添加测试集内部去重

@@ -80,6 +80,7 @@ python -m lyric_dedup.cli generate-eval-set \
   --lyrics-dir data/generated_eval/incoming \
   --csv data/generated_eval/eval_50000.csv \
   --index outputs/indexes/lyrics.pkl \
+  --eval-index data/generated_eval/eval_50000.csv.index.pkl \
   --size 50000 \
   --positive-ratio 0.3
 ```
@@ -88,10 +89,10 @@ python -m lyric_dedup.cli generate-eval-set \
 
 - 先扫描整个曲库，按有效歌词行数、语言类型、文件来源前缀做分层采样，不再按排序前缀取样。
 - `应去重` 样本只生成全曲歌词的样式变化，例如时间戳、标点、平台噪声、空行、重复副歌次数变化、附加中文翻译。
-- `不应去重` 样本包含同主题新歌词、hard negative、片段歌词、重复副歌碰撞、仅翻译相似、短歌词/占位边界样本。
+- `不应去重` 样本以真实 holdout 完整歌词为主，也包含片段歌词、重复副歌碰撞、仅翻译相似、同主题新歌词、短歌词/占位边界样本。
 - 片段歌词即使命中已有歌曲的一部分，也不应该输出 `duplicate`；最多进入 `review`。
-- 如果传入 `--index`，生成器会用现有索引构造更接近线上召回风险的 hard negative。
-- 同时会生成 `*.manifest.json`，记录 seed、曲库规模、样本类型分布、语言/来源分桶和样本来源覆盖数。
+- 生成器会额外写出 `--eval-index`，这个索引排除了 holdout 歌，评估生成 CSV 时应使用它。
+- 同时会生成 `*.manifest.json`，记录 seed、曲库规模、holdout 数、样本类型分布、语言/来源分桶和样本来源覆盖数。
 
 先准备一个 CSV，例如 `data/eval/eval.csv`：
 

@@ -103,6 +103,7 @@ python -m lyric_dedup.cli generate-eval-set \
   --lyrics-dir data/generated_eval/incoming \
   --csv data/generated_eval/eval_50000.csv \
   --index outputs/indexes/library_lyrics.pkl \
+  --eval-index data/generated_eval/eval_50000.csv.index.pkl \
   --size 50000 \
   --positive-ratio 0.3
 ```
@@ -120,24 +121,26 @@ python -m lyric_dedup.cli generate-eval-set \
 
 ```text
 positive_* = 应去重，全曲歌词样式变化
-negative_random_unrelated = 不应去重，同主题新歌词
-negative_hard_candidate = 不应去重，系统容易召回的短句/局部重合样本
+negative_real_holdout_full_song = 不应去重，完整真实歌词，已从评估索引中排除
 negative_fragment = 不应去重，单曲片段
 negative_shared_chorus = 不应去重，重复副歌碰撞
 negative_translation_only = 不应去重，仅翻译相似
+negative_same_theme_synthetic = 不应去重，同主题新歌词
 edge_short_or_placeholder = 不应去重，短歌词/占位边界样本
 ```
 
-生成器会扫描整个曲库并按有效歌词行数、语言类型、文件来源前缀分层采样。传入 `--index` 后会用现有索引生成 hard negative。每次还会输出：
+生成器会扫描整个曲库并按有效歌词行数、语言类型、文件来源前缀分层采样。它会分出一批 holdout 完整歌词作为真实新歌负样本，并生成一个排除 holdout 的评估索引。每次还会输出：
 
 ```text
 data/generated_eval/eval_50000.csv.manifest.json
+data/generated_eval/eval_50000.csv.index.pkl
 ```
 
 manifest 里重点看：
 
 ```text
 library_files          曲库歌词文件数
+holdout_records        从评估索引中排除、作为真实新歌负样本的数量
 sample_type_counts     各样本类型数量
 line_count_bucket_counts / language_bucket_counts / source_bucket_counts
 unique_source_records  本次评估覆盖了多少真实源文件
@@ -147,7 +150,7 @@ unique_source_records  本次评估覆盖了多少真实源文件
 
 ```bash
 python -m lyric_dedup.cli evaluate-csv \
-  --index outputs/indexes/library_lyrics.pkl \
+  --index data/generated_eval/eval_50000.csv.index.pkl \
   --csv data/generated_eval/eval_50000.csv \
   --base-dir data/generated_eval \
   --out outputs/results/library_eval_50000.csv
@@ -171,7 +174,7 @@ false_positive
 
 ```bash
 python -m lyric_dedup.cli evaluate-csv \
-  --index outputs/indexes/library_lyrics.pkl \
+  --index data/generated_eval/eval_50000.csv.index.pkl \
   --csv data/generated_eval/eval_50000.csv \
   --base-dir data/generated_eval \
   --positive-decisions duplicate,review \

@@ -96,16 +96,24 @@ class DuplicateChecker:
 
     def add_record(self, record: LyricRecord) -> None:
         indexed = self._index(record)
-        self._records[record.record_id] = indexed
-        self._exact_hash_to_ids.setdefault(indexed.exact_hash, set()).add(record.record_id)
+        self._add_indexed(record.record_id, indexed)
+
+    def add_normalized_record(self, record: LyricRecord, normalized: NormalizedLyrics) -> None:
+        """Add a record when normalized lyrics have already been computed."""
+        indexed = self._index_normalized(record, normalized)
+        self._add_indexed(record.record_id, indexed)
+
+    def _add_indexed(self, record_id: str, indexed: _IndexedRecord) -> None:
+        self._records[record_id] = indexed
+        self._exact_hash_to_ids.setdefault(indexed.exact_hash, set()).add(record_id)
         for line in indexed.normalized.unique_lines:
             if len(line) >= 4:
-                self._line_to_ids.setdefault(line, set()).add(record.record_id)
+                self._line_to_ids.setdefault(line, set()).add(record_id)
         for token in indexed.tokens:
-            self._token_to_ids.setdefault(token, set()).add(record.record_id)
+            self._token_to_ids.setdefault(token, set()).add(record_id)
         for token in indexed.fallback_tokens:
-            self._token_to_ids.setdefault(token, set()).add(record.record_id)
-        self._lsh.add(record.record_id, indexed.signature)
+            self._token_to_ids.setdefault(token, set()).add(record_id)
+        self._lsh.add(record_id, indexed.signature)
 
     def save(self, path: str | Path) -> None:
         """Persist the in-memory index for later checks."""
@@ -187,6 +195,9 @@ class DuplicateChecker:
 
     def _index(self, record: LyricRecord) -> _IndexedRecord:
         normalized = normalize_lyrics(record.lyrics)
+        return self._index_normalized(record, normalized)
+
+    def _index_normalized(self, record: LyricRecord, normalized: NormalizedLyrics) -> _IndexedRecord:
         tokens = lyric_tokens(normalized)
         primary_tokens = lyric_tokens(normalized, lines=normalized.primary_lines)
         translation_tokens = lyric_tokens(normalized, lines=normalized.translation_lines)

@@ -5,6 +5,7 @@ from __future__ import annotations
 import argparse
 import csv
 import json
+import sys
 from pathlib import Path
 
 from lyric_dedup.checker import DuplicateChecker
@@ -50,7 +51,8 @@ def main() -> None:
     generate.add_argument("--size", type=int, default=100)
     generate.add_argument("--positive-ratio", type=float, default=0.3)
     generate.add_argument("--seed", type=int, default=20260602)
-    generate.add_argument("--index", default="", help="optional existing index for hard-negative generation")
+    generate.add_argument("--index", default="", help="optional source index path recorded in the manifest")
+    generate.add_argument("--eval-index", default="", help="output index built from non-holdout records for this eval set")
 
     args = parser.parse_args()
     if args.command == "build-index":
@@ -77,6 +79,7 @@ def main() -> None:
             positive_ratio=args.positive_ratio,
             seed=args.seed,
             index_path=Path(args.index) if args.index else None,
+            eval_index_path=Path(args.eval_index) if args.eval_index else None,
         )
         print(json.dumps(summary, ensure_ascii=False))
 
@@ -155,52 +158,58 @@ def evaluate_csv(
     positive_decisions: set[str],
     max_candidates: int,
 ) -> None:
+    _progress(f"load index: {index_path}")
     checker = DuplicateChecker.load(index_path)
     rows: list[dict[str, object]] = []
+    total = _csv_data_row_count(csv_path)
+    _progress(f"evaluate csv: 0/{total}")
+    out_path.parent.mkdir(parents=True, exist_ok=True)
     with csv_path.open(encoding="utf-8-sig", newline="") as file:
         reader = csv.DictReader(file)
         if reader.fieldnames is None:
             raise ValueError("评估 CSV 需要表头")
-        for row_number, row in enumerate(reader, start=2):
-            sample_id = row.get("id") or row.get("sample_id") or str(row_number)
-            record, source = _record_from_eval_row(row, csv_path=csv_path, base_dir=base_dir)
-            expected_duplicate = _parse_expected(row.get("expected") or row.get("label") or row.get("target"))
-            result = checker.check_record(record, max_candidates=max_candidates)
-            predicted_duplicate = result.decision.value in positive_decisions
-            best = result.candidates[0] if result.candidates else None
-            rows.append(
-                {
-                    "id": sample_id,
-                    "source": source,
-                    "expected_duplicate": expected_duplicate,
-                    "decision": result.decision.value,
-                    "predicted_duplicate": predicted_duplicate,
-                    "correct": expected_duplicate == predicted_duplicate,
-                    "confidence": result.confidence,
-                    "reason": result.reason,
-                    "best_candidate_id": best.record_id if best else "",
-                    "best_candidate_decision": best.decision.value if best else "",
-                    "best_candidate_confidence": best.confidence if best else "",
-                    "best_candidate_jaccard": best.jaccard if best else "",
-                    "best_candidate_line_coverage": best.line_coverage if best else "",
-                    "best_candidate_primary_jaccard": best.primary_jaccard if best else "",
-                    "best_candidate_primary_line_coverage": best.primary_line_coverage if best else "",
-                    "best_candidate_translation_jaccard": best.translation_jaccard if best else "",
-                    "best_candidate_translation_line_coverage": best.translation_line_coverage if best else "",
-                    "best_candidate_reason": best.reason if best else "",
-                    "matched_unique_lines": " | ".join(best.matched_unique_lines) if best else "",
-                }
-            )
-
-    out_path.parent.mkdir(parents=True, exist_ok=True)
-    with out_path.open("w", encoding="utf-8", newline="") as file:
-        writer = csv.DictWriter(file, fieldnames=list(rows[0].keys()) if rows else ["id"])
+        fieldnames = [
+            "id",
+            "source",
+            "expected_duplicate",
+            "decision",
+            "predicted_duplicate",
+            "correct",
+            "confidence",
+            "reason",
+            "best_candidate_id",
+            "best_candidate_decision",
+            "best_candidate_confidence",
+            "best_candidate_jaccard",
+            "best_candidate_line_coverage",
+            "best_candidate_primary_jaccard",
+            "best_candidate_primary_line_coverage",
+            "best_candidate_translation_jaccard",
+            "best_candidate_translation_line_coverage",
+            "best_candidate_reason",
+            "matched_unique_lines",
+        ]
+        with out_path.open("w", encoding="utf-8", newline="") as out_file:
+            writer = csv.DictWriter(out_file, fieldnames=fieldnames)
             writer.writeheader()
-        writer.writerows(rows)
+            for index, row in enumerate(reader, start=1):
+                row_out = _evaluate_row(
+                    row,
+                    row_number=index + 1,
+                    checker=checker,
+                    csv_path=csv_path,
+                    base_dir=base_dir,
+                    positive_decisions=positive_decisions,
+                    max_candidates=max_candidates,
+                )
+                rows.append(row_out)
+                writer.writerow(row_out)
+                _progress_count("evaluate csv", index, total, step=1000)
 
     summary = _evaluation_summary(rows, positive_decisions=positive_decisions, out_path=out_path)
     summary_path = out_path.with_suffix(out_path.suffix + ".summary.json")
     summary_path.write_text(json.dumps(summary, ensure_ascii=False, indent=2), encoding="utf-8")
+    _progress("evaluation complete")
     print(json.dumps(summary, ensure_ascii=False))
 
 
@@ -229,6 +238,45 @@ def _result_to_dict(result, *, source: str) -> dict[str, object]:
     }
 
 
+def _evaluate_row(
+    row: dict[str, str],
+    *,
+    row_number: int,
+    checker: DuplicateChecker,
+    csv_path: Path,
+    base_dir: Path | None,
+    positive_decisions: set[str],
+    max_candidates: int,
+) -> dict[str, object]:
+    sample_id = row.get("id") or row.get("sample_id") or str(row_number)
+    record, source = _record_from_eval_row(row, csv_path=csv_path, base_dir=base_dir)
+    expected_duplicate = _parse_expected(row.get("expected") or row.get("label") or row.get("target"))
+    result = checker.check_record(record, max_candidates=max_candidates)
+    predicted_duplicate = result.decision.value in positive_decisions
+    best = result.candidates[0] if result.candidates else None
+    return {
+        "id": sample_id,
+        "source": source,
+        "expected_duplicate": expected_duplicate,
+        "decision": result.decision.value,
+        "predicted_duplicate": predicted_duplicate,
+        "correct": expected_duplicate == predicted_duplicate,
+        "confidence": result.confidence,
+        "reason": result.reason,
+        "best_candidate_id": best.record_id if best else "",
+        "best_candidate_decision": best.decision.value if best else "",
+        "best_candidate_confidence": best.confidence if best else "",
+        "best_candidate_jaccard": best.jaccard if best else "",
+        "best_candidate_line_coverage": best.line_coverage if best else "",
+        "best_candidate_primary_jaccard": best.primary_jaccard if best else "",
+        "best_candidate_primary_line_coverage": best.primary_line_coverage if best else "",
+        "best_candidate_translation_jaccard": best.translation_jaccard if best else "",
+        "best_candidate_translation_line_coverage": best.translation_line_coverage if best else "",
+        "best_candidate_reason": best.reason if best else "",
+        "matched_unique_lines": " | ".join(best.matched_unique_lines) if best else "",
+    }
+
+
 def _lyrics_from_eval_row(row: dict[str, str], *, csv_path: Path, base_dir: Path | None) -> tuple[str, str]:
     lyrics = (row.get("lyrics") or "").strip()
     if lyrics:
@@ -322,5 +370,23 @@ def _evaluation_summary(
     }
 
 
+def _csv_data_row_count(csv_path: Path) -> int:
+    with csv_path.open(encoding="utf-8-sig", newline="") as file:
+        reader = csv.reader(file)
+        next(reader, None)
+        return sum(1 for _ in reader)
+
+
+def _progress(message: str) -> None:
+    print(f"[eval] {message}", file=sys.stderr, flush=True)
+
+
+def _progress_count(label: str, current: int, total: int, *, step: int = 1000) -> None:
+    if total <= 0:
+        return
+    if current == 1 or current == total or current % step == 0:
+        _progress(f"{label}: {current}/{total}")
+
+
 if __name__ == "__main__":
     main()

@@ -308,9 +308,11 @@ def test_generated_eval_set_uses_stratified_production_mix(tmp_path) -> None:
     assert manifest["library_files"] == 12
     assert manifest["sample_size"] == 30
     assert manifest["unique_source_records"] > 1
+    assert manifest["holdout_records"] > 1
+    assert (tmp_path / "generated" / "eval.csv.index.pkl").exists()
     assert "positive_full_duplicate" in manifest["plan"]
+    assert "negative_real_holdout_full_song" in negative_types
     assert "negative_fragment" in negative_types
-    assert "negative_hard_candidate" in negative_types
     assert all(row["expected"] == "不应去重" for row in rows if row["sample_type"].startswith("negative_"))