prts-terrarchive 0.2.2 → 0.2.3

package/README.en.md

@@ -55,8 +55,9 @@ change as service capacity evolves.
 
 ## Install
 
-Web requires Node.js >= 22.19 and DSH >= 0.1.2-alpha.2. The current compatibility
-target is DSH 0.2.0-rc.1. An official Electron Desktop installation supplies
+Web requires Node.js >= 22.19 and DSH >= 0.2.0-rc.1 (declared as `engines.dsh`).
+The package has been audited point by point against the DSH 0.2.1-rc.2 source.
+An official Electron Desktop installation supplies
 its own runtime. Corpus disk usage is shown in Settings before download.
 
 ### Official Electron Desktop (recommended)
@@ -64,13 +65,13 @@ its own runtime. Corpus disk usage is shown in Settings before download.
 Recent DSH Desktop builds let you install from the UI without a terminal:
 
 1. Open **Plugins** in the sidebar, then click **+ Add plugin**.
-2. Paste the npm package name into **Package name or address**. Enter only the name, **not an `npm install` or `dsh plugin add` command**:
+2. Paste the versioned npm package name into **Package name or address**. Enter only the name, **not an `npm install` or `dsh plugin add` command**:
 
    ```text
-   prts-terrarchive
+   prts-terrarchive@0.2.3
    ```
 
-   Without a version, this installs npm's `latest`. To pin this release, enter `prts-terrarchive@0.2.2`.
+   Specify a version. With only `prts-terrarchive`, the preview may show npm's `latest` while the installer selects an older version; see the explanation below.
 3. Choose the official npm registry or the mainland China mirror under **Registry**, then click **Install**.
 4. Click **Enable now** after installation. If you close the dialog instead, turn on the plugin in the installed list.
 5. Open **Settings → Plugins → PRTS Corpus** to download the corpus and choose a skin, then start a new session in **PRTS mode**.
@@ -79,7 +80,7 @@ The same dialog accepts these alternatives. Choose one; you do not need to insta
 
 | Source | Text to enter | Use |
 | --- | --- | --- |
-| npm (recommended) | `prts-terrarchive` | Install a published release |
+| npm (recommended) | `prts-terrarchive@0.2.3` | Install the specified release |
 | GitHub | `https://github.com/HTian-qwq/prts-terrarchive.git` | Install repository source; requires access to GitHub |
 | Local directory | `D:\ds\prts-terrarchive` | Install a complete downloaded checkout; replace this with its actual absolute path |
 
@@ -91,6 +92,14 @@ does not require a separate `js-yaml` / `zod` installation or `npm ci`.
 If GitHub is unreachable, enter the npm package name and choose a mirror instead. If a mirror
 does not yet have a newly published version, try the official npm registry.
 
+**Installation preview and installed version differ:** DSH builds using pnpm 11 default to
+preferring versions published at least 24 hours ago, while the preview reads registry metadata.
+For example, shortly after 0.2.3 is published, entering only the package name can preview 0.2.3
+but install 0.2.1. This difference comes from the installer's version selection; see the
+[pnpm release notes](https://pnpm.io/blog/releases/11.0). Follow the upgrade steps below to
+uninstall the old version, install `prts-terrarchive@0.2.3`, then check the version in plugin
+details. You do not need to clear corpus or session data.
+
 **Upgrades:** the current Desktop installation UI does not provide automatic updates.
 For npm or GitHub installs, uninstall from the plugin details, then add and enable the plugin
 again; change any pinned version to your target release. For a local checkout, run `git pull`
@@ -134,7 +143,7 @@ Install the bundle directly from npm; no additional installer command
 or npm lifecycle hook is needed:
 
 ```bash
-dsh plugin --profile web add prts-terrarchive@0.2.2
+dsh plugin --profile web add prts-terrarchive@0.2.3
 ```
 
 Restart `dsh web`. On DSH 0.1.7 and later the plugin registers PRTS mode
@@ -163,8 +172,11 @@ Web and official Desktop share the default `$DSH_HOME` product data while
 installing their plugins separately. The plugin package includes its UI,
 map models, textures, skills, and presets. It does **not** bundle corpus data,
 Node/DSH runtimes, or user data. Open Settings → Plugins → PRTS 语料 and download
-the corpus explicitly before using local retrieval. Switching skins does not
-start a separate model or texture download.
+the corpus explicitly before using local retrieval. Downloads follow `enabledGames`:
+Endfield needs its two packs; Arknights needs its five packs. Enabling another game
+and downloading again adds the missing packs and reuses verified files. Disabling
+a game does not delete its installed data. Switching skins does not start a
+separate model or texture download.
 
 PRTS Portable distributions include the plugin, presets, and the complete corpus,
 so no separate corpus download is needed after extraction. Build the official
@@ -234,10 +246,11 @@ download.
   `{n}` repetitions; no grouping, alternation or variable quantifiers) on NFKC-normalized text, with resource-type /
   character / story / activity / entity / speaker / Wiki-field filters;
   keep the original conditions and pass `page.next_after` back unchanged as
-  `after` to continue, because the readable anchor includes the full
-  `data_version` and becomes invalid after a corpus release switch
-- `corpus_read` — read Arknights stages by `stage_code` (adding `story_part`
-  only when needed), operator records by character/name/segment, and character
+  `after` to continue; the anchor carries a short `data_version` prefix and
+  becomes invalid after a corpus release switch. `speakers` expands registered
+  aliases; hits keep original punctuation and mark the match with `⟦⟧`
+- `corpus_read` — read Arknights stages by `stage_code` (`story_part` is
+  optional; ambiguous codes default to the pre-battle part with a note), operator records by character/name/segment, and character
   materials by category; it can stream a whole Arknights activity or Endfield
   mission collection with readable, version-bound continuation positions;
   it can also read a tagged Wiki field or page through a whole document; story
@@ -247,8 +260,11 @@ download.
   when a character locator needs same-name disambiguation.
   Use exactly one primary locator per call. If search returns `document_uid`, it
   replaces `title`; use `{document_uid, line}` or
-  `{document_uid, mode:"document"}`, never both locators. `max_lines` and
-  `max_chars` only cap output and do not select a read mode
+  `{document_uid, mode:"document"}`, never both locators. A bare `title` reads
+  the document from line 1; titles tolerate missing separators and Wiki
+  sequence suffixes, and unknown titles list the closest candidates.
+  `activity_name` and `collection_name` reroute to the other game when the
+  name only exists there
 - `timeline_search` — query the Terra calendar timeline; source markers can
   be resolved back to full provenance
 - `cloud_search` / `cloud_inspect` — optional PRTS.cloud hybrid retrieval
@@ -260,7 +276,19 @@ download.
 Repeated and partially reused reads retain the normal page limits and
 continuation. Partial reads include `coverage` identifying previously visible
 lines and newly returned lines; `page.has_more` still determines whether the
-document has more content. Cloud answers remain available when local corpus
+document has more content. Reuse trusts the host's "currently visible surface":
+if earlier output was compacted or collapsed out of context, that page can look
+present while being unreadable — pass `reuse: false` to force a full resend
+instead of tuning `max_lines`.
+
+To continue a document read, pass `start_line` (not `line`): `line` means the
+centre line of an `around` window (three lines each way by default) and is
+rejected with `mode: "document"`. Every line hit in `corpus_search` also gets a
+handle such as `doc_xxxxxxxxxxxxxxxx#L12`; feed it back with
+`corpus_read({"citation":"doc_xxxxxxxxxxxxxxxx#L12"})` to return to those exact
+lines without looking up a `document_uid` first. Zero-hit searches report the
+covered scope (`scope`: documents and lines), and results carry a per-game hit
+distribution (`by_game`) so cross-game homonyms can be narrowed with `games`. Cloud answers remain available when local corpus
 mapping fails, with a warning to install or repair the corpus before verifying
 the original text.
 
@@ -282,8 +310,14 @@ for field semantics and query recipes.
 DSH 0.1.2-alpha.1 and 0.1.2-alpha.2 have completed historical real-host tests
 with the web profile. Alpha.1 was built from the official tag and passed
 installation, preset resolution, host startup, settings-route, and
-client-bundle checks. The current plugin compatibility target is DSH
-0.2.0-rc.1. The shared Connection Fetch transport supports Web and the official
+client-bundle checks. The declared minimum runtime is DSH 0.2.0-rc.1
+(`engines.dsh`), and every contract this plugin relies on — tool registration,
+systemPrompt contexts, Connection Fetch routes, skills, presets and the client
+module protocol — was checked against the 0.2.1-rc.2 source. Since 0.2.1 exact
+Fetch routes admit only the owner principal, so a desktop carrier must declare
+that principal; PTC sub-dispatches also persist `presentationMeta`, but the
+client never reads it, so evidence cards still come from the plugin's receipt.
+The shared Connection Fetch transport supports Web and the official
 Electron source implementation; Web also retains HTTP routes. Declarative
 registration and legacy file seeding pass this project's unit tests. Tests against
 0.1.7-rc.2 (`477b4f420553`) also exercise real Cordis, Tools, SystemPrompt,

package/README.md

@@ -71,7 +71,7 @@ npm run preview:rhine
 
 ## 环境要求
 
-- Web：Node.js **≥ 22.19**，DSH 运行时 **≥ 0.1.2-alpha.2**；当前适配目标为 **0.2.0-rc.1**
+- Web：Node.js **≥ 22.19**，DSH 运行时 **≥ 0.2.0-rc.1**（包内 `engines.dsh` 声明）；已按 **0.2.1-rc.2** 源码逐项核对
 - 官方 Electron Desktop：使用与插件兼容的 DSH 桌面版本；桌面自带运行时
 - 磁盘空间：语料大小以设置页与 release 清单为准，下载前不会自动占用完整语料空间
 
@@ -82,13 +82,13 @@ npm run preview:rhine
 在新版 DSH 桌面端，无需打开终端：
 
 1. 打开左侧 **插件**，点击右上角 **＋ 添加插件**。
-2. 在「包名或地址」输入框中粘贴下面的 npm 包名；**只填包名，不要填 `npm install` 或 `dsh plugin add` 命令**：
+2. 在「包名或地址」输入框中粘贴下面带版本号的 npm 包名；**只填包名，不要填 `npm install` 或 `dsh plugin add` 命令**：
 
    ```text
-   prts-terrarchive
+   prts-terrarchive@0.2.3
    ```
 
-   不带版本号时安装 npm 的 `latest`；需要固定本次发布版本时，填 `prts-terrarchive@0.2.2`。
+   建议指定版本。只填 `prts-terrarchive` 时，安装预览可能显示 npm 的 `latest`，实际安装却选择较旧的版本，原因见下方说明。
 3. 在 **安装源** 中按网络情况选择「npm 官方源」或「中国大陆镜像源」，点击 **安装**。
 4. 安装完成后点击 **立即启用**；如果直接关闭了对话框，请在已安装列表中打开插件开关。
 5. 打开 **设置 → 插件 → PRTS 语料** 下载资料并选择皮肤，然后新建会话，选择 **PRTS 模式**。
@@ -97,7 +97,7 @@ npm run preview:rhine
 
 | 来源 | 输入框中填写的内容 | 适用情况 |
 | --- | --- | --- |
-| npm（推荐） | `prts-terrarchive` | 安装已发布版本 |
+| npm（推荐） | `prts-terrarchive@0.2.3` | 安装指定发布版本 |
 | GitHub | `https://github.com/HTian-qwq/prts-terrarchive.git` | 安装仓库源码，需要本机能访问 GitHub |
 | 本地目录 | `D:\ds\prts-terrarchive` | 安装已下载的完整插件目录；替换为本机实际绝对路径 |
 
@@ -105,6 +105,8 @@ npm run preview:rhine
 
 **安装源只影响 npm 包及依赖下载，不能代理 GitHub 仓库或压缩包直链。** 如果 GitHub 无法连接，可以改填 npm 包名并选择镜像源。新发布版本在镜像中尚未可用时，可切换「npm 官方源」重试。
 
+**安装预览与已安装版本不同：** 使用 pnpm 11 的 DSH 默认优先安装发布满 24 小时的版本，而安装预览读取的是注册表版本。例如，0.2.3 刚发布时，只填包名可能预览 0.2.3、实际安装 0.2.2。这是安装器的版本选择差异。[pnpm 官方说明](https://pnpm.io/blog/releases/11.0)介绍了这项默认设置。请按下面的升级流程卸载旧版，再填 `prts-terrarchive@0.2.3` 安装，完成后在插件详情中核对版本；无需清除语料或会话数据。
+
 **升级插件：** 当前桌面安装界面不提供自动更新。需要升级 npm 或 GitHub 安装的插件时，在插件详情中卸载，再通过「添加插件」重新安装并启用；指定版本号时记得换成目标版本。本地源码安装先在原目录执行 `git pull`，再完全退出并重启桌面端。插件更新与语料更新分别进行，资料版本在「PRTS 语料」中管理。
 
 旧版桌面如只支持 npm 包名，请使用 `prts-terrarchive` 并按其界面提示启用。Electron 的 `desktop` profile 由应用管理，不要运行 `node bin/install.js desktop` 或 `dsh plugin --profile desktop`。
@@ -134,7 +136,7 @@ node bin/install.js web /path/to/prts-terrarchive
 可直接从 npm 安装插件 bundle，无需额外运行安装脚本：
 
 ```bash
-dsh plugin --profile web add prts-terrarchive@0.2.2
+dsh plugin --profile web add prts-terrarchive@0.2.3
 ```
 
 重启 `dsh web` 后，DSH `0.1.7` 起由插件注册「PRTS 模式」，禁用或卸载插件时
@@ -179,8 +181,7 @@ Web 本地安装需要可用的 `dsh.cmd`。安装器经 cmd.exe 调用它，插
 
 1. **重启** `dsh web` 或所用桌面应用；
 2. **设置 → 插件 →「PRTS 语料」**：选择皮肤（Harness 默认 / PRTS Agent / Endfield AIC / 莱茵生命资料馆）；皮肤模型、贴图与字体已随插件包安装；
-3. **版本管理**：Portable 已附带完整语料；其他安装方式下载双游戏资料（优先 ModelScope 镜像；大小以设置页与当前 release
-   清单为准）。未安装资料时
+3. **版本管理**：Portable 已附带完整语料；其他安装方式按设置中的启用游戏下载安装（优先 ModelScope 镜像）。只启用终末地时下载终末地官方资料和审校知识两包；只启用明日方舟时下载方舟的五包。之后启用另一款游戏，再点击下载或更新会补齐缺少的包，已校验文件会复用；关闭游戏不会删除已安装资料。未安装资料时
    PRTS 模式仍可进入；调用本地工具会提醒前往本设置页安装；
 4. **新建会话，顶部模式下拉选「PRTS 模式」**；
 5. 直接用自然语言提问，或让模型调用下列工具。
@@ -233,7 +234,7 @@ corpus_i18n({title: "<检索结果的完整标题>", line: 1, languages: ["EN"]}
 | `match_mode` | `literal`（默认，连续字面匹配）/ `regex`（线性安全子集：字面量、`^`/`$`、`.`、字符类、转义和固定次数 `{n}`；不支持分组、分支及可变量词） |
 | `character_names` / `story_names` / `activity_names` | 角色 / 篇章 / 活动展示名过滤 |
 | `entity_names` | 只返回出现指定实体的行（自动展开别名） |
-| `speakers` | 结构化说话人过滤，只匹配亲口台词 |
+| `speakers` | 结构化说话人过滤，只匹配亲口台词（自动展开已登记别名，如 银灰→恩希欧迪斯） |
 | `wiki_sections` | Wiki 标签字段过滤（相关活动、剧情总结、角色剧情概括等 16 种） |
 | `context_terms` | 要求命中附近（±3 行）同时出现的语境词，≤8 个 |
 | `after` | 下一页的版本绑定可读锚点 `{ data_version, resource_type, title, position }`；与原搜索条件一起原样提交 |
@@ -257,26 +258,35 @@ corpus_i18n({title: "<检索结果的完整标题>", line: 1, languages: ["EN"]}
 
 ### corpus_read — 原文阅读
 
-优先使用游戏内关卡代号、密录名、角色名、活动名或任务名；无法稳定定位时再使用搜索返回的完整标题。
+优先使用游戏内关卡代号、密录名、角色名、活动名或任务名；无法稳定定位时再使用搜索返回的标题；标题可省略分隔符与 Wiki 序号后缀，找不到时会列出最接近的完整标题。
 不使用内部 `document_id`、ref 或路径。搜索结果只在同名歧义时给出公开稳定的 `document_uid`；它替代 `title`，不得同时提交两者。同名多合集不会自动合并。
 
 | 用法 | 参数 |
 | --- | --- |
-| 明日方舟关卡 | `stage_code`；同代号多篇时加 `story_part`：`before` / `after` / `story`；加 `line` 可定点读上下文 |
+| 明日方舟关卡 | `stage_code`；同代号多篇时可加 `story_part`：`before` / `after` / `story`（省略时默认行动前并提示其他篇）；加 `line` 可定点读上下文 |
 | 干员密录 | `character_name` + `record_name`；多段密录加 `segment` |
 | 角色官方资料 | `character_name` + `material`，类别为 `profile/module/voice/skin/recruitment/potential` |
 | 双模块同名角色消歧 | 在角色定位参数之外加 `game: "arknights"` 或 `game: "endfield"`；无同名歧义时不填 |
 | 整活动连续阅读 | `activity_name` + `mode: "activity"` |
 | 终末地任务连续阅读 | `collection_name` + `mode: "collection"`，可选 `content_types` |
-| 其他资料定点上下文 | `title` + `line` |
+| 其他资料定点上下文 | `title` + `line`（`line` 是中心行，默认前后各 3 行；不用于全文续读） |
+| 终末地角色档案 | `character_name` + `material: "profile"`（与明日方舟同一套；如 `{"character_name":"管理员","material":"profile"}`） |
 | 读 Wiki 字段 | `title` + `section`（精确读取标签字段，不含标签行） |
-| 分页读其他全文 | `title` + `mode: "document"`（`max_lines` 默认 100、上限 500；`max_chars` 默认 12000、上限 100000） |
+| 分页读其他全文 | `title` + `mode: "document"`，要从中间开始时加 `start_line`（`max_lines` 默认 100、上限 500；`max_chars` 默认 12000、上限 100000） |
 | 同名单篇 | `document_uid` + `line`，或 `document_uid` + `mode: "document"`；不再提交 `title` |
-| 续读 | 原样提交 `page.continuation`；单篇使用 `line`，活动/任务使用版本绑定的 `position`；旧 `cursor` 仅兼容历史会话 |
+| 续读 | 复制结果末尾的续读调用；单篇续页给 `start_line`，活动/任务给 `position`，`data_version` 为短版本号且可省略；旧 `cursor` 仅兼容历史会话 |
+
+`mode: "document"` 不接受 `line`（除 `line: 1` 这类无害占位）：两者语义不同，混用会让续读停在同一段并误显示为「复用上文」。`start_line` 只用于全文连续读取。
 
 每次读取只选一种主定位方式。`max_lines/max_chars` 只限制输出量，不能代替 `line`、`section` 或 `mode`。
 
 重复读取与部分补读保持正常的预算和续页位置。部分补读会用 `coverage` 标明本页复用的上文行及新返回行；全文是否读完仍以 `page.has_more` 为准。
+复用只看宿主给出的「当前可见 surface」：长会话里上文若已被压缩或折叠，那一页会既看不到行、又被判定为可见。此时传 `reuse: false` 强制重发本页全文，不要改 `max_lines` 绕着试。
+
+每条命中还会带「句柄」：`doc_xxxxxxxxxxxxxxxx#L12`（区间为 `#L12-30`）。它同时是篇章定位和行号，直接把句柄塞回读取即可，不必先查 `document_uid`：
+`corpus_read({"citation":"doc_xxxxxxxxxxxxxxxx#L12"})`。`citation` 与 `title`/`document_uid`/`line` 互斥，格式错误会直接报 `INVALID_REQUEST`。
+
+零命中会给出检索覆盖范围（`scope`：本地篇数/行数），例如「已检查完整检索范围（本地 34007 篇 / 1133511 行），没有找到」——「没有」因此可以当作否定证据引用，但范围仍以当前安装的资料包为准。
 
 引用格式统一为「《篇章名》第 N 行」。剧情文档只返回请求的原文；剧情总结与活动
 时间线必须通过 `timeline_search` 或 Wiki 字段显式检索，不自动夹带。
@@ -349,6 +359,25 @@ Wiki 资料不是无差别文本池：工具会区分规范角色页、活动/
 重新定位到当前版本后再调用即可。旧任务不会回写新版本的正文或倒排缓存，缓存复用
 及跨篇章读取也受同一版本检查约束。
 
+### 实体别名与冲突
+
+识别用户问题里的实体时，一个别名字面只能归属一个实体。当前数据里别名有两个来源：
+
+- **明日方舟**：`references/char_alias.txt`（每行 `规范名;别名;别名…`，源自
+  `arknights_lore_wiki/data/char_alias.txt`，本意是生成剧情时定位有别名的人物）+ 实体资料里的
+  `aliases` 字段；
+- **终末地**：只有实体资料里的 `aliases` 字段。
+
+冲突按以下顺序收敛，绝不把同一个字面交给两个实体：
+
+1. 字面等于某实体自身规范名时归该实体（「海嗣」归海嗣，不再当作某个人的别名）；
+2. 否则若 `char_alias.txt` 明确声明了归属，按表裁决（「预言家」归博士）；
+3. 两处都没有裁决的冲突（共用称号、同一人物两条实体记录、被误写成别名的类别词）一律不归属任何实体，
+   并在本轮上下文里列出候选规范名，提示直呼实体名或分开检索。
+
+代码只负责"不许双重归属"，不猜谁才是对的。要真正消除第 3 类，需要修数据：合并重复实体记录，
+或从实体 JSON 的 `aliases` 里删掉误写的词。
+
 **可配置项**（`$DSH_HOME/prts-corpus.json`，设置页可视化编辑）：
 
 | 键 | 默认 | 说明 |
@@ -358,7 +387,7 @@ Wiki 资料不是无差别文本池：工具会区分规范角色页、活动/
 | `cloudBaseUrl` | `https://prts.chat` | 云端地址（https，或仅环回主机的 http） |
 | `cloudToken` | 空 | 静态 Bearer token；留空使用匿名 PoW 会话 |
 | `cloudGame` | `all` | `all`（一次检索两款游戏）/ `arknights` / `endfield` |
-| `enabledGames` | `["arknights","endfield"]` | 本地与云端工具允许检索的模块；单模块可只保留其中一项，新会话会装配对应 Skill |
+| `enabledGames` | `["arknights","endfield"]` | 本地与云端工具允许检索的模块，同时决定新下载所需资料包；单模块可只保留其中一项，新会话会装配对应 Skill |
 | `cloudUserId` | 空 | 匿名会话的稳定用户标识 |
 | `cloudTimeoutMs` | `90000` | 云端请求超时（1000–600000） |
 | `cloudMaxResponseBytes` | 32 MiB | 云端响应上限（1 KiB–64 MiB） |
@@ -440,7 +469,11 @@ presets/                 PRTS 模式声明及旧版用户预设初始化
 
 DSH `0.1.2-alpha.1` 与 `0.1.2-alpha.2`（web profile）已完成历史真机验证；其中
 `alpha.1` 通过官方 tag 构建并完成安装、预设解析、宿主启动、设置路由和客户端 bundle
-加载检查。当前插件适配目标为 DSH `0.2.0-rc.1`，共享 Connection Fetch 通道支持
+加载检查。当前声明的最低运行时为 DSH `0.2.0-rc.1`（`engines.dsh`），并已对照
+`0.2.1-rc.2` 源码逐个核对工具注册、systemPrompt 上下文、连接层 Fetch 路由、skill、
+preset 与客户端模块协议：0.2.1 起精确 Fetch 路由只放行 owner 主体，桌面载体需显式
+声明该主体；PTC 子调用也会持久化 `presentationMeta`，但客户端不读取它，因此证据卡
+仍由插件的 receipt 提供。共享 Connection Fetch 通道支持
 Web 与官方 Electron 源码实现；Web 同时保留 HTTP 路由。新版声明式预设与
 旧版文件预设均通过本项目单元测试。针对 `0.1.7-rc.2`（`477b4f420553`），另以真实
 Cordis、Tools、SystemPrompt、Storage Domain 与官方网页工具验证作用域、工具回执、

package/bin/build-mainline-catalog.py

@@ -0,0 +1,195 @@
+#!/usr/bin/env python3
+"""Build evidence-labelled Endfield navigation; standard library only.
+Community explicit lists supply display order, original assets supply classification.
+Same-name joins stay unresolved; ID suffixes never supply mission order.
+"""
+from __future__ import annotations
+import argparse
+from collections import defaultdict
+import json
+from pathlib import Path
+import re
+
+GAME_VERSION = '1.6.3_11010254-4'
+COMMUNITY = 'https://endfield.wiki.gg/wiki/Mission/Main?oldid=63259'
+INK = 'https://endfield.wiki.gg/wiki/Sanctuary_of_Ink?oldid=65256'
+# Verbatim ordered mission titles from the community page's explicit <ol> lists.
+PROCESSES = [
+ (1,0,'N/A',['e0'],'Process: Cold Start|Process: Awakening'),
+ (1,1,'The Broken Lands',['e1'],'Break the Siege|Valley Reboot|Rally and Unite|Westward into the Vale|Secure Sanctuary|Building an Outpost|Through the Haze|Fort Showdown'),
+ (1,2,'The Turbid Heavens',['e1'],'Paving the Way|Work Preparation|Maintenance Progress'),
+ (1,3,'Path of Ascension',['e2'],'The Mission Continues|Cross the Science Park|Park Way Party|Walk in the Park|Mystery in the Building|Down the Hill|Shattered Veins|Lodespring Crossing|Cradle of the Ages|Eve of Battle'),
+ (1,4,'Smoking Embers',['e3','e4'],'Entering the Endgame|Counterstrike Brewing|Scaldings of Utjug|Frontline Rush|Undying Cinders|Temporary Peace|Mysterious Signal|Conference of the Four|Work Preparation'),
+ (2,1,'Lost in the Marsh',['e5'],'Arrival at Wuling|Stumbling Into Qingbo Stockade|Destination Wuling'),
+ (2,2,'The Way of Water',['e6'],'A Glance of Astonishment|Ready the Easterlies|Of Grace and Gentle Might|The Surging Tide|Where Danger Sleeps'),
+ (2,3,'The Long Feud',['e7','c27'],"Unexpected Attack|Deep in the Mire|The Flow Splits Course|Surging From a Withered Spring|A River Restless in Still Winds|Another Wave Comes Crashing In|Ripples Spread Throughout the Flow|Purity Reveals the Stubborn Stone|'Twas One River That Gave Us Drink"),
+ (2,4,'Tales of Old',['e8'],'Like Fire and Ice|A Lethal Invitation|Phantom Wisps|Awaiting Thine Coming|Free Will'),
+ (2,5,'Wrothful Tide',['e9'],'Thus Rise the Wave and Wind|Tremors of the Crashing Tide|Where Friendship Was Made'),
+ (2,6,'Ponderous Load',['e10'],'A Breach Starts with a Crack|Patience Makes Greatness|Forged at the Very Beginning|Kindred Souls Who Share a Vision'),
+ (2,7,'Ruins in the Miasma',['e11'],'Bravely Breach the Hostile Winds|Thus Begins the Homeward March|A Past Shared by the Swordmancers|Same Old Faces in a Place Transformed|Warily Returns the Wandering Vigile|Finally Breaks the Hanging Clouds|"Homecoming"|Stubbornly Lingers the Setting Sun|Freedom of Expansion'),
+ (2,8,'Paint with Fresh Ink',['e12'],None),
+]
+VERSIONS = [('1.0','零号委托'),('1.1','新潮起，故渊离'),('1.2','春晓时'),('1.3','寻遗散记'),('1.4','向渊行'),('1.5','雪凇幽梦'),('1.6','丹青渡')]
+
+
+def load(path: Path):
+    return json.loads(path.read_text(encoding='utf-8'))
+
+
+def normalized(text: str) -> str:
+    return re.sub(r'\s+', ' ', re.sub(r'<[^>]*>', '', text).replace('’', "'")).strip()
+
+
+def build(game_root: Path, update_root: Path) -> dict:
+    table = game_root / 'tables' / GAME_VERSION / 'Table'
+    runtime = game_root / 'json' / GAME_VERSION / 'Data/Json/MissionRuntimeAsset'
+    index_path = update_root / 'wiki_generation/export/task_index.json'
+    tasks = load(index_path)
+    text = load(table / 'TextTable.json')
+    languages = {lang: load(table / f'I18nTextTable_{lang}.json') for lang in ('CN','EN')}
+
+    def localized(key, lang):
+        return normalized(languages[lang].get(str(text.get(key, {}).get('id')), ''))
+
+    def cn_for_en(english):
+        matches = {normalized(languages['CN'].get(k,'')) for k,v in languages['EN'].items() if normalized(v) == normalized(english)}
+        matches.discard('')
+        return next(iter(matches)) if len(matches) == 1 else ''
+
+    metadata = {}
+    by_name = defaultdict(list)
+    # Sort only for stable serialization, never as chronological evidence.
+    for path in sorted(runtime.glob('*.json')):
+        raw = load(path)
+        mid = raw.get('missionId')
+        if not mid or path.stem != mid:
+            continue
+        key = raw.get('missionName',{}).get('key','')
+        exported = tasks.get(mid,{})
+        row = {'id':mid, 'name':localized(key,'CN') or exported.get('name',''),
+               'english_name':localized(key,'EN'), 'mission_type':raw.get('missionType'),
+               'chapter_bitmask':raw.get('missionChapterBitmask'),
+               'char_id':raw.get('charId',''), 'sort_id':raw.get('sortId'),
+               'source':str(path), 'name_key':key,
+               'export_type':exported.get('type'), 'export_series':exported.get('series'),
+               'export_level':exported.get('level'),
+               'has_local_story':bool(exported.get('fragments')) and exported.get('level') != 'skip',
+               'char_linked':bool(raw.get('charId'))}
+        metadata[mid] = row
+        # Assets can reuse a parent's localization key. That proves a name
+        # alias, not a second independently named mission in the display list.
+        row['name_alias_of'] = key[:-5] if key.endswith('_name') and key != mid + '_name' else None
+        if row['english_name'] and row['name_alias_of'] is None:
+            by_name[normalized(row['english_name'])].append(mid)
+
+    chapters = []
+    for ordinal,(chapter,process,english,series,names) in enumerate(PROCESSES,1):
+        missions,unresolved = [],[]
+        if names is not None:
+            for position,name in enumerate(names.split('|'),1):
+                candidates = by_name.get(normalized(name),[])
+                if len(candidates) != 1:
+                    unresolved.append({'order':position,'english_name':name,'candidate_ids':candidates,
+                                       'reason':'ambiguous_name' if candidates else 'name_not_matched'})
+                    continue
+                mid = candidates[0]
+                missions.append({'id':mid,'name':metadata[mid]['name'],'english_name':name,
+                                 'order':position,'order_source':COMMUNITY,
+                                 'has_local_story':metadata[mid]['has_local_story']})
+        else:
+            # Local membership only: no latest-process mission chronology claimed.
+            for mid,row in metadata.items():
+                if row['export_series'] in series and row['has_local_story'] and row['mission_type'] == 0:
+                    missions.append({'id':mid,'name':row['name'],'english_name':row['english_name'],
+                                     'order':None,'order_source':None,'has_local_story':True})
+        summaries = list(dict.fromkeys(t['series_name'] for sid in series for t in tasks.values()
+                         if t.get('series') == sid and t.get('series_name')))
+        title_cn = cn_for_en(english) if english != 'N/A' else ''
+        summary_documents = [name + ' · 剧情总结' for name in summaries]
+        chapter_metadata = [metadata[m['id']] for m in missions]
+        chapters.append({'id':f'chapter{chapter}-process{process}','title':('序章' if process == 0 else title_cn or english),
+            'aliases':list(dict.fromkeys([english] + summaries + ([title_cn] if title_cn else []))),
+            'title_status':'display_alias' if process == 0 else 'localized_community_process_title',
+            'title_en':english,'chapter_number':chapter,
+            'chapter_title':localized(f'chapter{chapter}_name','CN'),
+            'process_number':process,'order':ordinal,
+            'collection_names':list(dict.fromkeys(m['name'] for m in missions if m['has_local_story'] and m['name'])),
+            'summary_title':summary_documents[0] if summary_documents else '', 'summary_titles':summary_documents,
+            'summary_series':series,
+            'mission_metadata':chapter_metadata,
+            'classification':{'basis':'MissionRuntimeAsset.missionType',
+                              'game_mission_types':sorted({r['mission_type'] for r in chapter_metadata if r['mission_type'] is not None}),
+                              'export_types':sorted({r['export_type'] for r in chapter_metadata if r['export_type']}),
+                              'char_linked_mission_ids':[r['id'] for r in chapter_metadata if r['char_linked']]},
+            'collection_order':'community_explicit_order' if names is not None else 'unknown',
+            'summary_coverage_note':'Exported series summaries: one series may cover several processes; one process may use several series.',
+            'version':'1.6' if (chapter,process)==(2,8) else None,
+            'missions':missions,'unresolved_missions':unresolved,
+            'mission_order_status':'community_explicit_order' if names is not None else 'unavailable_membership_only',
+            'classification_source':'MissionRuntimeAsset.missionType; export_type/series are separate ID-pattern grouping metadata',
+            'evidence':[{'kind':'community','url':COMMUNITY if names is not None else INK,
+                         'claim':'chapter/process and explicit mission list order' if names is not None else 'Chapter II Process VIII; local mission membership only, no mission order claimed'}]})
+    versions = []
+    official_bounds = {
+        '1.1': ('https://endfield.gryphline.com/en-us/news/8907', 3),
+        '1.2': ('https://endfield.gryphline.com/en-us/news/0751', 6),
+        '1.4': ('https://endfield.gryphline.com/en-us/news/5200', 7),
+    }
+    introduced_processes = {'1.1':[3], '1.2':[4,5,6], '1.4':[7], '1.6':[8]}
+    for version,title in VERSIONS:
+        evidence = [{'kind':'user_confirmed','claim':'version number and Chinese title'}]
+        status = 'unknown'
+        if version in official_bounds:
+            url, boundary = official_bounds[version]
+            evidence.append({'kind':'official','url':url,
+                             'claim':f'Main Mission available up to Chapter II Process {boundary}; announcement is a cumulative endpoint, not an explicit list of newly introduced processes.'})
+            status = 'official_supported'
+            if version == '1.2':
+                status = 'derived_between_official_boundaries'
+                evidence.append({'kind':'derivation','urls':[official_bounds['1.1'][0],url],
+                                 'claim':'Previous confirmed endpoint III, current endpoint VI: IV-VI are the intervening main-story units; version titles/numbers follow user confirmation.'})
+        elif version == '1.6':
+            status = 'community_supported'
+            evidence.append({'kind':'community','url':INK,'claim':'version 1.6, Chinese 丹青渡, Main Mission up to Chapter II Process VIII'})
+        versions.append({'version':version,'title':title,
+                         'chapter_ids':[f'chapter2-process{n}' for n in introduced_processes.get(version,[])],
+                         'chapter_ids_semantics':'new main-story navigation units, not cumulative availability',
+                         'mapping_status':status,'evidence':evidence})
+    by_id = {chapter['id']:chapter for chapter in chapters}
+    for version in versions:
+        for cid in version['chapter_ids']:
+            by_id[cid]['version'] = version['version']
+            by_id[cid]['version_mapping_status'] = version['mapping_status']
+            by_id[cid]['version_evidence'] = version['evidence']
+    relevant = {m['id'] for c in chapters for m in c['missions']}
+    relevant.update(mid for mid,row in metadata.items() if row['export_series'] in {f'e{i}' for i in range(13)} | {'c27'})
+    return {'schema_version':1,'source_game_version':GAME_VERSION,'versions':versions,'chapters':chapters,
+        'mission_metadata':[metadata[mid] for mid in sorted(relevant)],
+        'ordering_note':'Display order and pre-1.6 mission order cite community explicit tables. Latest mission order is unknown. Serialization, ID suffixes, sortId=0 and fragment order do not establish chronology.',
+        'classification_note':'Original mission_type is independent of export_type/series. c27 has mission_type=0 and char_id=chr_0027_tangtang despite export_type=c. chapter_bitmask is not a chapter ID.',
+        'sources':{'task_index':str(index_path),'mission_runtime_directory':str(runtime),
+                   'text_table':str(table/'TextTable.json'),'i18n_cn':str(table/'I18nTextTable_CN.json'),
+                   'i18n_en':str(table/'I18nTextTable_EN.json'),
+                   'chapter_bridge':str(table/'ChapterMissionChapterTable.json')},
+        'gaps':['No runtime chapter/process membership table located in this snapshot.',
+                'Version-to-process mapping remains unconfirmed for 1.0/1.3/1.5; unknown does not mean no new main story. Official notices give cumulative endpoints, while chapter_ids records new navigation units; 1.2 IV-VI is derived from adjacent confirmed endpoints.',
+                'Work Preparation and Freedom of Expansion preserve same-name ambiguity.',
+                'No cross-mission ordering evidence for e12; five story members have null order.']}
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--game-root',type=Path,default=Path('/home/cloudstack/sda/endfieldgamedata'))
+    parser.add_argument('--update-root',type=Path,default=Path('/home/cloudstack/ai-data/prts.chat/var/data_updates/endfield-1.6.3-11010254-4'))
+    parser.add_argument('--output',type=Path,default=Path(__file__).resolve().parents[1]/'resources/endfield-mainline.json')
+    args = parser.parse_args()
+    result = build(args.game_root,args.update_root)
+    args.output.parent.mkdir(parents=True,exist_ok=True)
+    args.output.write_text(json.dumps(result,ensure_ascii=False,indent=2)+'\n',encoding='utf-8')
+    print(json.dumps({'output':str(args.output),'processes':len(result['chapters']),
+                      'verified_missions':sum(len(c['missions']) for c in result['chapters']),
+                      'unresolved':sum(len(c['unresolved_missions']) for c in result['chapters'])},ensure_ascii=False))
+
+
+if __name__ == '__main__':
+    main()

package/contracts/corpus_i18n_v1.schema.json

@@ -73,14 +73,12 @@
         },
         "data_version": {
           "type": "string",
-          "description": "已有结果的资料版本"
+          "description": "已有结果的资料版本（可省略；续页给出的短版本号即可）"
         },
         "after": {
           "type": "object",
           "additionalProperties": false,
           "required": [
-            "data_version",
-            "request_hash",
             "position",
             "text_offset"
           ],

package/contracts/corpus_tools_v1.schema.json

@@ -67,9 +67,10 @@
           "minLength": 1
         },
         "position": {
+          "description": "不透明续页位置，绑定当前安装范围；请原样复制，补装或切换资料范围后重新搜索。",
           "type": "integer",
           "minimum": 0,
-          "maximum": 10000000
+          "maximum": 9007199254740991
         }
       }
     },
@@ -750,6 +751,34 @@
             "type": "string"
           }
         },
+        "by_game": {
+          "type": "object",
+          "additionalProperties": false,
+          "properties": {
+            "arknights": {
+              "type": "integer"
+            },
+            "endfield": {
+              "type": "integer"
+            }
+          }
+        },
+        "scope": {
+          "type": "object",
+          "additionalProperties": false,
+          "required": [
+            "documents",
+            "lines"
+          ],
+          "properties": {
+            "documents": {
+              "type": "integer"
+            },
+            "lines": {
+              "type": "integer"
+            }
+          }
+        },
         "warnings": {
           "type": "array",
           "items": {

package/docs/endfield-mainline-navigation.md

@@ -0,0 +1,45 @@
+# 终末地主线导航
+
+`corpus_navigate` 是确定性的本地主线目录入口，不调用云端检索，也不生成剧情内容。玩家不必先知道数据包中的系列名或任务名。
+
+```json
+{}
+{"query":"最新主线"}
+{"query":"1.6的主线"}
+{"query":"丹青渡主线"}
+{"query":"第二章第八进程"}
+{"query":"进程VIII"}
+{"query":"第二章第三进程"}
+```
+
+仅支持可选 `query` 和 `game:"endfield"`。省略 query 返回完整目录。目录包括序章和第一、二章共13个进程。返回值绑定当前 `data_version`，每个进程提供多个梗概及其 `corpus_search`/`corpus_read` 参数、任务集合入口、上下进程入口和逐项 `availability`。默认目录文本约2300字符，完整分类和来源元数据保留在结构化结果中。
+
+最新仅指目录中当前本地可定位的最新进程，不宣称实时游戏版本；旧资料包会降到实际覆盖边界，同时保留新进程的未安装说明。缺少正文不会被解释为没有剧情。读取时原样使用返回的 `read` 定位，合集续页使用 `page.continuation`，不推断文件名中的时间先后。
+
+## 三种信息分别取证
+
+- **任务分类**：原始 `MissionRuntimeAsset/{missionId}.json` 的 `missionType`、`charId`、`missionChapterBitmask`、`sortId`、`missionName.key`。导出器生成的 e/c 前缀分类独立保存。唐唐的 c27m1–c27m5 原始 `missionType=0`，属于第二章进程III的主线导航，且有角色关联。角色 Wiki 的文档类型不等于任务分类。
+- **显示目录与任务顺序**：社区显式有序目录提供旧进程的章/进程/任务顺序。英文任务名经 TextTable.id→官方 EN/CN I18n 表精确连接至原始任务ID，不凭英文同名强行选ID。`sortId` 全为0，不能用于排序；`missionChapterBitmask` 不当作章编号。两处 Work Preparation 与一处 Freedom of Expansion 保留候选和未决状态。第二章进程VIII目前只证实五篇正文成员，跨任务顺序 `unknown`，逐任务 `order=null`。
+- **公开版本归属**：版本主题与完整内部构建号分开。1.1进程III、1.4进程VII有官方开放边界；1.2进程IV–VI来自前后官方边界的推导；1.6进程VIII由社区资料支持。1.0、1.3、1.5的新增进程映射保留未知，不表示该版本无新增主线。返回 `mapping_status`、`evidence`，不截取内部构建号推断公开版本。
+
+第二章进程III跨“意外来袭”和“风平水不静”两个生成系列，因此返回两份梗概；第二章进程I的现有生成梗概名为“准备工作”，保留可验证的现有标题。任务名与进程名同名时，“上一段/下一段”不会把任务前后关系偷换成进程前后关系；结果明确 `navigation_level:"process"` 并说明锚点。
+
+## 重建与验证
+
+目录文件为 `resources/endfield-mainline.json`，构建器为 `bin/build-mainline-catalog.py`。构建器仅用Python标准库，按保存的来源快照和固定资料构建，不在运行工具时联网。默认输入为本仓库的最新导出和同机拆包目录，可通过脚本参数指定；先运行 `--help` 查看可配置路径。
+
+```sh
+python3 -B bin/build-mainline-catalog.py
+node --test test/navigation.test.js
+npm test
+```
+
+父仓库还提供 `agent/evals/endfield-mainline.mjs RELEASES_DIR [REPORT]`：用实际 CorpusStore 运行玩家问法，读取所有梗概以及最新进程的五个任务原文，验证返回定位器而非只验证结果非空。
+
+来源：
+
+- https://endfield.wiki.gg/wiki/Mission/Main?oldid=63259
+- https://endfield.wiki.gg/wiki/Sanctuary_of_Ink?oldid=65256
+- https://endfield.gryphline.com/en-us/news/8907
+- https://endfield.gryphline.com/en-us/news/0751
+- https://endfield.gryphline.com/en-us/news/5200