文笔层：AI 味机检、标点规范化、分块大量扩写（第 1 集成稿 10805 字）

- 修复上次提交把 dfannals/cli.py 写成 0 字节的问题（它是唯一入口，导致管道不可运行） - 性别：解析 <caste>，人物表与写作素材带性别（Ral Fastenhatchets 实为女性） - 新增 dfannals/deslop.py：AI 味机械诊断（硬伤词/句式/标点，按千字密度报告） - 新增 dfannals/normalize.py：标点与结构清理（引号配对、重复段落与句子、模型自加的小节标记） - 新增 dfannals/expand.py 与 prompts/literary-expander.md：按年份场景分块大量扩写 - 专名防幻觉：每块附史料专名白名单，事后按段自动修复可疑专名 - episode 命令并进文笔层（骨架稿另存 .skeleton.md），新增 expand 命令做 A/B 对照 - 新增 notes/switched-threads.md 与 test_deslop / test_normalize 回归测试 - 提交前拦截「已跟踪文件为空」，防止上述事故复发
2026-10-05 22:27:02 +08:00
parent 08a032120a
commit 8630dcad55
23 changed files with 1971 additions and 56 deletions
@@ -10,7 +10,7 @@
 <historical_figures>
  <historical_figure><id>10</id><name>Urist McMiner</name><race>DWARF</race><appeared>12</appeared><birth_year>12</birth_year><death_year>88</death_year><profession>MINER</profession></historical_figure>
  <historical_figure><id>11</id><name>Kogan Deathspear</name><race>GOBLIN</race><appeared>4</appeared><birth_year>4</birth_year><profession>AXEMAN</profession></historical_figure>
-  <historical_figure><id>12</id><name>liri fogbalded</name><race>ELF</race><appeared>20</appeared><birth_year>20</birth_year></historical_figure>
+  <historical_figure><id>12</id><name>liri fogbalded</name><race>ELF</race><caste>FEMALE</caste><appeared>20</appeared><birth_year>20</birth_year></historical_figure>
 </historical_figures>
 <entities>
  <entity><id>100</id><name>The Steel Confederacy</name><type>Civilization</type><race>DWARF</race></entity>
@@ -0,0 +1,90 @@
+"""AI 味诊断器的回归测试：能抓住植入的 AI 腔，也不冤枉正常句子。"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from dfannals import deslop
+
+# 故意堆满 AI 腔：禁用词、万能动宾、收束腔、破折号、省略号、认知直述
+AI_TEXT = (
+    "他心中一震，眼中闪过一丝惊讶，仿佛看见了命运的獠牙——这一刻，他终于明白，"
+    "一切都已经太晚了。他深吸一口气，嘴角勾起一抹冷笑，带着不容置疑的语气说："
+    "“走吧。”他知道，属于他的反击才刚刚开始……"
+)
+
+CLEAN_TEXT = (
+    "Ral把绳子绕了两圈，打了个结。她没说话，把账本推过去，指着第三行。"
+    "Guspu看了很久，把账本合上，还给她。窗外的风把灯吹得晃，两人谁都没去扶。"
+)
+
+
+def test_hard_hits_are_caught() -> None:
+    hits = deslop.hard_hits(deslop.check(AI_TEXT))
+    rules = {h.rule for h in hits}
+    for expected in ("心理模板", "眼中闪过", "一丝", "仿佛", "深吸一口气", "嘴角勾起",
+                     "不容置疑", "破折号", "省略号"):
+        assert expected in rules, f"漏检: {expected}（实际命中 {sorted(rules)}）"
+
+
+def test_cognitive_and_closure_patterns() -> None:
+    hits = deslop.hard_hits(deslop.check(AI_TEXT))
+    rules = {h.rule for h in hits}
+    assert "认知直述" in rules, "「他知道」应被抓住"
+    assert "收束腔" in rules, "「这一刻」「才刚刚开始」应被抓住"
+    assert "万能动宾" in rules, "「，带着……」应被抓住"
+    assert "抽象命运" in rules, "「命运的獠牙」应被抓住"
+
+
+def test_clean_text_has_no_hard_hits() -> None:
+    hits = deslop.hard_hits(deslop.check(CLEAN_TEXT))
+    assert not hits, f"正常句子被误判: {[h.line() for h in hits]}"
+
+
+def test_not_a_but_b_pattern() -> None:
+    hits = deslop.check("他不是冷漠，而是绝望。")
+    assert any(h.rule == "不是A而是B" for h in deslop.hard_hits(hits))
+
+
+def test_density_separates_ai_from_clean() -> None:
+    ai, clean = deslop.per_k(AI_TEXT), deslop.per_k(CLEAN_TEXT)
+    assert ai > clean, f"AI 稿密度应高于正常稿：{ai:.1f} vs {clean:.1f}"
+    assert clean == 0.0
+
+
+def test_report_and_digest_render() -> None:
+    rep = deslop.report(AI_TEXT)
+    assert "硬伤" in rep and "字" in rep
+    dig = deslop.digest(AI_TEXT)
+    assert dig.startswith("- "), dig[:80]
+    assert "机械检查没有发现" in deslop.digest(CLEAN_TEXT)
+
+
+def test_digest_lists_each_rule_once() -> None:
+    dig = deslop.digest(AI_TEXT)
+    rules = [line.split("『")[1].split("』")[0] for line in dig.splitlines()]
+    assert len(rules) == len(set(rules)), f"同一规则重复列出: {rules}"
+
+
+def _main() -> int:
+    tests = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    failed = 0
+    for fn in tests:
+        try:
+            fn()
+        except AssertionError as exc:
+            failed += 1
+            print(f"  ✗ {fn.__name__}: {exc}")
+        except Exception as exc:  # noqa: BLE001
+            failed += 1
+            print(f"  ✗ {fn.__name__}: {type(exc).__name__}: {exc}")
+        else:
+            print(f"  ✓ {fn.__name__}")
+    print(f"\n{len(tests) - failed}/{len(tests)} 通过")
+    return 1 if failed else 0
+
+
+if __name__ == "__main__":
+    sys.exit(_main())
@@ -75,6 +75,35 @@ def test_figure_ids_excludes_placeholder_and_dedups() -> None:
    assert len(battle.figure_ids()) == len(set(battle.figure_ids()))


+def test_caste_parsed_into_gender() -> None:
+    """回归：v53 导出用 <caste> 而非 <sex>，漏读会把女性主角写成"他"。"""
+    world = load_world()
+    assert world.figures[12].caste == "FEMALE"
+    assert world.figures[12].gender == "女"
+    assert world.figures[10].gender == "", "未载性别的应返回空串，不能猜"
+
+
+def test_cast_table_shows_gender() -> None:
+    """人物表必须带性别列，否则写作者只能猜。"""
+    from collections import Counter
+
+    from dfannals import cast
+    from dfannals import episodes as ep_mod
+    from dfannals.threads import Thread
+
+    world = load_world()
+    thread = Thread(
+        members=[10, 11, 12], interactions=5, turns=1, span=58, first_year=30, last_year=88,
+        types=Counter(), races=Counter(), non_person_races=[], events=world.events,
+    )
+    planned = ep_mod.plan_episodes(thread)
+    text, rows, suspicions = cast.build_cast(world, thread, planned, with_bios=False)
+    assert "| 人物 | 种族 | 性别 |" in text, text.splitlines()[4] if len(text.splitlines()) > 4 else text
+    assert not suspicions
+    elf = next(r for r in rows if r.fid == 12)
+    assert elf.gender == "女"
+
+
 def test_fake_names_are_caught() -> None:
    """反例：凭空编造的人名地名必须被抓出来。"""
    world = load_world()
@@ -0,0 +1,86 @@
+"""标点与结构清理的回归测试。"""
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from dfannals import normalize
+
+
+def test_ascii_quotes_become_chinese() -> None:
+    res = normalize.normalize('他说："好。" 她说："不。"')
+    assert res.text == "他说：“好。” 她说：“不。”", res.text
+    assert any("英文引号" in n for n in res.notes)
+
+
+def test_halfwidth_punct_between_cjk() -> None:
+    res = normalize.normalize("他说,好。她走,停。")
+    assert "，" in res.text and "," not in res.text, res.text
+
+
+def test_english_punctuation_untouched() -> None:
+    src = "Mon Sagus, The Plane of Dawn"
+    assert normalize.normalize(src).text == src
+
+
+def test_separator_lines_dropped() -> None:
+    res = normalize.normalize("甲段。\n\n---\n\n乙段。")
+    assert "---" not in res.text, res.text
+    assert any("分隔线" in n for n in res.notes)
+
+
+def test_stray_section_markings_dropped_but_title_kept() -> None:
+    res = normalize.normalize("# 守门人\n\n甲段。\n\n## 一\n\n乙段。\n\n七\n\n丙段。")
+    assert res.text.startswith("# 守门人"), res.text
+    assert "## 一" not in res.text, res.text
+    assert "\n七\n" not in res.text, res.text
+    assert any("小节标记" in n for n in res.notes)
+
+
+def test_duplicate_paragraph_merged() -> None:
+    res = normalize.normalize("同一段。\n同一段。\n不同段。")
+    assert res.text.count("同一段。") == 1, res.text
+    assert any("重复段落" in n for n in res.notes)
+
+
+def test_duplicate_sentences_merged() -> None:
+    res = normalize.normalize("他走了。他走了。她留下。")
+    assert res.text.count("他走了。") == 1, res.text
+    assert any("重复句子" in n for n in res.notes)
+
+
+def test_blank_lines_tidied() -> None:
+    res = normalize.normalize("甲。\n\n\n\n乙。")
+    assert "\n\n\n" not in res.text
+    assert res.text == "甲。\n\n乙。"
+
+
+def test_clean_text_untouched() -> None:
+    src = "Ral把绳子绕了两圈，打了个结。她没说话。"
+    res = normalize.normalize(src)
+    assert res.text == src
+    assert not res.changed, res.notes
+
+
+def _main() -> int:
+    tests = [v for k, v in sorted(globals().items()) if k.startswith("test_") and callable(v)]
+    failed = 0
+    for fn in tests:
+        try:
+            fn()
+        except AssertionError as exc:
+            failed += 1
+            print(f"  ✗ {fn.__name__}: {exc}")
+        except Exception as exc:  # noqa: BLE001
+            failed += 1
+            print(f"  ✗ {fn.__name__}: {type(exc).__name__}: {exc}")
+        else:
+            print(f"  ✓ {fn.__name__}")
+    print(f"\n{len(tests) - failed}/{len(tests)} 通过")
+    return 1 if failed else 0
+
+
+if __name__ == "__main__":
+    sys.exit(_main())