dsh-library 0.2.2 → 0.2.4

package/CHANGELOG.md

@@ -5,6 +5,17 @@ All notable changes to this project are documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.2.4] - 2026-09-04
+- Align devDeps pins to the published dsh 0.1.2-rc.1 line and move the compat CI probes from 0.1.1-rc.2 to 0.1.2-rc.1; no behavior change.
+
+## [0.2.3] - 2026-09-02
+- Align devDeps pins to the published dsh 0.1.2-alpha.5 line; no behavior change.
+
+### Fixed
+
+- The `library_search` relevance gate now filters on the hybrid score — the same value the diversity re-rank orders by — instead of re-scoring each chunk lexically, so semantic matches pass the `search.minRelevance` threshold ([#2](https://github.com/PerryLink/dsh-library/issues/2)).
+- CJK queries now score through unigram + adjacent-bigram tokens instead of whole-run tokens, so partial Chinese phrase overlaps pass the relevance gate ([#2](https://github.com/PerryLink/dsh-library/issues/2)).
+
 ## [0.2.2] - 2026-09-01
 
 ### Changed

package/README.es.md

@@ -25,7 +25,7 @@
 
 | Superficie | Estado |
 |---|---|
-| Harness | DeepSeek Harness `0.1.2-alpha.3` (compatibilidad declarada para `0.1.2-alpha.3`) |
+| Harness | DeepSeek Harness `0.1.2-rc.1` (compatibilidad declarada para `0.1.2-rc.1`) |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | Almacenamiento | Cualquier backend de storage-domain (JSON o SQLite); el índice vive en el dominio de almacenamiento del host |
 | Modelos | Ninguno requerido — el embedder integrado es hash determinista (cero descargas) |
@@ -120,8 +120,8 @@ Todos los ajustes son campos `Config` de Schemastery (modificables desde cordis.
 
 - **Permisos**: el plugin solo lee los archivos a los que apunta `library_add` (a través del servicio de archivos del harness y su política) y escribe en su propio dominio de almacenamiento `dsh_library`. Sin peticiones de red; un embedder externo opcional se ejecuta por `ctx.subprocess` sin interpretación de shell.
 - **Datos**: el texto de los chunks y los embeddings viven en el backend de almacenamiento del host (la misma confianza que el resto de los datos durables del despliegue); el plugin no añade cifrado. Las rutas de documentos y los embeddings nunca entran en el registro de sesión.
-- **Registro de sesión**: `library/inject` (id, consulta, ids de chunks, tamaño de página) y `library/purge` (veredicto) son eventos de auditoría solo-registro — la página inyectada visible para el modelo es reconstruible a partir de ellos. El append está condicionado por el host: los harnesses cuyo conjunto de tipos conocidos cubre el vocabulario reciben los eventos, las builds con envoltura `ignorable` los reciben con el marcador, y las builds sin envoltura (0.1.1-rc.2, 0.1.2-alpha.3) omiten el append — allí los eventos registrados `tool/call` + `tool/result` siguen siendo el rastro de auditoría reconstruible.
-0.1.2-alpha.3 (adaptado el 2026-09-01): el sobre de sesión conserva su campo ignorable solo para compatibilidad de lectura de logs almacenados - Session.append aún no puede estamparlo, por lo que el comportamiento de la puerta no cambia.
+- **Registro de sesión**: `library/inject` (id, consulta, ids de chunks, tamaño de página) y `library/purge` (veredicto) son eventos de auditoría solo-registro — la página inyectada visible para el modelo es reconstruible a partir de ellos. El append está condicionado por el host: los harnesses cuyo conjunto de tipos conocidos cubre el vocabulario reciben los eventos, las builds con envoltura `ignorable` los reciben con el marcador, y las builds sin envoltura (0.1.1-rc.2, 0.1.2-rc.1) omiten el append — allí los eventos registrados `tool/call` + `tool/result` siguen siendo el rastro de auditoría reconstruible.
+0.1.2-rc.1 (adaptado el 2026-09-02): el sobre de sesión conserva su campo ignorable solo para compatibilidad de lectura de logs almacenados - Session.append aún no puede estamparlo, por lo que el comportamiento de la puerta no cambia.
 
 ## Límites de seguridad
 
@@ -142,7 +142,7 @@ Todos los ajustes son campos `Config` de Schemastery (modificables desde cordis.
 ```sh
 pnpm install        # node ^22.19 || >=24
 pnpm run typecheck  # tsc: src + tests contra el checkout local del harness
-pnpm run typecheck:ci  # tsc contra los tipos publicados 0.1.2-alpha.3 (sin paths)
+pnpm run typecheck:ci  # tsc contra los tipos publicados 0.1.2-rc.1 (sin paths)
 pnpm test           # vitest: puertos de calidad, vocabulario núcleo, ensamblaje con pila real
 pnpm run build      # bundle tsdown + declaraciones tsc (lib/)
 pnpm run verify:self-contained  # las especificaciones de dependencias resuelven desde el registry
@@ -200,3 +200,7 @@ Este proyecto es uno de los [33 complementos de DeepSeek Harness](https://github
 ## License
 
 [Apache License 2.0](LICENSE) © 2026 dsh-library contributors
+
+### Instalar desde el mercado de DSH Desktop
+
+Todos los plugins de PerryLink pueden explorarse en el mercado integrado de DSH Desktop: **Market → Sources → add source → pegar** `https://perrylink-dsh-catalog.perrylink.workers.dev/catalog-source.json` **→ seleccionarlo**. La instalación sigue pasando por la verificación de identidad npm del mercado y tu confirmación.

package/README.hi.md

@@ -25,7 +25,7 @@
 
 | सतह | स्थिति |
 |---|---|
-| Harness | DeepSeek Harness `0.1.2-alpha.3` (`0.1.2-alpha.3` के लिए घोषित संगतता) |
+| Harness | DeepSeek Harness `0.1.2-rc.1` (`0.1.2-rc.1` के लिए घोषित संगतता) |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | भंडारण | कोई भी storage-domain बैकएंड (JSON या SQLite); सूचकांक होस्ट के भंडारण डोमेन में रहता है |
 | मॉडल | किसी की आवश्यकता नहीं — अंतर्निहित एम्बेडर नियतात्मक हैश है (शून्य डाउनलोड) |
@@ -120,8 +120,8 @@ dsh --profile web --dump-config | grep -A2 'id: dsh-library'
 
 - **अनुमतियाँ**: प्लगइन केवल वही फ़ाइलें पढ़ता है जिन्हें `library_add` इंगित करता है (harness फ़ाइल-सेवा और उसकी नीति से) और केवल अपने `dsh_library` भंडारण डोमेन में लिखता है। कोई नेटवर्क अनुरोध नहीं; वैकल्पिक बाहरी एम्बेडर `ctx.subprocess` से बिना शेल-व्याख्या चलता है।
 - **डेटा**: चंक पाठ और एम्बेडिंग होस्ट के भंडारण बैकएंड में रहते हैं (डिप्लॉयमेंट के बाकी स्थायी डेटा जैसा ही भरोसा); प्लगइन कोई एन्क्रिप्शन नहीं जोड़ता। दस्तावेज़ पथ और एम्बेडिंग कभी सत्र लॉग में नहीं जाते।
-- **सत्र लॉग**: `library/inject` (id, क्वेरी, चंक ids, पृष्ठ आकार) और `library/purge` (निर्णय) केवल-लॉग ऑडिट घटनाएँ हैं — मॉडल-दृश्य इंजेक्टेड पृष्ठ उनसे पुनर्निर्माण योग्य है। लिखावट होस्ट-गेटेड है: जिन हार्नेस का ज्ञात-प्रकार समूह शब्दावली को कवर करता है वे घटनाएँ पाते हैं, `ignorable`-लिफ़ाफ़े वाले बिल्ड उन्हें चिह्न के साथ पाते हैं, और लिफ़ाफ़ा-रहित बिल्ड (0.1.1-rc.2, 0.1.2-alpha.3) लिखावट छोड़ देते हैं — वहाँ लॉग किए गए `tool/call` + `tool/result` घटनाएँ पुनर्निर्माण योग्य ऑडिट-कड़ी बनी रहती हैं।
-0.1.2-alpha.3 (2026-09-01 को अनुकूलित): सत्र लिफ़ाफ़ा अपना ignorable फ़ील्ड केवल संग्रहीत-लॉग पठन संगतता के लिए रखता है - Session.append अभी भी इसे स्टैम्प नहीं कर सकता, इसलिए गेट व्यवहार अपरिवर्तित है।
+- **सत्र लॉग**: `library/inject` (id, क्वेरी, चंक ids, पृष्ठ आकार) और `library/purge` (निर्णय) केवल-लॉग ऑडिट घटनाएँ हैं — मॉडल-दृश्य इंजेक्टेड पृष्ठ उनसे पुनर्निर्माण योग्य है। लिखावट होस्ट-गेटेड है: जिन हार्नेस का ज्ञात-प्रकार समूह शब्दावली को कवर करता है वे घटनाएँ पाते हैं, `ignorable`-लिफ़ाफ़े वाले बिल्ड उन्हें चिह्न के साथ पाते हैं, और लिफ़ाफ़ा-रहित बिल्ड (0.1.1-rc.2, 0.1.2-rc.1) लिखावट छोड़ देते हैं — वहाँ लॉग किए गए `tool/call` + `tool/result` घटनाएँ पुनर्निर्माण योग्य ऑडिट-कड़ी बनी रहती हैं।
+0.1.2-rc.1 (2026-09-02 को अनुकूलित): सत्र लिफ़ाफ़ा अपना ignorable फ़ील्ड केवल संग्रहीत-लॉग पठन संगतता के लिए रखता है - Session.append अभी भी इसे स्टैम्प नहीं कर सकता, इसलिए गेट व्यवहार अपरिवर्तित है।
 
 ## सुरक्षा सीमाएँ
 
@@ -142,7 +142,7 @@ dsh --profile web --dump-config | grep -A2 'id: dsh-library'
 ```sh
 pnpm install        # node ^22.19 || >=24
 pnpm run typecheck  # tsc: src + tests स्थानीय हार्नेस चेकआउट के विरुद्ध
-pnpm run typecheck:ci  # tsc प्रकाशित 0.1.2-alpha.3 प्रकारों के विरुद्ध (बिना paths)
+pnpm run typecheck:ci  # tsc प्रकाशित 0.1.2-rc.1 प्रकारों के विरुद्ध (बिना paths)
 pnpm test           # vitest: गुणवत्ता पोर्ट, मूल शब्दावली, वास्तविक-स्टैक संयोजन
 pnpm run build      # tsdown बंडल + tsc घोषणाएँ (lib/)
 pnpm run verify:self-contained  # निर्भरता स्पेक registry से हल होती हैं
@@ -200,3 +200,7 @@ pnpm pack           # प्रकाशित tarball
 ## License
 
 [Apache License 2.0](LICENSE) © 2026 dsh-library contributors
+
+### DSH Desktop मार्केट से इंस्टॉल करें
+
+सभी PerryLink प्लगइन DSH Desktop के बिल्ट-इन मार्केट में देखे जा सकते हैं: **Market → Sources → add source → पेस्ट करें** `https://perrylink-dsh-catalog.perrylink.workers.dev/catalog-source.json` **→ चुनें**। इंस्टॉलेशन मार्केट के npm-identity सत्यापन और आपकी पुष्टि से ही होता है।

package/README.md

@@ -26,7 +26,7 @@
 
 | Surface | Status |
 |---|---|
-| Harness | DeepSeek Harness `0.1.2-alpha.3` (compat declared for `0.1.2-alpha.3`) |
+| Harness | DeepSeek Harness `0.1.2-rc.1` (compat declared for `0.1.2-rc.1`) |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | Storage | Any storage-domain backend (JSON or SQLite); the index lives in the host's storage domain |
 | Models | None required — the built-in embedder is deterministic hashing (zero downloads) |
@@ -121,8 +121,8 @@ All tunables are Schemastery `Config` fields (changeable from cordis.yml). An id
 
 - **Permissions**: the plugin only reads files you point `library_add` at (through the harness filesystem service and its policy) and writes into its own `dsh_library` storage domain. No network requests; an optional external embedder runs through `ctx.subprocess` without shell interpretation.
 - **Data**: chunk text and embeddings live in the host's storage backend (same trust as the deployment's other durable data); the plugin adds no encryption. Document paths and embeddings never enter the session log.
-- **Session log**: `library/inject` (id, query, chunk ids, page size) and `library/purge` (verdict) are log-only audit events — the model-visible injected page is reconstructable from them. The append is host-gated: harnesses whose known-type set covers the vocabulary get the events, `ignorable`-envelope builds get them with the marker, and envelope-less builds (0.1.1-rc.2, 0.1.2-alpha.3) skip the append — the logged `tool/call` + `tool/result` events remain the reconstructable audit trail there.
-0.1.2-alpha.3 (adapted 2026-09-01): the session envelope keeps its ignorable field for stored-log read compatibility only - Session.append still cannot stamp it, so audit-gate behavior is unchanged.
+- **Session log**: `library/inject` (id, query, chunk ids, page size) and `library/purge` (verdict) are log-only audit events — the model-visible injected page is reconstructable from them. The append is host-gated: harnesses whose known-type set covers the vocabulary get the events, `ignorable`-envelope builds get them with the marker, and envelope-less builds (0.1.1-rc.2, 0.1.2-rc.1) skip the append — the logged `tool/call` + `tool/result` events remain the reconstructable audit trail there.
+0.1.2-rc.1 (adapted 2026-09-02): the session envelope keeps its ignorable field for stored-log read compatibility only - Session.append still cannot stamp it, so audit-gate behavior is unchanged.
 
 ## Security boundaries
 
@@ -143,7 +143,7 @@ All tunables are Schemastery `Config` fields (changeable from cordis.yml). An id
 ```sh
 pnpm install        # node ^22.19 || >=24
 pnpm run typecheck  # tsc: src + tests against the local harness checkout
-pnpm run typecheck:ci  # tsc against the published 0.1.2-alpha.3 types (no paths)
+pnpm run typecheck:ci  # tsc against the published 0.1.2-rc.1 types (no paths)
 pnpm test           # vitest: quality ports, core vocabulary, real-stack assembly
 pnpm run build      # tsdown bundle + tsc declarations (lib/)
 pnpm run verify:self-contained  # dependency specs resolve from the registry
@@ -198,6 +198,10 @@ This project is one of the [33 DeepSeek Harness plugins](https://github.com/Perr
 | **[dsh-dsh-test-drive](https://github.com/PerryLink/dsh-dsh-test-drive)** | Isolated install-and-smoke test drives for DeepSeek Harness plugins. | |
 | **[dsh-dsh-translate](https://github.com/PerryLink/dsh-dsh-translate)** | Vendor parameter translation and deterministic JSON repair for DeepSeek Harness. | |
 
+### Install from the DSH Desktop Market
+
+All PerryLink plugins are browsable in the built-in DSH Desktop Market: **Market → Sources → add source → paste** `https://perrylink-dsh-catalog.perrylink.workers.dev/catalog-source.json` **→ select it**. Installation still goes through the Market's npm-identity verification and your confirmation.
+
 ## License
 
 [Apache License 2.0](LICENSE) © 2026 dsh-library contributors

package/README.pt.md

@@ -25,7 +25,7 @@
 
 | Superfície | Status |
 |---|---|
-| Harness | DeepSeek Harness `0.1.2-alpha.3` (compatibilidade declarada para `0.1.2-alpha.3`) |
+| Harness | DeepSeek Harness `0.1.2-rc.1` (compatibilidade declarada para `0.1.2-rc.1`) |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | Armazenamento | Qualquer backend de storage-domain (JSON ou SQLite); o índice vive no domínio de armazenamento do host |
 | Modelos | Nenhum necessário — o embedder integrado é hash determinístico (zero downloads) |
@@ -120,8 +120,8 @@ Todos os ajustes são campos `Config` do Schemastery (alteráveis pelo cordis.ym
 
 - **Permissões**: o plugin só lê os arquivos apontados pelo `library_add` (pelo serviço de arquivos do harness e sua política) e escreve no seu próprio domínio de armazenamento `dsh_library`. Sem requisições de rede; um embedder externo opcional executa via `ctx.subprocess` sem interpretação de shell.
 - **Dados**: o texto dos chunks e os embeddings vivem no backend de armazenamento do host (a mesma confiança do restante dos dados duráveis da implantação); o plugin não adiciona criptografia. Caminhos de documentos e embeddings nunca entram no registro de sessão.
-- **Registro de sessão**: `library/inject` (id, consulta, ids de chunks, tamanho da página) e `library/purge` (veredicto) são eventos de auditoria somente-registro — a página injetada visível ao modelo é reconstruível a partir deles. O append é condicionado pelo host: harnesses cujo conjunto de tipos conhecidos cobre o vocabulário recebem os eventos, builds com envelope `ignorable` os recebem com o marcador, e builds sem envelope (0.1.1-rc.2, 0.1.2-alpha.3) pulam o append — ali, os eventos registrados `tool/call` + `tool/result` continuam sendo a trilha de auditoria reconstruível.
-0.1.2-alpha.3 (adaptado em 2026-09-01): o envelope de sessão mantém seu campo ignorable apenas para compatibilidade de leitura de logs armazenados - o Session.append ainda não consegue estampá-lo, então o comportamento da porta não muda.
+- **Registro de sessão**: `library/inject` (id, consulta, ids de chunks, tamanho da página) e `library/purge` (veredicto) são eventos de auditoria somente-registro — a página injetada visível ao modelo é reconstruível a partir deles. O append é condicionado pelo host: harnesses cujo conjunto de tipos conhecidos cobre o vocabulário recebem os eventos, builds com envelope `ignorable` os recebem com o marcador, e builds sem envelope (0.1.1-rc.2, 0.1.2-rc.1) pulam o append — ali, os eventos registrados `tool/call` + `tool/result` continuam sendo a trilha de auditoria reconstruível.
+0.1.2-rc.1 (adaptado em 2026-09-02): o envelope de sessão mantém seu campo ignorable apenas para compatibilidade de leitura de logs armazenados - o Session.append ainda não consegue estampá-lo, então o comportamento da porta não muda.
 
 ## Limites de segurança
 
@@ -142,7 +142,7 @@ Todos os ajustes são campos `Config` do Schemastery (alteráveis pelo cordis.ym
 ```sh
 pnpm install        # node ^22.19 || >=24
 pnpm run typecheck  # tsc: src + tests contra o checkout local do harness
-pnpm run typecheck:ci  # tsc contra os tipos publicados 0.1.2-alpha.3 (sem paths)
+pnpm run typecheck:ci  # tsc contra os tipos publicados 0.1.2-rc.1 (sem paths)
 pnpm test           # vitest: portas de qualidade, vocabulário núcleo, montagem com pilha real
 pnpm run build      # bundle tsdown + declarações tsc (lib/)
 pnpm run verify:self-contained  # especificações de dependências resolvem pelo registry
@@ -200,3 +200,7 @@ Este projeto é um dos [33 plugins de DeepSeek Harness](https://github.com/Perry
 ## License
 
 [Apache License 2.0](LICENSE) © 2026 dsh-library contributors
+
+### Instalar a partir do mercado do DSH Desktop
+
+Todos os plugins PerryLink podem ser explorados no mercado integrado do DSH Desktop: **Market → Sources → add source → colar** `https://perrylink-dsh-catalog.perrylink.workers.dev/catalog-source.json` **→ selecionar**. A instalação continua passando pela verificação de identidade npm do mercado e pela sua confirmação.

package/README.zh.md

@@ -25,7 +25,7 @@
 
 | 方面 | 状态 |
 |---|---|
-| Harness | DeepSeek Harness `0.1.2-alpha.3`（声明兼容 `0.1.2-alpha.3`） |
+| Harness | DeepSeek Harness `0.1.2-rc.1`（声明兼容 `0.1.2-rc.1`） |
 | Node | `^22.19.0 \|\| >=24.0.0` |
 | 存储 | 任意 storage-domain 后端（JSON 或 SQLite）；索引存放在宿主的存储域中 |
 | 模型 | 无需任何模型 —— 内置嵌入为确定性哈希（零下载） |
@@ -120,8 +120,8 @@ dsh --profile web --dump-config | grep -A2 'id: dsh-library'
 
 - **权限**：插件只读取你让 `library_add` 指向的文件（经 harness 文件系统服务及其策略），并只写入自己的 `dsh_library` 存储域。无网络请求；可选外部嵌入命令经 `ctx.subprocess` 执行、无 shell 解释。
 - **数据**：切片文本与嵌入存放在宿主的存储后端（与部署的其他持久数据同级信任）；插件不额外加密。文档路径与嵌入向量绝不进入会话日志。
-- **会话日志**：`library/inject`（id、查询、切片 id、页大小）与 `library/purge`（判定）是仅日志审计事件 —— 模型可见的注入页可从中重建。写入受宿主门控：已知类型集合覆盖该词汇的宿主直接写入；带 `ignorable` 信封的构建加标记写入；无信封构建（0.1.1-rc.2、0.1.2-alpha.3）跳过写入 —— 此时已记录的 `tool/call` + `tool/result` 事件仍是可重建的审计链。
-0.1.2-alpha.3（2026-09-01 已适配）：会话信封保留 ignorable 字段但仅用于存量日志读取兼容——Session.append 仍无法盖章，门控行为不变。
+- **会话日志**：`library/inject`（id、查询、切片 id、页大小）与 `library/purge`（判定）是仅日志审计事件 —— 模型可见的注入页可从中重建。写入受宿主门控：已知类型集合覆盖该词汇的宿主直接写入；带 `ignorable` 信封的构建加标记写入；无信封构建（0.1.1-rc.2、0.1.2-rc.1）跳过写入 —— 此时已记录的 `tool/call` + `tool/result` 事件仍是可重建的审计链。
+0.1.2-rc.1（2026-09-02 已适配）：会话信封保留 ignorable 字段但仅用于存量日志读取兼容——Session.append 仍无法盖章，门控行为不变。
 
 ## 安全边界
 
@@ -142,7 +142,7 @@ dsh --profile web --dump-config | grep -A2 'id: dsh-library'
 ```sh
 pnpm install        # node ^22.19 || >=24
 pnpm run typecheck  # tsc：src + tests 对照本地 harness checkout
-pnpm run typecheck:ci  # tsc：对照已发布的 0.1.2-alpha.3 类型（无 paths）
+pnpm run typecheck:ci  # tsc：对照已发布的 0.1.2-rc.1 类型（无 paths）
 pnpm test           # vitest：八移植回归、核心词汇、真实栈装配
 pnpm run build      # tsdown bundle + tsc 声明（lib/）
 pnpm run verify:self-contained  # 依赖声明全部来自 registry
@@ -200,3 +200,7 @@ pnpm pack           # 发布用 tarball
 ## License
 
 [Apache License 2.0](LICENSE) © 2026 dsh-library contributors
+
+### 从 DSH Desktop 市场安装
+
+所有 PerryLink 插件均可在 DSH Desktop 内置市场中浏览：**市场 → 来源 → 添加来源 → 粘贴** `https://perrylink-dsh-catalog.perrylink.workers.dev/catalog-source.json` **→ 选中**。安装仍需通过市场的 npm 身份校验与你的确认。

package/lib/index.js

@@ -4263,16 +4263,42 @@ function tokenize(text) {
 	return tokens;
 }
 /**
-* Term-frequency map of one text: token → occurrence count.
+* CJK-aware scoring tokenization: like {@link tokenize}, but every CJK run
+* contributes its unigrams plus its adjacent bigrams, so a query and a
+* document overlap on sub-phrases instead of only on identical whole runs.
 * @param text - input text.
+* @returns the expanded token list.
+*/
+function cjkTokenize(text) {
+	const tokens = [];
+	for (const match of normalize(text).matchAll(/[a-z0-9]+|[\u00c0-\uffff]+/giu)) {
+		const run = match[0];
+		if (/^[\u4e00-\u9fff]+$/u.test(run)) {
+			for (const char of run) tokens.push(char);
+			for (let index = 0; index < run.length - 1; index += 1) tokens.push(run.slice(index, index + 2));
+		} else tokens.push(run);
+	}
+	return tokens;
+}
+/**
+* Term-frequency map of a token list: token → occurrence count.
+* @param tokens - the token list.
 * @returns the frequency map (plain object, no prototype).
 */
-function termFrequencies(text) {
+function termFrequenciesOf(tokens) {
 	const frequencies = Object.create(null);
-	for (const token of tokenize(text)) frequencies[token] = (frequencies[token] ?? 0) + 1;
+	for (const token of tokens) frequencies[token] = (frequencies[token] ?? 0) + 1;
 	return frequencies;
 }
 /**
+* Term-frequency map of one text: token → occurrence count.
+* @param text - input text.
+* @returns the frequency map (plain object, no prototype).
+*/
+function termFrequencies(text) {
+	return termFrequenciesOf(tokenize(text));
+}
+/**
 * Character n-grams of a normalized text, for the built-in hash embedder.
 * A space-padded form adds leading/trailing boundary markers so prefixes and
 * suffixes still contribute.
@@ -5026,7 +5052,9 @@ function middlePenalty(bins) {
 * retrieval. The upstream cross-encoder model is replaced by a local-rules
 * lexical scorer (zero downloads): token overlap weighted toward rarer
 * query terms, normalized to 0..1 — same `(document, score, passed)` API
-* shape as the upstream `score_batch`/`filter_documents`.
+* shape as the upstream `score_batch`/`filter_documents`. Scoring tokenizes
+* CJK runs into unigrams plus adjacent bigrams so partial phrase overlaps
+* count, instead of whole runs matching only on exact equality.
 * @module dsh-library/quality/relevance
 */
 /**
@@ -5040,9 +5068,10 @@ function middlePenalty(bins) {
 * @returns relevance within 0..1 (0 for an empty query or document).
 */
 function scoreRelevance(query, document) {
-	if (tokenize(query).length === 0) return 0;
-	const documentTerms = termFrequencies(document);
-	const queryTerms = termFrequencies(query);
+	const queryTokens = cjkTokenize(query);
+	if (queryTokens.length === 0) return 0;
+	const documentTerms = termFrequenciesOf(cjkTokenize(document));
+	const queryTerms = termFrequenciesOf(queryTokens);
 	let covered = 0;
 	let total = 0;
 	for (const [token, count] of Object.entries(queryTerms)) {
@@ -5684,7 +5713,7 @@ var LibraryStore = class {
 			relevance: candidate.hybrid,
 			chunk: candidate.chunk,
 			similarityTo: (other) => cosine(candidate.chunk.embedding, other.chunk.embedding)
-		})), this.config.search.diversityLambda, Math.min(candidates.length, MAX_CANDIDATE_BOUND)).filter((entry) => scoreRelevance(query, entry.chunk.text) >= this.config.search.minRelevance).slice(0, 500);
+		})), this.config.search.diversityLambda, Math.min(candidates.length, MAX_CANDIDATE_BOUND)).filter((entry) => entry.relevance >= this.config.search.minRelevance).slice(0, 500);
 		const reordered = avoidLostMiddle(filtered.map((entry) => ({
 			id: entry.chunk.chunkId,
 			score: entry.relevance

package/lib/types/quality/relevance.d.ts

@@ -4,7 +4,9 @@
  * retrieval. The upstream cross-encoder model is replaced by a local-rules
  * lexical scorer (zero downloads): token overlap weighted toward rarer
  * query terms, normalized to 0..1 — same `(document, score, passed)` API
- * shape as the upstream `score_batch`/`filter_documents`.
+ * shape as the upstream `score_batch`/`filter_documents`. Scoring tokenizes
+ * CJK runs into unigrams plus adjacent bigrams so partial phrase overlaps
+ * count, instead of whole runs matching only on exact equality.
  * @module dsh-library/quality/relevance
  */
 /**

package/lib/types/quality/relevance.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"relevance.d.ts","sourceRoot":"","sources":["../../../src/quality/relevance.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAIH;;;;;;;;;GASG;AACH,wBAAgB,cAAc,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,CAatE;AAED,uEAAuE;AACvE,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;IACzB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;IACtB,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAA;CACzB;AAED;;;;;;GAMG;AACH,wBAAgB,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,MAAM,EAAE,EAAE,SAAS,SAAM,GAAG,cAAc,EAAE,CAKzG;AAED;;;;;;;GAOG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,MAAM,EAAE,EAAE,SAAS,SAAM,GAAG,MAAM,EAAE,CAItG"}
+{"version":3,"file":"relevance.d.ts","sourceRoot":"","sources":["../../../src/quality/relevance.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAIH;;;;;;;;;GASG;AACH,wBAAgB,cAAc,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,MAAM,CAatE;AAED,uEAAuE;AACvE,MAAM,WAAW,cAAc;IAC7B,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;IACzB,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;IACtB,QAAQ,CAAC,MAAM,EAAE,OAAO,CAAA;CACzB;AAED;;;;;;GAMG;AACH,wBAAgB,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,MAAM,EAAE,EAAE,SAAS,SAAM,GAAG,cAAc,EAAE,CAKzG;AAED;;;;;;;GAOG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,SAAS,MAAM,EAAE,EAAE,SAAS,SAAM,GAAG,MAAM,EAAE,CAItG"}

package/lib/types/text.d.ts

@@ -18,8 +18,22 @@ export declare function normalize(text: string): string;
  * @returns the token list.
  */
 export declare function tokenize(text: string): string[];
+/**
+ * CJK-aware scoring tokenization: like {@link tokenize}, but every CJK run
+ * contributes its unigrams plus its adjacent bigrams, so a query and a
+ * document overlap on sub-phrases instead of only on identical whole runs.
+ * @param text - input text.
+ * @returns the expanded token list.
+ */
+export declare function cjkTokenize(text: string): string[];
 /** Deduplicated token set of one text. */
 export declare function tokenSet(text: string): ReadonlySet<string>;
+/**
+ * Term-frequency map of a token list: token → occurrence count.
+ * @param tokens - the token list.
+ * @returns the frequency map (plain object, no prototype).
+ */
+export declare function termFrequenciesOf(tokens: readonly string[]): Record<string, number>;
 /**
  * Term-frequency map of one text: token → occurrence count.
  * @param text - input text.

package/lib/types/text.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"text.d.ts","sourceRoot":"","sources":["../../src/text.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH;;;;;GAKG;AACH,wBAAgB,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAE9C;AAED;;;;;GAKG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,CAO/C;AAED,0CAA0C;AAC1C,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,CAE1D;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAIpE;AAED;;;;;;;GAOG;AACH,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAQ5D;AAED;;;;;GAKG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAQ1E;AAED;;;;GAIG;AACH,wBAAgB,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAO1C"}
+{"version":3,"file":"text.d.ts","sourceRoot":"","sources":["../../src/text.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH;;;;;GAKG;AACH,wBAAgB,SAAS,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAE9C;AAED;;;;;GAKG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,CAO/C;AAED;;;;;;GAMG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,CAalD;AAED,0CAA0C;AAC1C,wBAAgB,QAAQ,CAAC,IAAI,EAAE,MAAM,GAAG,WAAW,CAAC,MAAM,CAAC,CAE1D;AAED;;;;GAIG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAInF;AAED;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAEpE;AAED;;;;;;;GAOG;AACH,wBAAgB,UAAU,CAAC,IAAI,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAQ5D;AAED;;;;;GAKG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,SAAS,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,GAAG,MAAM,EAAE,CAQ1E;AAED;;;;GAIG;AACH,wBAAgB,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAO1C"}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "dsh-library",
-  "version": "0.2.2",
-  "description": "Local document knowledge base for DeepSeek Harness: library_add/remove/list, hybrid semantic+keyword library_search with diversity re-ranking, relevance filtering and lost-in-the-middle avoidance, citation-aware injection, library_cite_check and library_diagnose 闁炽儲鎽琎Lite-backed index via the storage domain, local embedding, zero model downloads",
+  "version": "0.2.4",
+  "description": "Local document knowledge base for DeepSeek Harness: library_add/remove/list, hybrid semantic+keyword library_search with diversity re-ranking, relevance filtering and lost-in-the-middle avoidance, citation-aware injection, library_cite_check and library_diagnose 閳ユ摬QLite-backed index via the storage domain, local embedding, zero model downloads",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/PerryLink/dsh-library.git"
@@ -67,7 +67,7 @@
     ],
     "compatibility": {
       "dshVersions": [
-        "0.1.2-alpha.3"
+        "0.1.2-rc.1"
       ]
     },
     "capability": {
@@ -119,22 +119,23 @@
     "typescript": "^5.9.0",
     "zod": "^4.4.3"
   },
-  "devDependencies": {"@deepseek-ai/dsh-attachment": "0.1.2-alpha.3", 
+  "devDependencies": {
+    "@deepseek-ai/dsh-attachment": "0.1.2-rc.1",
     "@deepseek-ai/cordis": "^4.0.2",
-    "@deepseek-ai/cordis-plugin-include": "1.0.6",
-    "@deepseek-ai/cordis-plugin-loader": "1.0.2",
-    "@deepseek-ai/dsh-agent": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-commands": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-fs": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-fs-local": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-llm": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-session": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-storage": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-storage-domain": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-storage-json": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-subprocess": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-subprocess-local": "0.1.2-alpha.3",
-    "@deepseek-ai/dsh-tools": "0.1.2-alpha.3",
+    "@deepseek-ai/cordis-plugin-include": "^1.0.7",
+    "@deepseek-ai/cordis-plugin-loader": "^1.0.3",
+    "@deepseek-ai/dsh-agent": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-commands": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-fs": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-fs-local": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-llm": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-session": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-storage": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-storage-domain": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-storage-json": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-subprocess": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-subprocess-local": "0.1.2-rc.1",
+    "@deepseek-ai/dsh-tools": "0.1.2-rc.1",
     "@deepseek-ai/schemastery": "^3.18.2",
     "@types/node": "^22.19.0",
     "@vitest/coverage-v8": "3.2.0",