diff --git a/app/schemas/subscribe.py b/app/schemas/subscribe.py
index f0b511c8f..f1be43ca6 100644
--- a/app/schemas/subscribe.py
+++ b/app/schemas/subscribe.py
@@ -1,6 +1,7 @@
+import json
 from typing import Optional, List, Dict, Any, ClassVar
 
-from pydantic import BaseModel, Field, ConfigDict, model_validator
+from pydantic import BaseModel, Field, ConfigDict, model_validator, field_validator
 
 from app.schemas.media import OptionalMediaIdentityMixin
 from app.schemas.types import MediaSource, MediaType
@@ -154,6 +155,28 @@ class Subscribe(OptionalMediaIdentityMixin, BaseModel):
 
     model_config = ConfigDict(from_attributes=True)
 
+    @field_validator("note", mode="before")
+    @classmethod
+    def _normalize_legacy_note(cls, value: Any) -> Any:
+        """
+        兼容历史字符串型 note。
+
+        2.0 时代旧代码对 JSON 列显式做了 ``json.dumps``，历史数据可能是一层
+        或两层 JSON 编码的字符串（如 ``'[1, 2, 3]'``），不解析会触发响应
+        校验 500；解析失败按空值处理，避免脏数据阻塞整个订阅列表接口。
+        """
+        if not isinstance(value, str):
+            return value
+        parsed = value
+        while isinstance(parsed, str):
+            try:
+                parsed = json.loads(parsed)
+            except (TypeError, ValueError):
+                return None
+        if isinstance(parsed, list):
+            return [item for item in parsed if isinstance(item, int)]
+        return None
+
     @model_validator(mode="before")
     @classmethod
     def _normalize_empty_strings(cls, data: Any) -> Any:
diff --git a/database/versions/73370ce9bab7_3_0_7.py b/database/versions/73370ce9bab7_3_0_7.py
new file mode 100644
index 000000000..9004005fd
--- /dev/null
+++ b/database/versions/73370ce9bab7_3_0_7.py
@@ -0,0 +1,76 @@
+"""3.0.7
+修复历史订阅 note 字段的双重 JSON 编码
+
+Revision ID: 73370ce9bab7
+Revises: f4c8d2a7b1e6
+Create Date: 2026-08-17
+"""
+
+import json
+
+from alembic import op
+import sqlalchemy as sa
+
+
+revision = "73370ce9bab7"
+down_revision = "f4c8d2a7b1e6"
+branch_labels = None
+depends_on = None
+
+
+def _parse_legacy_note(value: object):
+    """解析历史写入的字符串型 note，兼容一层或两层 JSON 编码。
+
+    2.0 时代旧代码对 JSON 列显式做了 ``json.dumps(note)``，SQLite 下 2.0.3
+    的列类型迁移又是空操作，导致读回的值是 ``'[1, 2, 3]'`` 这类字符串而非列表；
+    字符串会被响应模型按 ``List[int]`` 校验并触发 500。解析失败按空值处理，
+    不丢弃整型数组以外的历史内容。
+    """
+    parsed = value
+    while isinstance(parsed, str):
+        try:
+            parsed = json.loads(parsed)
+        except (TypeError, ValueError):
+            return None
+    if isinstance(parsed, list) and all(isinstance(item, int) for item in parsed):
+        return parsed
+    return None
+
+
+def _repair_subscribe_note() -> None:
+    """把 subscribe 表中字符串型 note 回写为真正的 JSON 数组。
+
+    通过 sa.JSON 类型绑定参数写回，SQLite 与 PostgreSQL 都会按各自方言
+    序列化，保证后续按模型读回时得到整数列表。
+    """
+    subscribe = sa.table(
+        "subscribe",
+        sa.column("id", sa.Integer()),
+        sa.column("note", sa.JSON()),
+    )
+    connection = op.get_bind()
+    rows = connection.execute(
+        sa.select(subscribe.c.id, subscribe.c.note)
+    ).mappings().all()
+    for row in rows:
+        note = row["note"]
+        if not isinstance(note, str):
+            continue
+        repaired = _parse_legacy_note(note)
+        connection.execute(
+            subscribe.update()
+            .where(subscribe.c.id == row["id"])
+            .values(note=repaired)
+        )
+
+
+def upgrade() -> None:
+    """修复历史订阅 note 字段的双重 JSON 编码数据。"""
+    if "subscribe" not in sa.inspect(op.get_bind()).get_table_names():
+        return
+    _repair_subscribe_note()
+
+
+def downgrade() -> None:
+    """数据修复不可逆，降级不做任何操作。"""
+    pass
diff --git a/tests/test_subscribe_endpoint.py b/tests/test_subscribe_endpoint.py
index 53cff0141..93bac6072 100644
--- a/tests/test_subscribe_endpoint.py
+++ b/tests/test_subscribe_endpoint.py
@@ -1424,3 +1424,34 @@ def test_create_subscribe_accepts_music_payload_with_empty_strings():
     assert payload["type"] == MediaType.MUSIC.value
     assert payload["music_type"] == "album"
     assert payload["total_tracks"] == 13
+
+
+class _LegacyNoteRow:
+    """携带历史字符串 note 的最小 ORM 替身。"""
+
+    def __init__(self, note):
+        self.note = note
+
+
+@pytest.mark.parametrize(
+    ("raw", "expected"),
+    [
+        ("[1, 2, 3, 4]", [1, 2, 3, 4]),      # 双重 JSON 编码的整型数组
+        ("[1, 2]", [1, 2]),
+        ("[]", []),                          # 双重编码的空数组
+        ("null", None),                      # 双重编码的 null
+        ("not json", None),                  # 无法解析的历史脏数据
+        ([1, 2, 3], [1, 2, 3]),              # 正常列表原样保留
+        (None, None),                        # 空值
+    ],
+)
+def test_subscribe_note_normalizes_legacy_json_string(raw, expected):
+    """历史字符串型 note 应被解析为整数列表，避免响应校验 500。"""
+    subscribe = Subscribe.model_validate(_LegacyNoteRow(note=raw))
+    assert subscribe.note == expected
+
+
+def test_subscribe_note_strips_non_int_items_from_legacy_string():
+    """历史脏数据中混入非整数元素时只保留整数，不阻塞整个订阅列表接口。"""
+    subscribe = Subscribe.model_validate(_LegacyNoteRow(note='[1, 2, "x", 3]'))
+    assert subscribe.note == [1, 2, 3]