diff --git a/.claude/skills/parse-book/scripts/parse_ingest.py b/.claude/skills/parse-book/scripts/parse_ingest.py index 1c695da..0c66674 100644 --- a/.claude/skills/parse-book/scripts/parse_ingest.py +++ b/.claude/skills/parse-book/scripts/parse_ingest.py @@ -32,9 +32,14 @@ TENANT, ACTOR = 1, "1" PATTERN_TYPES = {"craft", "combat", "emotion", "scene_pattern", "trope"} # 拍板①:首轮只拆五型 NGRAM = 15 # 脱敏红线:≥15 连续字与原文重合=违规(parse-book skill) # 版权 IP 与系统专名词表(复抽实证:机战无限为高达系同人,IP 背景词不在实体名池)。 -# 注意"浮游炮"不入表——已泛化为中文机甲文通用武器品类词(同"光剑"),入表实测误伤 3 卡 +# 词形边界(opus 终检教训): +# - "浮游炮"不入表——已泛化为通用武器品类词(同"光剑"),入表实测误伤 3 卡; +# - "战功"/"负能"不入表——通用词(立下战功/负能量)子串误伤面大,其系统义 +# (机战货币"战功"、星环设定"负能")由批次清洗按语境甄别; +# - 原创书自造宇宙观术语(伪造物主/幽畸/锐眼)同属泄漏,一并列管。 IP_LEAK_WORDS = ("GN粒子", "太阳炉", "扎古", "高达", "夏亚", "阿姆罗", "米诺夫斯基", - "脑量子波", "GN-Bit", "影印人", "殖装", "战功点", "宇宙世纪") + "脑量子波", "GN-Bit", "影印人", "殖装", "战功点", "宇宙世纪", + "爆种", "次元兽", "伪造物主", "幽畸", "锐眼") EMBED_MODEL = "Qwen/Qwen3-Embedding-8B" SIM_MERGE = 0.85 # 嵌入判重:≥该值触发 M3 归并判定(阈值未校准,实战收集中) SIM_MARK = 0.75 # [MARK, MERGE) 区间只标记不判定(校准带宽,审核环可见)