs(startTs).format('YYYY-MM-DD')
-  const end = dayjs(endTs).format('YYYY-MM-DD')
-  return start === end ? start : `${start} ~ ${end}`
-}
-
 function viewSource(messageId: number): void {
   layoutStore.openChatRecordDrawer({ scrollToMessageId: messageId })
 }
@@ -143,13 +128,7 @@ function viewSource(messageId: number): void {
             <span class="rounded px-1.5 py-0.5 text-[11px] font-medium" :class="statusBadgeClass[group.status]">
               {{ t(`ai.chat.evidence.group.${group.status}`) }}
             </span>
-            <span class="truncate text-xs font-medium text-gray-700 dark:text-gray-200">{{ group.title }}</span>
-            <span
-              v-if="formatGroupTimeRange(group)"
-              class="ml-auto shrink-0 text-[11px] text-gray-400 dark:text-gray-500"
-            >
-              {{ formatGroupTimeRange(group) }}
-            </span>
+            <span class="ml-auto shrink-0 text-[11px] text-gray-400 dark:text-gray-500">{{ group.title }}</span>
           </div>
 
           <div v-if="group.reason" class="mt-1 text-[11px] text-gray-500 dark:text-gray-400">{{ group.reason }}</div>