Remove the retired artist overrides layer and the mixed artist type

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

README.md

@@ -24,8 +24,7 @@ Hot 100（2016–2025）為例的內容分析。
    ../data/source/yearend_hot100_2016_2025.csv
    ../data/derived
    --lyrics-dir ../data/captures/lyrics
-   --wikidata-csv ../data/captures/artists-wikidata.csv
+   --wikidata-csv ../data/captures/artists-wikidata.csv`）。
-   --overrides-csv ../data/manual/artists_overrides.csv`）。
    LLM 步驟使用 `claude-sonnet-4-6`、temperature=0、
    thinking 關閉；每步驟獨立執行兩次後由仲裁步驟合併
    （「2+1」協定）。

docs/decision_log.md

@@ -180,3 +180,18 @@
   推測不確定且無佐證者列疑慮清單；查證屬資料策展而非
   分析，以 Claude Code 輔助、不走分析 API；QS 批次以
   私人工作檔留存，不隨論文發布。
+
+## 2026-08-04
+
+- **移除人工 overrides 層**：`build-db` 刪去
+  `--overrides-csv` 選項與套用邏輯，文件同步移除
+  `data/manual/artists_overrides.csv`。理由：「先補完
+  Wikidata、再捕捉快照」工作流實跑後，469 位歌手全數
+  在上游查證補齊，快照即完整，本地覆蓋層已無存在事實；
+  `data/manual/` 層保留（供日後黃金標準編碼）。
+- **歌手型態刪去 mixed 值**：`ArtistType` 只留 solo／group。
+  mixed 是先導研究「男／女／混合團體」單一欄位的殘留，
+  正式設計拆成 gender＋type 後從未定義其指涉；署名一律
+  拆成個人後，男女混合是歌曲層（song_artists＋各歌手
+  gender）可推導的事實，不屬歌手實體。非人非團體者
+  （Pinkfong）type 留空由人工判定，維持現狀。

docs/project_structure.md

@@ -21,7 +21,7 @@ pop-fem-audit/
 │   │   └── lyrics/                        # 歌詞 .txt 快取
 │   │                                      #   （gitignored，版權）
 │   ├── manual/                # 人工著作：只由研究者手寫
-│   │   └── artists_overrides.csv          # 人工核定 / 深度背景
+│   │                          #   （黃金標準編碼等）
 │   └── derived/               # 衍生：只由 build-db 寫入
 │       ├── songs.csv          # 歌曲報表（人讀；進 git）
 │       └── artists.csv        # 歌手報表（人讀；進 git）

docs/research_plan.md

@@ -33,7 +33,7 @@
   ——一律以文字格式 commit。格式跟著層次走，不跟著偏好走。
 - **分層**：
   - 源頭：原始榜單 CSV（進 git）。
-  - 捕捉：Wikidata 快照 CSV、人工 overrides CSV、你的編碼
+  - 捕捉：Wikidata 快照 CSV、你的編碼
     CSV、`runs/` JSONL（皆進 git）；歌詞 `.txt` 快取
     （版權因素 gitignored，為已知的稽核缺口）。
   - 工作儲存：SQLite 單檔（`tools/instance/`，generated、
@@ -79,7 +79,7 @@
 | 階段 | 內容 | 方式 | 時程 |
 |---|---|---|---|
 | 0 | 基礎建設：git init、目錄結構、.gitignore、決策日誌、runner script（含 Batch API）、codebook v0 骨架 | script + 討論 | 7/30–7/31 |
-| 1 | 資料準備：`run_llm` 改走統一設定 → `build-db`（解析榜單成 songs/chart_entries/artists/song_artists）→ `import-lyrics`（pilot 2018–2025）→ `fetch-lyrics`（2016–17 與缺漏，Lyrics.ovh / LRCLIB）→ `fetch-artists`（Wikidata 快照 + 人工 overrides）→ `export-llm-input` | 子命令 | 7/31–8/3 |
+| 1 | 資料準備：`run_llm` 改走統一設定 → `build-db`（解析榜單成 songs/chart_entries/artists/song_artists）→ `import-lyrics`（pilot 2018–2025）→ `fetch-lyrics`（2016–17 與缺漏，Lyrics.ovh / LRCLIB）→ `fetch-artists`（Wikidata 快照）→ `export-llm-input` | 子命令 | 7/31–8/3 |
 | 2 | 候選篩選：全部唯一歌曲高召回 women-power 候選篩選（寧可多抓，人工剔除） | API 2+1 | 8/2–8/3 |
 | 3 | 黃金標準：依 codebook 人工逐首判定 genuine/peripheral/fake，附引用歌詞證據表（LLM 只做摘錄，不給判定建議）；先以 10–15 首校準樣本試編並修訂 codebook 後凍結；同批校準樣本實測 Sonnet 4.6 vs Opus 5 一致率 | 人工 + script 輔助 | 8/4–8/8 |
 | 4 | 受控比較（盲點實驗）：條件 A（詞彙層提示）vs 條件 B（框架感知提示），各 2+1，對照黃金標準計算假陽／假陰率 | API | 8/6–8/9 |

tools/src/pop_fem_audit_tools/build_db.py

@@ -8,8 +8,8 @@ Rebuilds the working store from scratch out of the committed
 inputs: the year-end chart CSV and the output directory for the
 review CSV files, given as the two positional command-line
 arguments, and the optional capture inputs, each given as an
-option: the lyrics cache directory, the Wikidata artist snapshot
+option: the lyrics cache directory and the Wikidata artist
-CSV, and the manual artist overrides CSV.  An omitted option
+snapshot CSV.  An omitted option
 leaves its capture layer unloaded; a given option whose path does
 not exist fails the build.  Missing tables
 are created on a fresh store; existing tables are never altered,
@@ -188,9 +188,6 @@ def parse_args(argv: list[str] | None) -> argparse.Namespace:
     parser.add_argument(
         "--wikidata-csv", type=Path, default=None,
         help="the Wikidata artist snapshot CSV file to apply")
-    parser.add_argument(
-        "--overrides-csv", type=Path, default=None,
-        help="the manual artist override CSV file to apply")
     return parser.parse_args(argv)
 
 
@@ -215,8 +212,7 @@ def parse_artist_credit(credit: str) -> list[tuple[str, Role]]:
     word or punctuation.
 
     Known limitation: a compound act name that contains one of the
-    delimiters, other than the protected names, is over-split;
+    delimiters, other than the protected names, is over-split.
-    such cases are corrected later via the human override layer.
 
     :param credit: The combined artist credit string.
     :return: The (name, role) pairs in credit order, primary side
@@ -407,8 +403,7 @@ def apply_artist_csv(session: Session, path: Path) -> None:
     """Apply an artist attribute CSV onto the artist rows.
 
     Artists match by exact name.  Only the non-empty cells are
-    applied, so a later CSV overrides an earlier one field by
+    applied, field by field.  The note column is ignored.
-    field.  The note column is ignored.
 
     :param session: The database session, with the artists
         flushed.
@@ -727,8 +722,6 @@ def main(argv: list[str] | None = None) -> int:
             load_lyrics(session, args.lyrics_dir)
         if args.wikidata_csv is not None:
             apply_artist_csv(session, args.wikidata_csv)
-        if args.overrides_csv is not None:
-            apply_artist_csv(session, args.overrides_csv)
         session.flush()
         violations: list[str] = find_violations(
             session, YEARS, RANKS_PER_YEAR)

tools/src/pop_fem_audit_tools/fetch_artists.py

@@ -106,9 +106,6 @@ class ArtistType(enum.StrEnum):
     """A solo artist: a human."""
     GROUP = "group"
     """A musical ensemble."""
-    MIXED = "mixed"
-    """A mixed act, assigned manually via the overrides; never
-    derived by the fetcher."""
 
 
 @dataclass

tools/src/pop_fem_audit_tools/models.py

@@ -78,7 +78,7 @@ class Artist(Base):
     gender: Mapped[str | None]
     """The gender of the artist."""
     type: Mapped[str | None]
-    """The artist type: solo, group, or mixed."""
+    """The artist type: solo or group."""
     genre: Mapped[str | None]
     """The music genre of the artist."""
     country: Mapped[str | None]

tools/tests/test_build_db.py

@@ -258,8 +258,6 @@ class TestBuildDB(unittest.TestCase):
         self.__lyrics: Path = self.__dir / "lyrics"
         self.__wikidata: Path = \
             self.__dir / "artists_wikidata.csv"
-        self.__overrides: Path = \
-            self.__dir / "artists_overrides.csv"
         self.__write_chart(self.CHART_CSV)
         url: str = f"sqlite:///{self.__dir}/store.sqlite3"
         config.set_settings(config.Settings(
@@ -573,42 +571,6 @@ class TestBuildDB(unittest.TestCase):
             list(session.scalars(sa.select(ChartEntry))), [])
         self.assertEqual(list(session.scalars(sa.select(Song))), [])
 
-    def test_overrides_apply_over_wikidata(self) -> None:
-        """Test that the overrides win over the Wikidata snapshot."""
-        self.__wikidata.write_text(
-            "name,qid,gender,type,genre,country,note\n"
-            "Adele,Q2831,female,solo,pop,GB,\n",
-            encoding="utf-8")
-        self.__overrides.write_text(
-            "name,qid,gender,type,genre,country,note\n"
-            "Adele,,,,soul,,manually checked\n",
-            encoding="utf-8")
-        self.assertEqual(self.__run_build(
-            "--wikidata-csv", str(self.__wikidata),
-            "--overrides-csv", str(self.__overrides))[0], 0)
-        session: Session = self.__session()
-        artist: Artist | None = session.scalar(
-            sa.select(Artist).where(Artist.name == "Adele"))
-        assert artist is not None
-        self.assertEqual(artist.genre, "soul")
-        self.assertEqual(artist.gender, "female")
-        self.assertEqual(artist.wikidata_qid, "Q2831")
-        self.assertEqual(artist.country, "GB")
-
-    def test_unknown_override_name_fails(self) -> None:
-        """Test that an unknown override name fails the build."""
-        self.__overrides.write_text(
-            "name,qid,gender,type,genre,country,note\n"
-            "Adel,,female,,,,typo\n", encoding="utf-8")
-        status: int
-        stderr: str
-        status, stderr = self.__run_build(
-            "--overrides-csv", str(self.__overrides))
-        self.assertNotEqual(status, 0)
-        self.assertIn("Adel", stderr)
-        session: Session = self.__session()
-        self.assertEqual(list(session.scalars(sa.select(Song))), [])
-
     def test_lyrics_loaded(self) -> None:
         """Test loading the lyrics cache into the songs."""
         self.__lyrics.mkdir()
@@ -676,18 +638,6 @@ class TestBuildDB(unittest.TestCase):
         session: Session = self.__session()
         self.assertEqual(list(session.scalars(sa.select(Song))), [])
 
-    def test_missing_overrides_csv_fails(self) -> None:
-        """Test that a given but missing override CSV fails."""
-        status: int
-        stderr: str
-        status, stderr = self.__run_build(
-            "--overrides-csv", str(self.__overrides))
-        self.assertNotEqual(status, 0)
-        self.assertIn("error:", stderr)
-        self.assertIn(str(self.__overrides), stderr)
-        session: Session = self.__session()
-        self.assertEqual(list(session.scalars(sa.select(Song))), [])
-
     REVIEW_CHART_CSV: str = (
         "year,rank,title,artist\n"
         "2016,1,banana,Artist B\n"