diff --git a/api/monitor/douyin_api.py b/api/monitor/douyin_api.py
index 3fc98a1..3047b3d 100644
--- a/api/monitor/douyin_api.py
+++ b/api/monitor/douyin_api.py
@@ -41,6 +41,7 @@ import os
 import time
 from dataclasses import dataclass
 from typing import Any, Dict, List, Optional, Tuple
+from urllib.parse import urlencode
 
 import config
 import httpx
@@ -281,14 +282,44 @@ def _cookie_value(cookie: str, name: str) -> str:
     return ""
 
 
+def _sign(params: Dict[str, Any], path: str, user_agent: str) -> Dict[str, Any]:
+    """给一组参数补上 ``a_bogus`` 签名，返回新 dict。
+
+    **按需 import**：那个模块在 import 的那一瞬间就把 ``libs/douyin.js`` 交给 execjs
+    编译（还要读相对路径），把它拖进监控层的热路径不合适。
+
+    签名算在**不含 a_bogus 的那串 query 上**，追加到末尾 —— 和爬虫那条路一致，也是
+    实测能过的形态。
+    """
+    from media_platform.douyin.help import get_a_bogus_from_js
+
+    try:
+        return {
+            **params,
+            "a_bogus": get_a_bogus_from_js(path, urlencode(params), user_agent),
+        }
+    except Exception as exc:  # execjs 起不来 / JS 抛错，都算签名失败
+        raise DouyinApiError(f"算 a_bogus 签名失败：{exc}") from exc
+
+
 async def _get(
-    path: str, params: Dict[str, Any], identity: BrowserIdentity
+    path: str,
+    params: Dict[str, Any],
+    identity: BrowserIdentity,
+    *,
+    signed: bool = False,
 ) -> Dict[str, Any]:
     """发一个 GET，返回 JSON。
 
     只带调用方给的参数 —— **不要往里加 webid / msToken / browser_version 那一堆**，
     那正是爬虫那条路失败的原因。
+
+    ``signed=True`` 时补一个 ``a_bogus``。**只有评论接口需要它**：作品、详情、博主资料
+    三个不带签名也照常返回，而给它们加签名是没验证过的改动，不做。
     """
+    if signed:
+        params = _sign(params, path, identity.user_agent)
+
     url = f"{API_ORIGIN}{path}"
     async with httpx.AsyncClient(timeout=REQUEST_TIMEOUT_SECONDS) as client:
         response = await client.get(
@@ -487,6 +518,10 @@ async def video_comments(
             "aid": 6383,
         },
         identity,
+        # 这个接口**必须**签名。不签的话网关回 200 + 空 body，会被读成「这条没评论」，
+        # 而它其实只是被挡了 —— 和登录失效长得一模一样。（实测：带上签名 200/9960 字节
+        # 真评论，不带就是空的。）
+        signed=True,
     )
 
     records = [
diff --git a/api/routers/monitor.py b/api/routers/monitor.py
index 574e983..7d87558 100644
--- a/api/routers/monitor.py
+++ b/api/routers/monitor.py
@@ -18,7 +18,7 @@
 
 """HTTP API for scheduled monitoring tasks."""
 
-from datetime import date, timedelta
+from datetime import date, datetime, timedelta
 from typing import Any, Dict, List, Optional
 
 from fastapi import APIRouter, HTTPException, Query, Response
@@ -486,20 +486,29 @@ def _export_columns(kind: str) -> List[tuple[str, str]]:
     """(key, header) pairs per export kind."""
     if kind == "notes":
         return [
-            ("note_id", "作品ID"),
+            # 先放「这是谁」：导出来是拿去比对和汇报的，一行只有作品 ID 没法用。
+            # 备注优先 —— 昵称常常认不出是谁（见 notes 表那一层的说明）。
+            ("creator_alias", "博主备注"),
+            ("creator_name", "博主昵称"),
+            ("note_alias", "作品备注"),
             ("title", "标题"),
+            ("note_id", "作品ID"),
             ("note_url", "链接"),
-            ("liked_count", "点赞"),
-            ("comment_count", "评论"),
-            ("collected_count", "收藏"),
-            ("share_count", "分享"),
-            ("liked_count_delta", "点赞增量"),
-            ("comment_count_delta", "评论增量"),
+            ("published_at", "发布时间"),
+            # 指标嵌在 row["metrics"] 里，所以这里必须写成路径 —— 写成裸键名的话这几列
+            # 全空（见 _lookup）。
+            ("metrics.liked_count", "点赞"),
+            ("metrics.comment_count", "评论"),
+            ("metrics.collected_count", "收藏"),
+            ("metrics.share_count", "分享"),
+            ("deltas.liked_count", "点赞增量"),
+            ("deltas.comment_count", "评论增量"),
             ("first_seen_at", "首次发现"),
             ("last_seen_at", "最近采集"),
         ]
     if kind == "comments":
         return [
+            ("note_creator_name", "博主昵称"),
             ("note_title", "所属作品"),
             ("note_id", "作品ID"),
             ("comment_id", "评论ID"),
@@ -521,6 +530,42 @@ def _export_columns(kind: str) -> List[tuple[str, str]]:
     ]
 
 
+# 表里存的是毫秒时间戳。直接倒进 CSV 就是一串 13 位数字 —— 打开 Excel 的人没法看，
+# 也没法排序。这几个键统一格式化成人能读的形态。
+_TIME_KEYS = {"published_at", "first_seen_at", "last_seen_at", "create_time"}
+
+
+def _fmt_time(value: Any) -> str:
+    """毫秒 → ``YYYY-MM-DD HH:MM``（服务器本地时区）。"""
+    try:
+        return datetime.fromtimestamp(int(value) / 1000).strftime("%Y-%m-%d %H:%M")
+    except (TypeError, ValueError, OSError, OverflowError):
+        return ""
+
+
+def _cell_for(row: Dict[str, Any], key: str) -> Any:
+    """一列的值：时间键格式化成人能读的，其余照原样（None 变空串）。"""
+    value = _lookup(row, key)
+    if key.rsplit(".", 1)[-1] in _TIME_KEYS:
+        return _fmt_time(value)
+    return _cell(value)
+
+
+def _lookup(row: Dict[str, Any], key: str) -> Any:
+    """取一列的值。键可以是 ``metrics.liked_count`` 这种路径。
+
+    作品行的指标是**嵌在** ``metrics`` / ``deltas`` 里的，而 ``_export_columns`` 里写的
+    是 ``liked_count`` —— 照顶层键直接 ``row.get()`` 的话，点赞/评论/收藏/分享四列连带
+    两个增量列**永远是空的**，导出来的表看着有这几列，其实一格都没有。
+    """
+    value: Any = row
+    for part in key.split("."):
+        if not isinstance(value, dict):
+            return None
+        value = value.get(part)
+    return value
+
+
 def _cell(value: Any) -> Any:
     if value is None:
         return ""
@@ -537,7 +582,7 @@ def _to_csv(rows: List[Dict[str, Any]], columns: List[tuple[str, str]]) -> bytes
     writer = csv.writer(buffer)
     writer.writerow([header for _, header in columns])
     for row in rows:
-        writer.writerow([_cell(row.get(key)) for key, _ in columns])
+        writer.writerow([_cell_for(row, key) for key, _ in columns])
 
     # utf-8-sig: without the BOM Excel opens Chinese CSV as mojibake, which is
     # the single most common complaint about CSV exports here.
@@ -554,7 +599,7 @@ def _to_xlsx(rows: List[Dict[str, Any]], columns: List[tuple[str, str]], sheet:
     worksheet.title = {"notes": "作品", "comments": "评论"}.get(sheet, "报表")
     worksheet.append([header for _, header in columns])
     for row in rows:
-        worksheet.append([_cell(row.get(key)) for key, _ in columns])
+        worksheet.append([_cell_for(row, key) for key, _ in columns])
 
     output = io.BytesIO()
     workbook.save(output)
diff --git a/tests/test_douyin_api.py b/tests/test_douyin_api.py
index aef3cc0..a67be0c 100644
--- a/tests/test_douyin_api.py
+++ b/tests/test_douyin_api.py
@@ -274,3 +274,128 @@ class TestGet:
         assert asyncio.run(douyin_api._get("/x", {}, self._identity())) == {
             "user": {"nickname": "x"}
         }
+
+
+@pytest.fixture
+def fake_signer(monkeypatch):
+    """把真正的 ``a_bogus`` 签名换成假的。
+
+    真的那个在 import 的瞬间就要把 ``libs/douyin.js`` 喂给 execjs（还得有 node 和正确的
+    相对路径），单元测试不该依赖这些。**签名本身是实测过的**：带上它是 200 + 真评论，
+    不带是 200 + 空 body。这里只负责钉住「有没有带上、传对了没有」。
+    """
+    import sys
+    import types
+
+    module = types.ModuleType("media_platform.douyin.help")
+    calls: list = []
+
+    def get_a_bogus_from_js(url: str, params: str, user_agent: str) -> str:
+        calls.append({"url": url, "params": params, "user_agent": user_agent})
+        return "FAKE-BOGUS"
+
+    module.get_a_bogus_from_js = get_a_bogus_from_js
+    monkeypatch.setitem(sys.modules, "media_platform.douyin.help", module)
+    return calls
+
+
+@pytest.fixture
+def recording_client(monkeypatch):
+    """记下实际发出去的那次请求。"""
+    sent: dict = {}
+
+    class _Response:
+        status_code = 200
+        text = '{"comments": []}'
+
+        def json(self):
+            return {"comments": []}
+
+    class _Client:
+        async def __aenter__(self):
+            return self
+
+        async def __aexit__(self, *exc):
+            return False
+
+        async def get(self, url, **kwargs):
+            sent["url"] = url
+            sent.update(kwargs)
+            return _Response()
+
+    monkeypatch.setattr(httpx, "AsyncClient", lambda **kwargs: _Client())
+    return sent
+
+
+class TestCommentSigning:
+    """评论接口**必须**带 a_bogus。
+
+    不带的话网关回 200 + 空 body —— 那在下游会变成「这条作品没有评论」，把一次被挡住的
+    请求伪装成一条正常的空结果。和登录失效长得一模一样，查起来能查半天。
+    """
+
+    def _identity(self):
+        return douyin_api.BrowserIdentity(
+            cookie="sessionid=s", user_agent="UA", client_hints={}
+        )
+
+    def test_the_signature_is_computed_over_the_unsigned_params(
+        self, fake_signer, recording_client
+    ):
+        asyncio.run(
+            douyin_api._get(
+                douyin_api.COMMENT_PATH,
+                {"aweme_id": "123", "count": 20},
+                self._identity(),
+                signed=True,
+            )
+        )
+
+        assert len(fake_signer) == 1
+        # 签名算在**不含 a_bogus** 的那串上 —— 把它自己也算进去是循环的。
+        assert fake_signer[0]["params"] == "aweme_id=123&count=20"
+        assert fake_signer[0]["url"] == douyin_api.COMMENT_PATH
+        assert fake_signer[0]["user_agent"] == "UA"
+
+    def test_the_signature_goes_out_with_the_request(self, fake_signer, recording_client):
+        asyncio.run(
+            douyin_api._get(
+                douyin_api.COMMENT_PATH, {"aweme_id": "123"}, self._identity(), signed=True
+            )
+        )
+
+        assert recording_client["params"]["a_bogus"] == "FAKE-BOGUS"
+        assert recording_client["params"]["aweme_id"] == "123"
+
+    def test_unsigned_calls_never_touch_the_signer(self, fake_signer, recording_client):
+        """作品 / 详情 / 博主资料三个接口不带签名也照常返回。
+
+        给它们加签名是**没验证过的改动** —— 所以这里钉住「不签」，防止有人图省事把
+        signed=True 改成全局默认。
+        """
+        asyncio.run(
+            douyin_api._get(douyin_api.POSTS_PATH, {"sec_user_id": "x"}, self._identity())
+        )
+
+        assert fake_signer == []
+        assert "a_bogus" not in recording_client["params"]
+
+    def test_a_broken_signer_is_reported_as_such(self, monkeypatch, recording_client):
+        """execjs 起不来时要说出是签名失败，而不是让它变成「没评论」。"""
+        import sys
+        import types
+
+        module = types.ModuleType("media_platform.douyin.help")
+
+        def boom(url, params, user_agent):
+            raise RuntimeError("node 没装")
+
+        module.get_a_bogus_from_js = boom
+        monkeypatch.setitem(sys.modules, "media_platform.douyin.help", module)
+
+        with pytest.raises(douyin_api.DouyinApiError, match="a_bogus"):
+            asyncio.run(
+                douyin_api._get(
+                    douyin_api.COMMENT_PATH, {"aweme_id": "1"}, self._identity(), signed=True
+                )
+            )
diff --git a/tests/test_monitor_comments.py b/tests/test_monitor_comments.py
index 6fc83ee..9a689e8 100644
--- a/tests/test_monitor_comments.py
+++ b/tests/test_monitor_comments.py
@@ -20,6 +20,7 @@
 
 import csv
 import io
+import re
 
 import httpx
 import pytest
@@ -264,7 +265,8 @@ class TestExport:
         workbook = load_workbook(io.BytesIO(response.content))
         sheet = workbook.active
         assert sheet.max_row == 5  # header + four comments
-        assert sheet.cell(row=1, column=1).value == "所属作品"
+        assert sheet.cell(row=1, column=1).value == "博主昵称"
+        assert sheet.cell(row=1, column=2).value == "所属作品"
 
     @pytest.mark.asyncio
     async def test_report_export(self, client):
@@ -292,3 +294,79 @@ class TestExport:
             "/api/monitor/export", params={"kind": "comments", "note_id": "no-such-note"}
         )
         assert response.status_code == 404
+
+
+class TestNotesExportColumns:
+    """作品导出的列 —— 「有列名」和「列里有数」是两回事。
+
+    原先这几列写的是裸键名 `liked_count`，而作品行的指标是嵌在 `metrics` 里的，
+    于是导出来的表有「点赞/评论/收藏/分享」四列，**每一格都是空的**，还没人发现 ——
+    因为原来的测试只断言了 `作品ID`。
+    """
+
+    @pytest.mark.asyncio
+    async def test_the_metric_columns_actually_contain_numbers(self, client):
+        from api.monitor.models import MonitorNoteMetric
+
+        async with monitor_db.get_session() as session:
+            from sqlalchemy import select
+
+            task_id = (await session.scalar(select(MonitorTask.id))).__int__()
+            session.add(
+                MonitorNoteMetric(
+                    task_id=task_id, note_id="note-a", run_id=1,
+                    captured_at=1_700_000_000_000,
+                    liked_count=123, comment_count=45,
+                    collected_count=6, share_count=7,
+                )
+            )
+
+        response = await client.get(
+            "/api/monitor/export", params={"kind": "notes", "format": "csv"}
+        )
+        rows = list(csv.DictReader(io.StringIO(response.content.decode("utf-8-sig"))))
+        by_id = {row["作品ID"]: row for row in rows}
+
+        assert by_id["note-a"]["点赞"] == "123"
+        assert by_id["note-a"]["评论"] == "45"
+        assert by_id["note-a"]["收藏"] == "6"
+        assert by_id["note-a"]["分享"] == "7"
+
+    @pytest.mark.asyncio
+    async def test_a_missing_metric_is_left_empty_not_zero(self, client):
+        """没采到的指标留空。写 0 的话，导出来的表会声称这条作品零互动。"""
+        response = await client.get(
+            "/api/monitor/export", params={"kind": "notes", "format": "csv"}
+        )
+        rows = list(csv.DictReader(io.StringIO(response.content.decode("utf-8-sig"))))
+
+        assert rows[0]["点赞"] == ""
+
+    @pytest.mark.asyncio
+    async def test_the_export_says_who_the_creator_is(self, client):
+        """一行只有作品 ID 没法用 —— 导出来是拿去比对和汇报的。
+
+        备注优先：昵称常常认不出是谁，而备注是人自己起的名字。
+        """
+        response = await client.get(
+            "/api/monitor/export", params={"kind": "notes", "format": "csv"}
+        )
+        rows = list(csv.DictReader(io.StringIO(response.content.decode("utf-8-sig"))))
+
+        assert {row["博主昵称"] for row in rows} == {"博主甲", "博主乙"}
+
+    @pytest.mark.asyncio
+    async def test_times_are_readable_not_raw_milliseconds(self, client):
+        """毫秒时间戳倒进 CSV 就是 13 位数字，打开 Excel 的人没法看、也没法排序。"""
+        response = await client.get(
+            "/api/monitor/export", params={"kind": "notes", "format": "csv"}
+        )
+        rows = list(csv.DictReader(io.StringIO(response.content.decode("utf-8-sig"))))
+        by_id = {row["作品ID"]: row for row in rows}
+
+        # 断言**形状**而不是具体时刻：格式化用的是服务器本地时区，写死一个字符串的话
+        # 换个时区的机器上就会红。
+        assert re.fullmatch(r"\d{4}-\d{2}-\d{2} \d{2}:\d{2}", by_id["note-a"]["发布时间"])
+        assert re.fullmatch(r"\d{4}-\d{2}-\d{2} \d{2}:\d{2}", by_id["note-a"]["首次发现"])
+        # 而且不能是原始毫秒。
+        assert by_id["note-a"]["发布时间"] != str(PUBLISHED_A)
diff --git a/webui/src/components/monitor/MonitorDashboard.tsx b/webui/src/components/monitor/MonitorDashboard.tsx
index 9f94cc4..10312b6 100644
--- a/webui/src/components/monitor/MonitorDashboard.tsx
+++ b/webui/src/components/monitor/MonitorDashboard.tsx
@@ -1,10 +1,11 @@
 import { useState } from 'react'
-import { Activity, BellRing, FileText, MessageSquare, Plus } from 'lucide-react'
+import { Activity, BellRing, Download, FileText, MessageSquare, Plus } from 'lucide-react'
 
 import { Badge } from '@/components/ui/badge'
 import { Button } from '@/components/ui/button'
 import { Tabs, TabsContent, TabsList, TabsTrigger } from '@/components/ui/tabs'
 import { useMonitorOverview, useMonitorTasks } from '@/hooks/useMonitor'
+import { monitorApi } from '@/lib/api'
 import type { MonitorTask } from '@/types/monitor'
 import { useCookieStatus, useWebhookStatus } from '@/hooks/useMonitor'
 import { useCurrentPlatform } from '@/hooks/usePlatform'
@@ -166,6 +167,23 @@ export function MonitorDashboard() {
                   >
                     只看新增
                   </Button>
+                  {/* 导出的是**这个任务**的全部作品，不是屏幕上这 200 条 —— 屏幕上那份是
+                      为了好看才截断的，导出跟着截断就成了「导出来的比看到的少」。
+                      走 window.open 而不是 blob：鉴权在 Cookie 上，浏览器自己会带上。 */}
+                  <Button
+                    variant="outline"
+                    size="sm"
+                    className="ml-auto"
+                    onClick={() =>
+                      window.open(
+                        monitorApi.getExportUrl({ kind: 'notes', taskId: selectedTask.id }),
+                        '_blank',
+                      )
+                    }
+                  >
+                    <Download className="w-3 h-3 mr-1" />
+                    导出 CSV
+                  </Button>
                 </div>
                 <NotesTable taskId={selectedTask.id} onlyNew={onlyNew} />
               </TabsContent>