dsh-novel-writer 2.0.2 → 2.5.0

package/README.en.md

@@ -1,104 +1,26 @@
-# 📚 dsh-novel-writer — Novel Writing Assistant
+# dsh-novel-writer — Novel Writing Assistant Plugin (v2.5.0)
 
-English | [**中文**](./README.md)
+A DSH (DeepSeek Harness) bundle plugin. **Ships a local semantic embedding engine (bge-small-zh-v1.5 ONNX, bundled, CPU inference, 0 token cost)** — semantic capabilities without any API fees.
 
-[![npm version](https://img.shields.io/npm/v/dsh-novel-writer.svg?style=flat-square&color=blue)](https://www.npmjs.com/package/dsh-novel-writer)
-[![License: MIT](https://img.shields.io/badge/license-MIT-green.svg?style=flat-square)](LICENSE)
-[![GitHub stars](https://img.shields.io/github/stars/siweina/dsh-novel-writer.svg?style=flat-square&color=orange)](https://github.com/siweina/dsh-novel-writer/stargazers)
-[![GitHub release](https://img.shields.io/github/v/release/siweina/dsh-novel-writer.svg?style=flat-square)](https://github.com/siweina/dsh-novel-writer/releases)
-[![DSH plugin](https://img.shields.io/badge/DSH-plugin-4b8bbe.svg?style=flat-square)](https://github.com/deepseek-ai/deepseek-harness)
+## Tools (15, each with an independent UI switch)
+novel_books / novel_chapters / novel_read / novel_keywords / novel_new_chapter / novel_import / novel_sentence_analysis / novel_sentence_config / novel_style_check / novel_plot / novel_settings / novel_summary / novel_continuity_check / novel_semantic_search (new in v2.0.0) / **novel_style_report (new in v2.5.0)**
 
-A novel-writing assistant plugin for **DeepSeek Harness (DSH)**: chapter library management, sentence-pattern analysis, emotion purification & quantification, style checking, plot tracking, settings management, worldview/pragmatics detection, **local semantic search (0 token)**, batch import, and AI-assisted continuation writing. Zero third-party dependencies on host.
+## v2.0.0 — Semantic Layer
+- **Local engine** (lib/embedding.js): Xenova/bge-small-zh-v1.5 (quantized), 512-dim Chinese vectors, ~23MB, bundled; @huggingface/tokenizers + onnxruntime-web (WASM), local CPU, 0 token / 0 API cost;
+- **Lazy load** (~0.3s first call); automatic fallback to pure-rule mode if unavailable;
+- **Index cache**: per-book vectors at `.novel-writer/embedding/<book>.json` — repeated searches are instant;
+- **novel_semantic_search**: natural-language search across the whole book (foreshadowing clues, emotional scenes, setting mentions) — matches by meaning even when keywords differ;
+- **novel_style_check upgrade**: adds semantic similarity (target chapter vs rest of book) alongside rule-based fingerprint similarity;
+- **Switch: semanticEmbedding** — default ON (probe-based: auto-enabled when the model is present); turn it off in the sidebar「写作助手功能」panel to fully return to rule-only mode.
 
----
+## History (v0.3 → v1.6)
+Sentence-pattern analysis; emotion purification + AI review (strong/weak lexicon, pollution caveat); emotion quantification (Valence sliding window: variance/slope/conflict + implicit imagery + complexity score); worldview/genre/theme detection (modern/western/eastern + 15 genres + 35 themes); pragmatic-style review; settings tables (5) + foreshadowing registry + summaries + continuity audit + style check; analysis cache & report export (`.novel-writer/analysis/`).
 
-## Installation
+## v2.5.0: Style Portrait Report (measurement/judgment separation)
+- **novel_style_report** (15th tool): aggregates 6 measurement dimensions (style fingerprint / keywords / theme / emotion quantification / 12-axis vibe / semantic style distances);
+- Plugin only reports numbers, never labels — style judgment is left to the LLM; AI judgment can be saved via `aiJudgment` (stored in .novel-writer/style-reports/) and read with `action=get`;
+- Vibe spectrum: 12 axes (added 文艺唯美 Aesthetic / 情欲暧昧 Sensual), rule-based conclusions removed;
+- Semantic style distances: 12 style prototype sets × book vector (local embedding, 0 token).
 
-**Option 1: npm (recommended)**
-
-```sh
-dsh plugin --profile web add dsh-novel-writer
-# or
-npm install dsh-novel-writer
-```
-
-**Option 2: From GitHub**
-
-```sh
-dsh plugin --profile web add github:siweina/dsh-novel-writer#main
-```
-
-After installing, **restart the web app** to activate (host registers 14 tools + state/reveal routes; browser mounts the "Writing Assistant" sidebar toggle panel).
-
----
-
-## v2.0.0 New: Local Semantic Engine (0 token)
-
-- 🧠 **Built-in Chinese semantic model** (bge-small-zh-v1.5 quantized, 23MB, shipped with the plugin): 512-dim vectors, local CPU inference, **zero API cost**;
-- 🔍 **novel_semantic_search**: search the whole book with natural language for semantically related passages (foreshadowing/emotional scenes/settings), even without matching keywords;
-- 📊 **Semantic style comparison**: novel_style_check adds a semantic-similarity dimension beyond rule-based fingerprints;
-- 💭 **Semantic implicit emotion**: emotion prototype sentences scan the book index to find "imagery passages outside the word list";
-- 🪶 Lazy loading + automatic fallback to pure-rule mode (never breaks existing features).
-
----
-
-## Features
-
-1. **Chapter library**: chapters in `novels/<book>/第N章.md` (or .txt/.markdown); auto encoding detection (UTF-8/UTF-16/GBK).
-2. **Sentence-pattern analysis**: 9 categories, arrangement patterns, rhythm, emotion curve, style fingerprint + guidance, with cache & report export.
-3. **Emotion purification**: strong/weak emotion-word grading, pollution-source detection, caveat warning + AI re-verification, cleanDominant true baseline.
-4. **Emotion quantification** (v1.6.0): Valence mapping + sliding window → variance V / delta Δ / conflict index C; implicit imagery carriers + explicit-implicit conflict; complexity score + composite emotion pairs.
-5. **Style check**: chapter vs book → similarity + deviation list + advice.
-6. **Plot tracking**: foreshadowing/plot-hook registry (open/done), typed fields + auto mention tracking.
-7. **Settings management**: five tables — characters / locations / items / timeline / worldview.
-8. **Worldview & pragmatics detection**: auto cultural-baseline detection with confidence; speechStyle title/honorifics/rituals/tone norms.
-9. **Genre & theme detection**: primary/secondary themes + genre, low-frequency noise filtered.
-10. **Chapter summaries**: per-chapter digest + key events for long-book continuation.
-11. **Continuity audit**: settings conflicts + pragmatics conflicts with replacement suggestions.
-12. **Batch import**: auto-detect book names & chapter numbers, classified import.
-13. **Continuation writing**: read-first workflow; keeps style/foreshadowing/worldview consistent.
-14. **Per-tool UI toggles**: "Writing Assistant" sidebar panel (master + per-tool + feature toggles).
-
----
-
-## Provided Tools (14)
-
-| Tool | Description |
-|------|-------------|
-| `novel_books` | List all books in library |
-| `novel_chapters` | List a book's chapters |
-| `novel_read` | Read a chapter (paginated) |
-| `novel_keywords` | Keywords: bigram/trigram/name candidates |
-| `novel_new_chapter` | Create new chapter file |
-| `novel_import` | Batch import manuscripts |
-| `novel_sentence_analysis` | Sentence-pattern analysis |
-| `novel_sentence_config` | View/set tool & feature toggles |
-| `novel_style_check` | Style check (rule + semantic) |
-| `novel_plot` | Plot/foreshadowing tracker |
-| `novel_settings` | Settings management (+worldview) |
-| `novel_summary` | Chapter summaries |
-| `novel_continuity_check` | Continuity audit |
-| `novel_semantic_search` | **Semantic search** (v2.0.0, local embedding, 0 token) |
-
----
-
-## Configuration
-
-```yaml
-- id: novel-writer
-  config:
-    root: 'D:/my-novel-library'
-    allowLanState: false   # true = allow state save from LAN GUI access
-```
-
----
-
-## Data Directory
-
-Under `<library-root>/.novel-writer/`: `plots` / `settings` / `summaries` / `analysis` / `audits` / `embedding` (semantic index cache).
-
----
-
-## License
-
-[MIT](./LICENSE)
+## Size
+~95MB total (23MB quantized model + 70MB runtime); zip 31MB. If disk is tight, disable the semantic switch and delete `lib/models/` + `node_modules/` to go back to rule-only.

package/README.md

@@ -1,104 +1,73 @@
-# 📚 dsh-novel-writer — 小说写作助手
+# dsh-novel-writer — 小说写作助手插件（v2.0.0）
 
-[**English**](./README.en.md) | 中文
+DSH（DeepSeek Harness）小说写作助手 bundle 插件。**v2.0.0 集成本地语义嵌入引擎（bge-small-zh-v1.5 ONNX，随插件分发，本地 CPU 推理，0 token 成本）**，在不增加任何 API 费用的前提下获得语义级能力。
 
-[![npm version](https://img.shields.io/npm/v/dsh-novel-writer.svg?style=flat-square&color=blue)](https://www.npmjs.com/package/dsh-novel-writer)
-[![License: MIT](https://img.shields.io/badge/license-MIT-green.svg?style=flat-square)](LICENSE)
-[![GitHub stars](https://img.shields.io/github/stars/siweina/dsh-novel-writer.svg?style=flat-square&color=orange)](https://github.com/siweina/dsh-novel-writer/stargazers)
-[![GitHub release](https://img.shields.io/github/v/release/siweina/dsh-novel-writer.svg?style=flat-square)](https://github.com/siweina/dsh-novel-writer/releases)
-[![DSH plugin](https://img.shields.io/badge/DSH-plugin-4b8bbe.svg?style=flat-square)](https://github.com/deepseek-ai/deepseek-harness)
+## 工具清单（14 个，全部带独立 UI 开关）
 
-为 **DeepSeek Harness (DSH)** 打造的小说写作助手插件：章节库管理、句式模式分析、情感净化与量化、风格自检、伏笔登记、设定管理、世界观/语用检测、**本地语义检索（0 token）**、批量导入与 AI 续写辅助。宿主端零第三方依赖，浏览器端仅依赖 Web GUI 自带的 react。
+novel_books / novel_chapters / novel_read / novel_keywords / novel_new_chapter / novel_import / novel_sentence_analysis / novel_sentence_config / novel_style_check / novel_plot / novel_settings（设定管理）/ novel_summary（章节摘要）/ novel_continuity_check（连贯性审计）/ **novel_semantic_search（语义检索，v2.0.0 新增）**
 
----
+## v2.0.0 新增能力（语义层）
 
-## 安装
+### 本地语义嵌入引擎（embedding.js）
+- 模型：Xenova/bge-small-zh-v1.5（512 维中文向量，约 91MB），随插件分发，开箱即用；
+- 运行时：**@huggingface/tokenizers**（官方轻量分词，294KB 零依赖）+ onnxruntime-web（WASM 推理），本地 CPU，0 token、0 API 费用；
+- 懒加载：首次调用才加载（约 0.3s）；加载失败自动回退纯规则，不影响任何既有功能；
+- 索引缓存：每书向量落盘 `.novel-writer/embedding/<书>.json`，重复检索秒开（47 万字约 3100 段，首次建索引约 20-30s，之后命中缓存）。
 
-**方式一：npm（推荐）**
+### 新工具：novel_semantic_search（语义检索）
+用自然语言描述在全书中检索语义相关段落——找伏笔线索、情感场景、设定提及，**即使原文没有相同关键词也能命中**（如搜「压抑克制的时刻」能命中没有"压抑"二字的段落）。
 
-```sh
-dsh plugin --profile web add dsh-novel-writer
-# 或
-npm install dsh-novel-writer
-```
-
-**方式二：从 GitHub 安装**
-
-```sh
-dsh plugin --profile web add github:siweina/dsh-novel-writer#main
-```
-
-安装后**重启 web 应用**生效（宿主端注册 14 个工具与 state/reveal 路由，浏览器端挂载侧边栏「写作助手功能」开关面板）。
-
----
-
-## v2.0.0 新增：本地语义引擎（0 token）
-
-- 🧠 **内置中文语义模型**（bge-small-zh-v1.5 quantized，23MB，随插件分发）：512 维向量，本地 CPU 推理，**零 API 费用**；
-- 🔍 **novel_semantic_search**：用自然语言搜全书语义相关段落（伏笔/情感场景/设定提及），即使原文没有相同关键词也能命中；
-- 📊 **语义级风格对比**：novel_style_check 在规则指纹之外新增语义相似度维度；
-- 💭 **语义隐性情感**：情感原型句扫全书索引，发现"词表外疑似意象段落"；
-- 🪶 懒加载 + 失败自动回退纯规则（不影响任何既有功能）。
-
----
-
-## 功能
+### novel_style_check 升级：语义级风格对比
+规则指纹相似度（句式/句长/情绪）之外，新增**语义相似度**（目标章 vs 全书其余部分向量余弦），双维度判定风格一致性。
 
-1. **章节库管理**：章节存放于 `novels/<书名>/第N章.md`（或 .txt/.markdown），编码自动探测（UTF-8/UTF-16/GBK）。
-2. **句式模式分析**：九类句式分布、排列规律、句长节奏、情感曲线、风格指纹与节奏建议，带缓存与报告导出。
-3. **情感净化**：强/弱情绪词分级、污染源检测、caveat 预警 + AI 复核闭环、cleanDominant 真实基调。
-4. **情感量化**（v1.6.0）：Valence 效价映射 + 滑动窗口 → 方差 V / 斜率 Δ / 矛盾指数 C；隐性意象载体 + 显隐冲突；复杂度评分 + 复合情感共现。
-5. **风格自检**：章节 vs 全书 → 相似度 + 偏差清单 + 续写建议。
-6. **伏笔登记表**：open/done 状态、字段化登记、章节提及自动追踪。
-7. **设定管理**：人物/地点/道具/时间线/世界观 五张表。
-8. **世界观与语用检测**：自动判断文化基准（西/东/混合）+ 置信度；speechStyle 称谓/客套/仪式/语气规范。
-9. **题材与流派检测**：主副题材 + 流派识别，低频噪音自动过滤。
-10. **章节摘要**：每章摘要 + 关键事件，长书续写先读摘要。
-11. **连贯性审计**：设定矛盾 + 语用冲突候选，带建议替换。
-12. **批量导入**：原稿件自动识别书名与章节号分类导入。
-13. **续写辅助**：先读后写、保持文风/伏笔/世界观一致。
-14. **全工具 UI 开关**：侧边栏「写作助手功能」面板（总开关 + 每工具开关 + 功能开关）。
+### 功能开关：语义增强（semanticEmbedding）
+- **默认开启（探测式）**：模型存在即自动启用；用户可在侧边栏「写作助手功能」面板关闭，关闭后语义检索返回提示、风格检查跳过语义维度；
+- 关闭后插件完整回退到 v1.6.0 纯规则能力。
 
----
+## 历史能力（v0.3 → v1.6 累积）
+- 句式模式分析（九类句式分布/排列/段落/句长/情感曲线/风格指纹/密度）；
+- 情感净化 + AI 复核（强/弱情绪词分级、污染预警 caveat、强制抽查）；
+- 情感量化（Valence 滑动窗口：方差/斜率/矛盾指数 + 隐性意象 + 复杂度评分）；
+- 世界观/流派/题材检测（modern/western/eastern + 15 流派 + 35 题材，主副题材）；
+- 语用级审查（称谓/客套/仪式/语气，中西/现代禁用词）；
+- 设定管理五张表 + 伏笔登记 + 章节摘要 + 连贯性审计 + 风格自检；
+- 分析缓存与报告导出（`.novel-writer/analysis/`）。
 
-## 提供的工具（14 个）
+## 功能开关（侧边栏「写作助手功能」面板）
+- 总开关 enabled；autoAnalyze；14 个工具级开关；
+- 功能级：emotionCaveat（情感净化预警）/ genreTheme（题材与流派检测）/ emotionComplexity（情感量化）/ **semanticEmbedding（语义增强，v2.0.0）**。
 
-| 工具 | 说明 |
-|------|------|
-| `novel_books` | 列出章节库全部作品 |
-| `novel_chapters` | 列出某作品章节清单 |
-| `novel_read` | 阅读某章正文（分段） |
-| `novel_keywords` | 关键词：二字组/三字组/疑似人名 |
-| `novel_new_chapter` | 创建新章节文件 |
-| `novel_import` | 原稿件批量导入/分类 |
-| `novel_sentence_analysis` | 句式模式分析（九类/情感净化/量化/曲线/指纹） |
-| `novel_sentence_config` | 查看/修改工具与功能开关 |
-| `novel_style_check` | 风格自检（规则+语义双维度） |
-| `novel_plot` | 伏笔/剧情线登记表 |
-| `novel_settings` | 设定管理（人物/地点/道具/时间线/世界观） |
-| `novel_summary` | 章节摘要（长书续写辅助） |
-| `novel_continuity_check` | 连贯性审计（设定矛盾+语用冲突候选） |
-| `novel_semantic_search` | **语义检索**（v2.0.0，本地 embedding，0 token） |
-
----
-
-## 配置
-
-```yaml
-- id: novel-writer
-  config:
-    root: 'D:/我的小说库'
-    allowLanState: false   # true=局域网访问 GUI 时也允许保存开关
+## 目录结构
+```
+dsh-novel-writer-v2.0.0/
+├── lib/
+│   ├── index.js        # 宿主端（14 工具 + state 路由 + 开关门禁）
+│   ├── analysis.js     # 规则引擎（句式/情感净化/情感量化/题材）
+│   ├── embedding.js    # 本地语义嵌入引擎（bge-small-zh ONNX）
+│   ├── models/         # 模型文件（onnx/model.onnx 91MB + tokenizer）
+│   └── client.js       # 浏览器端（侧边栏面板 + 全部开关）
+├── node_modules/       # @huggingface/tokenizers + onnxruntime-web（随插件分发）
+├── test/               # 引擎 + client + e2e 测试
+└── skills/novel-writing/SKILL.md  # 模型使用指南
 ```
 
----
-
-## 数据目录
-
-`<书库根>/.novel-writer/`：`plots`（伏笔）/ `settings`（设定）/ `summaries`（摘要）/ `analysis`（分析报告）/ `audits`（审计）/ `embedding`（语义索引缓存）。
-
----
-
-## 许可证
-
-[MIT](./LICENSE)
+## v2.1.0 新增：氛围光谱 + 多方向意象表
+- **氛围光谱（vibe）**：detect 输出 10 轴氛围坐标（噩梦感/焦虑压抑/温馨治愈/甜宠日常/催泪虐心/黑暗残酷/悬疑神秘/热血激昂/荒诞无厘头/孤独疏离）+ 组合结论 + 置信度 + 证据链——设定型气质（类克苏鲁等）以数值表达，不贴死标签；
+- **多方向意象表**：变色龙词（雨/烛火/火光/夜/灯…）按语境触发词裁决方向，触发词未命中时用本地语义引擎与正/负原型句比余弦，仍不确定则双计分并标记歧义——修复"甜宠文误判负向 100%"类问题；
+- **词表**：内置大连理工七类情感词表（27,413 词）供意象/情感扩展。
+
+## v2.2.0 新增：网文信号层（vibe 三路补强）
+- **动作/套路词群**：164 词（含 webnovel-writer genre-tropes 现成套路词：退婚/打脸/秒杀/追妻火葬场…），按 10 轴分组，封顶权重（每轴 ≤0.3）；
+- **题材联动**：题材命中 → 相关轴加分（豪门总裁→甜宠、系统流→荒诞/热血、玄幻→热血…），仅当该轴已有基础信号时触发；
+- **情感直给词密度**：joy/fear/sorrow/anger 词计数 → 对应轴密度加分；
+- **防污染四道防线**：加信号不换引擎 / 低权重封顶 / 来源标注可审计 / features.webnovelVibe 开关可回退；三组回归（公版文学 8 本 + 手写样本 6 本 + 网文 6 本）验证无污染；
+- **词表资源**：lib/lexicons/webnovel-tropes.md（现成套路模板）+ dutir_seven.json（大连理工 27,413 情感词）。
+
+## v2.5.0 新增：风格画像报告（测量与判断分离）
+- **novel_style_report（第 15 个工具）**：聚合 6 维测量数据（文风指纹/高频词汇/题材流派/情感量化/氛围光谱 12 轴/语义风格距离）输出结构化报告；
+- **插件只报数不贴标签**：氛围光谱 12 轴（新增 文艺唯美/情欲暧昧），砍掉规则结论文案——判断交给大模型；
+- **AI 判断回传存盘**：AI 结合报告给出风格气质判断 → aiJudgment 参数回传 → 存入 .novel-writer/style-reports/<书>.json，后续 action=get 读取，续写保持风格一致；
+- **语义风格距离**：12 组风格原型句 × 全书向量（本地 embedding，0 token）——实测示例书：克苏鲁诡异 0.79/压抑致郁 0.79。
+
+## 体积说明
+**zip 约 31MB**（含本地语义模型与 WASM 运行时，解压后代码+模型约 130MB）；已移除 transformers.js/onnxruntime-node/sharp 全家桶（省 85%）。如磁盘紧张可在「语义增强」开关关闭后删除 `lib/models/` 与 `node_modules/`，插件回到纯规则模式。

package/lib/analysis.js

@@ -347,6 +347,28 @@ function emotionOf(text) {
       cleanDominant = emotion;
     }
   }
+  // v2.5 修复：dutir_seven.json（大连理工 27,413 词）兜底——小词表未覆盖的情感词也计分
+  // 性能安全：只用"文本中出现的相邻二字组"查表（O(句长)），不做 27k 词全表扫描
+  try {
+    const seenWords = new Set();
+    for (const list of Object.values(words)) for (const w of list) seenWords.add(w);
+    dutirEmotion(""); // 确保 dutirLookup 已构建（懒加载）
+    const bigrams = text.match(/[\u4e00-\u9fa5]{2}/g) || [];
+    for (const w of new Set(bigrams)) {
+      if (seenWords.has(w)) continue;
+      const emo = dutirLookup.get(w);
+      if (!emo) continue;
+      scores[emo] += 1;
+      cleanScores[emo] += 1;
+      words[emo].push(w);
+      seenWords.add(w);
+    }
+    dominant = "neutral";
+    let best = 0;
+    for (const emotion of ["joy", "anger", "sorrow", "fear", "surprise"]) {
+      if (scores[emotion] > best) { best = scores[emotion]; dominant = emotion; }
+    }
+  } catch { /* dutir 兜底失败不影响原结果 */ }
   return {
     scores,
     words,
@@ -474,18 +496,143 @@ const IMPLICIT_CARRIERS = Object.freeze([
   { words: ["垂手", "放下手", "手垂落"], valence: -0.15, label: "无力·放弃" }
 ]);
 
+/**
+ * v2.1.0 多方向意象表：同一载体在不同语境可表达不同情感（"变色龙词"）。
+ * 每个词允许多个方向条目，各自带触发语境词（规则层裁决）：
+ *  - 命中触发词 → 按该方向计分
+ *  - 无触发词 → 交给语义层（与正/负原型句比相似度）
+ *  - 语义层也不确定 → 双计分 + 歧义标记
+ */
+const AMBIGUOUS_CARRIERS = Object.freeze({
+  "雨": [
+    { valence: -0.3, label: "压抑·萧瑟", triggers: ["冷", "夜", "秋", "寒", "阴", "灰", "敲", "不停", "绵", "细", "孤"] },
+    { valence: 0.2, label: "清新·复苏", triggers: ["晴", "彩虹", "洗净", "春", "润", "后", "停", "甘"] }
+  ],
+  "烛火": [
+    { valence: -0.2, label: "诡异·不安", triggers: ["摇曳", "昏", "暗", "影", "鬼", "摇", "颤", "燃尽", "跳"] },
+    { valence: 0.25, label: "温馨", triggers: ["暖", "炉", "家", "围", "饭", "柔", "亮"] }
+  ],
+  "火光": [
+    { valence: -0.25, label: "灾难·恐惧", triggers: ["烧", "浓烟", "废墟", "惨叫", "逃", "夜", "红"] },
+    { valence: 0.2, label: "希望·温暖", triggers: ["暖", "黎明", "亮", "驱散", "炉", "温"] }
+  ],
+  "夜": [
+    { valence: -0.2, label: "孤独·恐惧", triggers: ["深", "黑", "静", "冷", "无眠", "怕", "漫长", "沉"] },
+    { valence: 0.15, label: "安宁·静谧", triggers: ["星", "月", "静好", "温柔", "安"] }
+  ],
+  "风": [
+    { valence: -0.15, label: "萧瑟·离别", triggers: ["冷", "寒", "秋", "吹散", "凛冽", "呜咽"] },
+    { valence: 0.15, label: "清爽·自由", triggers: ["暖", "春", "清新", "拂", "轻", "晴"] }
+  ],
+  "海": [
+    { valence: -0.3, label: "深邃·恐惧", triggers: ["黑", "沉", "浪", "吞", "潮", "深", "暗", "涌"] },
+    { valence: 0.2, label: "开阔·浪漫", triggers: ["蓝", "晴", "暖", "浪花", "笑", "沙滩", "夕阳"] }
+  ],
+  "灯": [
+    { valence: -0.15, label: "孤独·守望", triggers: ["孤", "昏", "暗", "残", "灭", "夜"] },
+    { valence: 0.2, label: "温暖·归处", triggers: ["暖", "亮", "家", "等", "柔", "光"] }
+  ],
+  "影子": [
+    { valence: -0.25, label: "不安·诡异", triggers: ["长", "暗", "摇晃", "鬼", "拖", "黑"] },
+    { valence: 0.1, label: "陪伴", triggers: ["暖", "短", "依偎"] }
+  ],
+  "笑": [
+    { valence: -0.15, label: "苦笑·强颜", triggers: ["苦", "勉", "僵", "假", "惨", "涩", "硬"] },
+    { valence: 0.25, label: "欢乐", triggers: ["开怀", "灿烂", "暖", "甜", "大", "爽朗", "咯咯"] }
+  ],
+  "眼泪": [
+    { valence: -0.3, label: "悲伤", triggers: ["落", "流", "止不住", "擦", "咸", "含", "忍"] },
+    { valence: 0.15, label: "感动·释然", triggers: ["感动", "幸福", "喜极", "温暖", "笑"] }
+  ],
+  "沉默": [
+    { valence: -0.2, label: "压抑·隔阂", triggers: ["久", "冷", "尴尬", "低头", "死寂", "不开口"] },
+    { valence: 0.1, label: "默契·安宁", triggers: ["温柔", "懂", "默契", "安静", "并肩"] }
+  ],
+  "花开": [
+    { valence: -0.1, label: "易逝·伤春", triggers: ["落", "谢", "春尽", "残"] },
+    { valence: 0.25, label: "美好·希望", triggers: ["盛", "香", "春", "灿烂", "暖"] }
+  ],
+  "黄昏": [
+    { valence: -0.25, label: "迟暮·萧瑟", triggers: ["残", "落", "暗", "孤", "冷", "尽"] },
+    { valence: 0.15, label: "温柔·归家", triggers: ["暖", "金", "柔", "炊烟", "并肩"] }
+  ],
+  "奔跑": [
+    { valence: -0.2, label: "逃离·慌乱", triggers: ["逃", "拼命", "慌", "追", "喘", "夜"] },
+    { valence: 0.2, label: "自由·奔赴", triggers: ["向", "奔", "扑", "迎", "笑", "阳光"] }
+  ]
+});
+
+/** v2.1.0：查找多方向载体命中（返回所有方向条目 + 上下文命中触发词）。 */
+export function matchAmbiguousCarriers(text) {
+  const hits = [];
+  for (const [word, entries] of Object.entries(AMBIGUOUS_CARRIERS)) {
+    let idx = 0;
+    while ((idx = text.indexOf(word, idx)) !== -1) {
+      // 上下文窗口：前后 30 字
+      const ctx = text.slice(Math.max(0, idx - 30), idx + word.length + 30);
+      const matched = entries.map((e) => {
+        const trigger = e.triggers.find((t) => ctx.includes(t));
+        return { ...e, triggerHit: trigger || null };
+      });
+      hits.push({ word, ctx, entries: matched });
+      idx += word.length;
+    }
+  }
+  return hits;
+}
+
 /** v1.6.0 隐性载体扫描：意象/动作 → 负向/正向占比 + top 载体 + 脆弱标记。 */
-export function implicitEmotionScan(text) {
+export function implicitEmotionScan(text, semResolver = null) {
   let negHits = 0, posHits = 0, fragileHits = 0;
+  let ambHits = 0;
   const carrierCounts = new Map();
+  const ambiguous = [];
+  const see = (label, word, n, valence) => {
+    if (valence < 0) negHits += n * -valence;
+    else { posHits += n * valence; if (false) fragileHits += n; }
+    carrierCounts.set(label + ":" + word, (carrierCounts.get(label + ":" + word) ?? 0) + n);
+  };
+  // ① 单方向表（原有）
   for (const carrier of IMPLICIT_CARRIERS) {
     for (const word of carrier.words) {
       const n = text.split(word).length - 1;
       if (n > 0) {
-        if (carrier.valence < 0) negHits += n * -carrier.valence;
-        else { posHits += n * carrier.valence; if (carrier.fragile) fragileHits += n; }
-        carrierCounts.set(carrier.label + ":" + word, (carrierCounts.get(carrier.label + ":" + word) ?? 0) + n);
+        see(carrier.label, word, n, carrier.valence);
+        if (carrier.valence >= 0 && carrier.fragile) fragileHits += n;
+      }
+    }
+  }
+  // ② 多方向表（变色龙词）：规则触发 → 定方向；无触发 → 语义层；都不确定 → 双计分+歧义
+  for (const [word, entries] of Object.entries(AMBIGUOUS_CARRIERS)) {
+    let idx = 0;
+    while ((idx = text.indexOf(word, idx)) !== -1) {
+      const ctx = text.slice(Math.max(0, idx - 30), idx + word.length + 30);
+      const triggerMatched = entries.filter((e) => e.triggers.some((t) => ctx.includes(t)));
+      if (triggerMatched.length === 1) {
+        const e = triggerMatched[0];
+        see(e.label, word, 1, e.valence);
+      } else if (triggerMatched.length > 1) {
+        // 多个触发同时命中：双计分 + 歧义
+        for (const e of triggerMatched) see(e.label, word, 1, e.valence * 0.5);
+        ambHits += 1;
+        ambiguous.push({ word, reason: "多触发冲突", ctx: ctx.trim().slice(0, 40) });
+      } else if (semResolver) {
+        // 无触发 → 语义层裁决（异步由调用方包装）
+        const r = semResolver(ctx, entries);
+        if (r && r.resolved) see(r.label, word, 1, r.valence);
+        else {
+          // 语义不确定 → 双计分 + 歧义标记
+          for (const e of entries) see(e.label, word, 1, e.valence * 0.5);
+          ambHits += 1;
+          ambiguous.push({ word, reason: "语义不确定", ctx: ctx.trim().slice(0, 40) });
+        }
+      } else {
+        // 无语义层可用 → 按词表首条目弱计分 + 歧义
+        for (const e of entries.slice(0, 2)) see(e.label, word, 1, e.valence * 0.5);
+        ambHits += 1;
+        ambiguous.push({ word, reason: "未裁决", ctx: ctx.trim().slice(0, 40) });
       }
+      idx += word.length;
     }
   }
   const total = negHits + posHits;
@@ -493,8 +640,10 @@ export function implicitEmotionScan(text) {
   return {
     negative: total === 0 ? 0 : Math.round((negHits / total) * 100) / 100,
     positive: total === 0 ? 0 : Math.round((posHits / total) * 100) / 100,
+    ambiguousRatio: total === 0 ? 0 : Math.round((ambHits / (total + ambHits)) * 100) / 100,
     fragile: fragileHits > 0,
     totalHits: negHits + posHits,
+    ambiguous,
     topCarriers: top
   };
 }
@@ -675,9 +824,9 @@ export function emotionComplexity(perChapter) {
 /**
  * v1.6.0 情感量化入口：三指标 + 显隐对比 + 复杂度 + 复合共现（纯规则 0 token）。
  */
-export function emotionalQuantification(text, perChapter, blocks) {
+export function emotionalQuantification(text, perChapter, blocks, semResolver = null) {
   const stats = valenceStats(text);
-  const implicit = implicitEmotionScan(text);
+  const implicit = implicitEmotionScan(text, semResolver);
   const meanValence = stats.meanValence ?? 0;
   // perChapter 为空（单章/无分章输入）时，用当前文本自身 clean 计数聚合，保证 complexity 恒有值
   if (!Array.isArray(perChapter) || perChapter.length === 0) {
@@ -942,7 +1091,7 @@ export function analyzeText(text, options = {}) {
       .slice(0, 12),
     curve: emotionCurve(blockMeta, curveSegments),
     // v1.6.0：情感量化（Valence 三指标 + 显隐对比 + 复杂度 + 复合共现）
-    quantification: emotionalQuantification(text, [], blockMeta.map((m) => m.sentences).filter((s) => s.length > 0))
+    quantification: emotionalQuantification(text, [], blockMeta.map((m) => m.sentences).filter((s) => s.length > 0), options.semResolver || null)
   };
 
   // 主观性指数（启发式 0-100）
@@ -1150,3 +1299,34 @@ export function densityOf(text) {
     sense: { ...sense }
   };
 }
+
+
+// v2.5 修复：接入 lib/lexicons/dutir_seven.json（大连理工七类情感词表 27,413 词）
+// 懒加载 + 内存缓存；七类 → 五情感映射，供 sentimentCounts 扩展计数（词表未覆盖的词也能计分）
+import { createRequire } from "node:module";
+const __req = createRequire(import.meta.url);
+let DUTIR = null;
+function loadDutir() {
+  if (DUTIR) return DUTIR;
+  try {
+    const p = __req.resolve("./lexicons/dutir_seven.json");
+    DUTIR = JSON.parse(__req("node:fs").readFileSync(p, "utf8"));
+  } catch {
+    DUTIR = {};
+  }
+  return DUTIR;
+}
+const DUTIR_TO_EMOTION = { 乐: "joy", 好: "joy", 怒: "anger", 哀: "sorrow", 惧: "fear", 恶: "sorrow", 惊: "surprise" };
+const dutirLookup = new Map();
+function dutirEmotion(word) {
+  if (dutirLookup.size === 0) {
+    const dutir = loadDutir();
+    for (const [cat, words] of Object.entries(dutir)) {
+      const emo = DUTIR_TO_EMOTION[cat];
+      if (!emo) continue;
+      for (const w of words) if (typeof w === "string" && w.length >= 2) dutirLookup.set(w, emo);
+    }
+  }
+  return dutirLookup.get(word);
+}
+export function dutirEmotionOf(word) { return dutirEmotion(word); }