init: edu-content 内容库骨架（L1-L4 四层数据模型 + B类迁移 + lint/CI 工具）

2026-08-22 00:16:49 +08:00
--- a/structured/exams/ket/INDEX.md
+++ b/structured/exams/ket/INDEX.md
@@ -0,0 +1,11 @@
+# exams/ket — 结构化数据（L3）
+
+本目录存放清洗后的可用数据。每条记录须带 `source.source_id`（指向 `sources/`）与 `schema` 版本。
+
+## 消费须知
+- 数据来源与 license：见 `sources/` 与根 `LICENSES.md`
+- 格式 schema：见 `../../schemas/`（qb-1.1 / vocab-1.1）
+- 新增记录请先确认 source 已在 `sources/` 登记
+
+## 当前内容
+（待填充）
--- a/structured/exams/ket/questions/INDEX.md
+++ b/structured/exams/ket/questions/INDEX.md
@@ -0,0 +1,44 @@
+# 题库 INDEX — 数据总览与使用指南
+
+> 更新：2026-08-21 ｜ 格式规范见 [SCHEMA.md](SCHEMA.md)（qb-1.0）
+
+## 一、现有数据
+
+| 文件 | 级别 | 套数 | 空格/小题 | 多答案小题 | 来源 | 许可 |
+|---|---|---|---|---|---|---|
+| `cepoc-a2-key.jsonl` | A2 Key | 6 | 68 | 14 | CEPOC（剑桥 ALTA 官方开源） | `cc-nc-education` |
+| `cepoc-b1-preliminary.jsonl` | B1 Preliminary | 21 | 150 | 24 | CEPOC | `cc-nc-education` |
+| `p2/out/*.jsonl`（6 份） | A2+B1 | 6 卷 | 131 | — | 剑桥官方 2022 样卷 | `official-free` |
+| **合计** | | **33** | **349** | | | |
+
+> P2 详情（成功率 112/112 对账、token 成本、风险点）见 [p2/P2-REPORT.md](p2/P2-REPORT.md)。解析器：pymupdf4llm。
+
+## 二、待入库（已规划）
+
+| 优先级 | 数据 | 预估规模 | 许可 |
+|---|---|---|---|
+| ~~P3~~ | ~~官方词表 → 词库~~ **已完成** → 见 [../vocab/INDEX.md](../vocab/INDEX.md)（A2 1740 + B1 3172 词条） | — | `official-free` |
+| ~~P3.5~~ | ~~词库中文释义+音标~~ **已完成** → 见 [../vocab/INDEX.md](../vocab/INDEX.md) §三·五（ECDICT 4804 + LLM 98，音标覆盖 93.3%） | 4902 条 | ECDICT `CC-BY-SA-4.0` / LLM 生成 |
+| P4 | tape script + mp3 精听语料 | 2 套对齐文本 | `official-free` |
+| P5 | SkoolPlus 教材（ECS 下载后） | 20+ 套模拟题 | `restricted`（仅本地） |
+
+## 三、下游 agent 使用须知
+
+1. **读库**：JSONL 逐行 `json.loads`；一条记录 = 一个任务（语篇+多小题），不是单题
+2. **判题**：必须按 SCHEMA.md 第二节规则——`strip` + 大小写不敏感 + `answers_accept` 集合匹配；38 个空有多个可接受答案（如 `["because","as","since"]`），字符串全等会误判
+3. **许可红线**（按 `license` 字段过滤）：
+   - `official-free` → 可进部署的英语门户
+   - `cc-nc-education`（CEPOC 全部数据）→ 自用/练习工具可以，**禁止商业化与再分发**
+   - `restricted` → 任何对外输出都不允许出现
+4. **完形上下文切分**：`passage.split(blank_token)`，段数 = 空数 + 1
+5. **CEPOC 的 `part` 为 null**：数据混有 2020 改版前真题（KET 11-12 空/套、PET 6-8 空/套），与新版题型空数不符，见记录内 `note` 字段
+6. **P2 记录的媒体引用**：`meta.media[].path` 与 `meta.audio` 为相对路径（前者相对 question-bank 目录，后者相对 english-site 目录）；图片选项题（听力 Part 1）`options=null`，必须渲染 `meta.media` 图片作答
+
+## 四、复现
+
+```bash
+cd data/question-bank
+python3 scripts/convert_cepoc.py <cepoc-main 绝对路径> .
+```
+
+转换脚本内置校验：空格数与小题数一致、题号不重复、答案非空、id 唯一。2026-08-21 校验全部通过。
--- a/structured/exams/ket/questions/cepoc-a2-key.jsonl
+++ b/structured/exams/ket/questions/cepoc-a2-key.jsonl
@@ -0,0 +1,6 @@
+{"schema": "qb-1.1", "id": "a2k-cepoc-0001", "exam": "A2 Key", "cefr": "A2", "paper": "reading_writing", "part": null, "type": "open_cloze", "task_prompt": "For each gap, write ONE word which best completes the text.", "title": null, "passage": "Dear Melissa\n\nHow are things _ London?  I had a great time staying with you.  Thank you for looking after _ .  You are _ great cook and your family were really kind to me.  And my English is much better now _ it was before.  \n\nMy flight home was fine – there were _ delays at all.  It’s good to be back, but I miss you all very _ !\n\n_ you remember my friend Jenny?  I think she came _ visit me at your house once _ twice.  Well, she took some great photos _ us on our last day at the language school.  I’m sending you one with this email.  I hope you like \n\n_ .\n\nRosario", "blank_token": "_", "items": [{"no": 1, "stem": null, "options": null, "answers_accept": ["in"], "answer_official": "in", "explain": null, "tags": []}, {"no": 2, "stem": null, "options": null, "answers_accept": ["me"], "answer_official": "me", "explain": null, "tags": []}, {"no": 3, "stem": null, "options": null, "answers_accept": ["a"], "answer_official": "a", "explain": null, "tags": []}, {"no": 4, "stem": null, "options": null, "answers_accept": ["than"], "answer_official": "than", "explain": null, "tags": []}, {"no": 5, "stem": null, "options": null, "answers_accept": ["no"], "answer_official": "no", "explain": null, "tags": []}, {"no": 6, "stem": null, "options": null, "answers_accept": ["much"], "answer_official": "much", "explain": null, "tags": []}, {"no": 7, "stem": null, "options": null, "answers_accept": ["Do"], "answer_official": "Do", "explain": null, "tags": []}, {"no": 8, "stem": null, "options": null, "answers_accept": ["to"], "answer_official": "to", "explain": null, "tags": []}, {"no": 9, "stem": null, "options": null, "answers_accept": ["or"], "answer_official": "or", "explain": null, "tags": []}, {"no": 10, "stem": null, "options": null, "answers_accept": ["of"], "answer_official": "of", "explain": null, "tags": []}, {"no": 11, "stem": null, "options": null, "answers_accept": ["it"], "answer_official": "it", "explain": null, "tags": []}], "source": {"origin": "cepoc", "file": "assets/open-source/cepoc-main/KET.json", "orig_id": 1, "url": "https://github.com/CambridgeALTA/cepoc", "retrieved": "2026-08-21", "source_id": "cepoc", "source_item_id": "cepoc-ket-json"}, "license": "cc-nc-education", "tags": ["cloze", "open_cloze"], "note": "CEPOC 未标注套题年份与卷内 Part 号；空格数与 2020 版题型不完全对应（KET 11-12空/套、PET 6-8空/套），推断混有 2020 改版前真题。完形题型本身对应 A2 Key Reading & Writing / B1 Preliminary Reading 的 open cloze 题型。", "meta": {"multi_answer_items": 0}}
+{"schema": "qb-1.1", "id": "a2k-cepoc-0002", "exam": "A2 Key", "cefr": "A2", "paper": "reading_writing", "part": null, "type": "open_cloze", "task_prompt": "For each gap, write ONE word which best completes the text.", "title": null, "passage": "Dear Aunt Fabienne,\n\nMum asked _ to let you know that I arrived home okay. I enjoyed travelling back _ the train, and I was surprised that _ only took me three hours to get home. Thank you very _ for looking after me _ I was in France. \n\nI’m going back to school _ week. My French is a lot better _ it was before, so I hope my marks will improve!\n\nI _ like to keep practising French and I’m looking _ a pen friend. I thought of your friend’s son Elian. We could email each _ once a week. Let me know _ you think.\n\nI hope _ see you soon. \n\nDaniel", "blank_token": "_", "items": [{"no": 1, "stem": null, "options": null, "answers_accept": ["me"], "answer_official": "me", "explain": null, "tags": []}, {"no": 2, "stem": null, "options": null, "answers_accept": ["on"], "answer_official": "on", "explain": null, "tags": []}, {"no": 3, "stem": null, "options": null, "answers_accept": ["it"], "answer_official": "it", "explain": null, "tags": []}, {"no": 4, "stem": null, "options": null, "answers_accept": ["much"], "answer_official": "much", "explain": null, "tags": []}, {"no": 5, "stem": null, "options": null, "answers_accept": ["when", "while"], "answer_official": "when", "explain": null, "tags": []}, {"no": 6, "stem": null, "options": null, "answers_accept": ["next", "this"], "answer_official": "next", "explain": null, "tags": []}, {"no": 7, "stem": null, "options": null, "answers_accept": ["than"], "answer_official": "than", "explain": null, "tags": []}, {"no": 8, "stem": null, "options": null, "answers_accept": ["would", "'d"], "answer_official": "would", "explain": null, "tags": []}, {"no": 9, "stem": null, "options": null, "answers_accept": ["for"], "answer_official": "for", "explain": null, "tags": []}, {"no": 10, "stem": null, "options": null, "answers_accept": ["other"], "answer_official": "other", "explain": null, "tags": []}, {"no": 11, "stem": null, "options": null, "answers_accept": ["what"], "answer_official": "what", "explain": null, "tags": []}, {"no": 12, "stem": null, "options": null, "answers_accept": ["to", "I'll", "we'll"], "answer_official": "to", "explain": null, "tags": []}], "source": {"origin": "cepoc", "file": "assets/open-source/cepoc-main/KET.json", "orig_id": 2, "url": "https://github.com/CambridgeALTA/cepoc", "retrieved": "2026-08-21", "source_id": "cepoc", "source_item_id": "cepoc-ket-json"}, "license": "cc-nc-education", "tags": ["cloze", "open_cloze"], "note": "CEPOC 未标注套题年份与卷内 Part 号；空格数与 2020 版题型不完全对应（KET 11-12空/套、PET 6-8空/套），推断混有 2020 改版前真题。完形题型本身对应 A2 Key Reading & Writing / B1 Preliminary Reading 的 open cloze 题型。", "meta": {"multi_answer_items": 4}}
+{"schema": "qb-1.1", "id": "a2k-cepoc-0003", "exam": "A2 Key", "cefr": "A2", "paper": "reading_writing", "part": null, "type": "open_cloze", "task_prompt": "For each gap, write ONE word which best completes the text.", "title": null, "passage": "Hi Marta,\n\nI hope _ are well.  I was trying _ call you yesterday but you didn’t answer.  Is your phone broken, _ have you lost it again? \n\nAnyway, I wanted to ask _ you can come to my birthday party next Saturday.  My birthday was actually two weeks _ , but I couldn’t have the party then _ lots of people were _ holiday.  \n\nI’ve invited _ few people from our swimming club, as well as my friends from school.  _ you think your brother _ free that day?  Please invite _ to come too.  The party’s at my house and starts at 8p.m.  \n\nParveen", "blank_token": "_", "items": [{"no": 1, "stem": null, "options": null, "answers_accept": ["you"], "answer_official": "you", "explain": null, "tags": []}, {"no": 2, "stem": null, "options": null, "answers_accept": ["to"], "answer_official": "to", "explain": null, "tags": []}, {"no": 3, "stem": null, "options": null, "answers_accept": ["or"], "answer_official": "or", "explain": null, "tags": []}, {"no": 4, "stem": null, "options": null, "answers_accept": ["if", "whether"], "answer_official": "if", "explain": null, "tags": []}, {"no": 5, "stem": null, "options": null, "answers_accept": ["ago"], "answer_official": "ago", "explain": null, "tags": []}, {"no": 6, "stem": null, "options": null, "answers_accept": ["because", "as"], "answer_official": "because", "explain": null, "tags": []}, {"no": 7, "stem": null, "options": null, "answers_accept": ["on"], "answer_official": "on", "explain": null, "tags": []}, {"no": 8, "stem": null, "options": null, "answers_accept": ["a"], "answer_official": "a", "explain": null, "tags": []}, {"no": 9, "stem": null, "options": null, "answers_accept": ["Do"], "answer_official": "Do", "explain": null, "tags": []}, {"no": 10, "stem": null, "options": null, "answers_accept": ["is"], "answer_official": "is", "explain": null, "tags": []}, {"no": 11, "stem": null, "options": null, "answers_accept": ["him"], "answer_official": "him", "explain": null, "tags": []}], "source": {"origin": "cepoc", "file": "assets/open-source/cepoc-main/KET.json", "orig_id": 3, "url": "https://github.com/CambridgeALTA/cepoc", "retrieved": "2026-08-21", "source_id": "cepoc", "source_item_id": "cepoc-ket-json"}, "license": "cc-nc-education", "tags": ["cloze", "open_cloze"], "note": "CEPOC 未标注套题年份与卷内 Part 号；空格数与 2020 版题型不完全对应（KET 11-12空/套、PET 6-8空/套），推断混有 2020 改版前真题。完形题型本身对应 A2 Key Reading & Writing / B1 Preliminary Reading 的 open cloze 题型。", "meta": {"multi_answer_items": 2}}
+{"schema": "qb-1.1", "id": "a2k-cepoc-0004", "exam": "A2 Key", "cefr": "A2", "paper": "reading_writing", "part": null, "type": "open_cloze", "task_prompt": "For each gap, write ONE word which best completes the text.", "title": null, "passage": "Hi Marcia, \n\nDo you remember I wanted _ invite you, Sasha and John to my new house?  Well, _ about next Saturday evening?  We _ a beautiful garden with lots _ trees.  We may have a barbeque outside _ it doesn’t rain.  My mother is _ really good cook and I _ going to help her make the food.  \n\nAfterwards, _ don’t we watch a film?  _ you bring a DVD with you as I haven’t got many here?  \n\nJohn’s coming _ bike and Sasha is getting a lift _ her father.  I hope you’ll be able to come too.  \n\nLet me know soon!\n\nConny", "blank_token": "_", "items": [{"no": 1, "stem": null, "options": null, "answers_accept": ["to"], "answer_official": "to", "explain": null, "tags": []}, {"no": 2, "stem": null, "options": null, "answers_accept": ["what", "how"], "answer_official": "what", "explain": null, "tags": []}, {"no": 3, "stem": null, "options": null, "answers_accept": ["have"], "answer_official": "have", "explain": null, "tags": []}, {"no": 4, "stem": null, "options": null, "answers_accept": ["of"], "answer_official": "of", "explain": null, "tags": []}, {"no": 5, "stem": null, "options": null, "answers_accept": ["if"], "answer_official": "if", "explain": null, "tags": []}, {"no": 6, "stem": null, "options": null, "answers_accept": ["a"], "answer_official": "a", "explain": null, "tags": []}, {"no": 7, "stem": null, "options": null, "answers_accept": ["am", "'m"], "answer_official": "am", "explain": null, "tags": []}, {"no": 8, "stem": null, "options": null, "answers_accept": ["why"], "answer_official": "why", "explain": null, "tags": []}, {"no": 9, "stem": null, "options": null, "answers_accept": ["Can", "Could"], "answer_official": "Can", "explain": null, "tags": []}, {"no": 10, "stem": null, "options": null, "answers_accept": ["by"], "answer_official": "by", "explain": null, "tags": []}, {"no": 11, "stem": null, "options": null, "answers_accept": ["from", "with"], "answer_official": "from", "explain": null, "tags": []}], "source": {"origin": "cepoc", "file": "assets/open-source/cepoc-main/KET.json", "orig_id": 4, "url": "https://github.com/CambridgeALTA/cepoc", "retrieved": "2026-08-21", "source_id": "cepoc", "source_item_id": "cepoc-ket-json"}, "license": "cc-nc-education", "tags": ["cloze", "open_cloze"], "note": "CEPOC 未标注套题年份与卷内 Part 号；空格数与 2020 版题型不完全对应（KET 11-12空/套、PET 6-8空/套），推断混有 2020 改版前真题。完形题型本身对应 A2 Key Reading & Writing / B1 Preliminary Reading 的 open cloze 题型。", "meta": {"multi_answer_items": 4}}
+{"schema": "qb-1.1", "id": "a2k-cepoc-0005", "exam": "A2 Key", "cefr": "A2", "paper": "reading_writing", "part": null, "type": "open_cloze", "task_prompt": "For each gap, write ONE word which best completes the text.", "title": null, "passage": "How _ you?  I’m writing to you _ I need your help.\n\nI’m starting a business course at college _ term.  The teacher has asked us to do some office work during _ summer.  Your Dad’s _ manager, isn’t he?  Can you ask him _ I can work with him?  I only need to work _ a few weeks, and he will _ have to pay me anything.\n\nAlso, can I stay _ you while I’m working?  Mum says it’s much _ far to go by bike _ day.  Tell your mum that I can make _ own meals, so she won’t have to cook for me.\n\nThanks.", "blank_token": "_", "items": [{"no": 1, "stem": null, "options": null, "answers_accept": ["are"], "answer_official": "are", "explain": null, "tags": []}, {"no": 2, "stem": null, "options": null, "answers_accept": ["because", "as", "since"], "answer_official": "because", "explain": null, "tags": []}, {"no": 3, "stem": null, "options": null, "answers_accept": ["this", "next"], "answer_official": "this", "explain": null, "tags": []}, {"no": 4, "stem": null, "options": null, "answers_accept": ["the"], "answer_official": "the", "explain": null, "tags": []}, {"no": 5, "stem": null, "options": null, "answers_accept": ["a"], "answer_official": "a", "explain": null, "tags": []}, {"no": 6, "stem": null, "options": null, "answers_accept": ["if"], "answer_official": "if", "explain": null, "tags": []}, {"no": 7, "stem": null, "options": null, "answers_accept": ["for"], "answer_official": "for", "explain": null, "tags": []}, {"no": 8, "stem": null, "options": null, "answers_accept": ["not"], "answer_official": "not", "explain": null, "tags": []}, {"no": 9, "stem": null, "options": null, "answers_accept": ["with"], "answer_official": "with", "explain": null, "tags": []}, {"no": 10, "stem": null, "options": null, "answers_accept": ["too"], "answer_official": "too", "explain": null, "tags": []}, {"no": 11, "stem": null, "options": null, "answers_accept": ["each", "every"], "answer_official": "each", "explain": null, "tags": []}, {"no": 12, "stem": null, "options": null, "answers_accept": ["my"], "answer_official": "my", "explain": null, "tags": []}], "source": {"origin": "cepoc", "file": "assets/open-source/cepoc-main/KET.json", "orig_id": 5, "url": "https://github.com/CambridgeALTA/cepoc", "retrieved": "2026-08-21", "source_id": "cepoc", "source_item_id": "cepoc-ket-json"}, "license": "cc-nc-education", "tags": ["cloze", "open_cloze"], "note": "CEPOC 未标注套题年份与卷内 Part 号；空格数与 2020 版题型不完全对应（KET 11-12空/套、PET 6-8空/套），推断混有 2020 改版前真题。完形题型本身对应 A2 Key Reading & Writing / B1 Preliminary Reading 的 open cloze 题型。", "meta": {"multi_answer_items": 3}}
+{"schema": "qb-1.1", "id": "a2k-cepoc-0006", "exam": "A2 Key", "cefr": "A2", "paper": "reading_writing", "part": null, "type": "open_cloze", "task_prompt": "For each gap, write ONE word which best completes the text.", "title": null, "passage": "Welcome _ my blog! My name’s Lauren and I’m 22 years old. I come from\n\nIllinois in the USA, but _ the moment I’m studying in Florence, in Italy.\n\nI’m living _ a friend of mine from university. Our apartment _ in the centre of town, so we don’t _ to use the buses – we can walk everywhere.  I have classes most days, but in _ afternoon, I explore the city or _ swimming. When _ is sunny I visit the Boboli Gardens. _ are beautiful views of Florence from here.\n\nI _ you enjoy my blog. _  you have any questions, please leave me a message.", "blank_token": "_", "items": [{"no": 1, "stem": null, "options": null, "answers_accept": ["to"], "answer_official": "to", "explain": null, "tags": []}, {"no": 2, "stem": null, "options": null, "answers_accept": ["at"], "answer_official": "at", "explain": null, "tags": []}, {"no": 3, "stem": null, "options": null, "answers_accept": ["with"], "answer_official": "with", "explain": null, "tags": []}, {"no": 4, "stem": null, "options": null, "answers_accept": ["is"], "answer_official": "is", "explain": null, "tags": []}, {"no": 5, "stem": null, "options": null, "answers_accept": ["have", "need"], "answer_official": "have", "explain": null, "tags": []}, {"no": 6, "stem": null, "options": null, "answers_accept": ["the"], "answer_official": "the", "explain": null, "tags": []}, {"no": 7, "stem": null, "options": null, "answers_accept": ["go"], "answer_official": "go", "explain": null, "tags": []}, {"no": 8, "stem": null, "options": null, "answers_accept": ["it"], "answer_official": "it", "explain": null, "tags": []}, {"no": 9, "stem": null, "options": null, "answers_accept": ["There"], "answer_official": "There", "explain": null, "tags": []}, {"no": 10, "stem": null, "options": null, "answers_accept": ["hope"], "answer_official": "hope", "explain": null, "tags": []}, {"no": 11, "stem": null, "options": null, "answers_accept": ["If"], "answer_official": "If", "explain": null, "tags": []}], "source": {"origin": "cepoc", "file": "assets/open-source/cepoc-main/KET.json", "orig_id": 6, "url": "https://github.com/CambridgeALTA/cepoc", "retrieved": "2026-08-21", "source_id": "cepoc", "source_item_id": "cepoc-ket-json"}, "license": "cc-nc-education", "tags": ["cloze", "open_cloze"], "note": "CEPOC 未标注套题年份与卷内 Part 号；空格数与 2020 版题型不完全对应（KET 11-12空/套、PET 6-8空/套），推断混有 2020 改版前真题。完形题型本身对应 A2 Key Reading & Writing / B1 Preliminary Reading 的 open cloze 题型。", "meta": {"multi_answer_items": 1}}
--- a/structured/exams/ket/vocab/INDEX.md
+++ b/structured/exams/ket/vocab/INDEX.md
@@ -0,0 +1,101 @@
+# 词库 INDEX — 数据总览与使用指南
+
+> 更新：2026-08-21 ｜ 解析器：`scripts/parse_vocab.py`（pymupdf，纯规则解析，零 LLM 成本）
+> 对应计划：题库路线 P3（官方词表 → 结构化词库）+ P3.5（词库增补：中文释义 + IPA 音标）
+> 增补管线：`scripts/enrich_gloss.py`（ECDICT 匹配）+ `scripts/make_llm.py`（LLM 增补）+ `scripts/finalize_vocab.py`（音标格式统一 + 补常用词音标）
+
+## 一、现有数据
+
+| 文件 | 级别 | 主表词条 | 仅主题表补充词 | 合计 | 其中 B1 新增 | 来源 | 许可 |
+|---|---|---|---|---|---|---|---|
+| `a2-key-vocab.jsonl` | A2 Key / for Schools | 1727 | 13 | **1740** | — | 剑桥官方词汇表（2025-08 版） | `official-free` |
+| `b1-preliminary-vocab.jsonl` | B1 Preliminary / for Schools | 3055 | 117 | **3172** | 1517（`new_at_b1=true`） | 剑桥官方词汇表（2020 版） | `official-free` |
+| `word-sets.json` | A2+B1 通用 | — | — | 6 个词集 | — | Appendix 1（规则生成） | `official-free` |
+
+主题（topic）覆盖：A2 25 个 / B1 23 个（Clothes、Education、Food and Drink、Travel and Transport、Work and Jobs 等）。
+
+## 二、记录结构（每行一个词条）
+
+```json
+{
+  "id": "a2k-vocab-0091",
+  "level": "A2",
+  "headword": "attractive",            // 原样保留, 含变体斜杠 "café/cafe"、缩写 "(kg)"
+  "pos": ["adj"],                      // 词性数组; 多词词性整体保留: "phr v"/"prep phr"/"aux v"
+  "pos_notes": null,                   // 用法/地区注记: ["+ ing"], ["(Br Eng) (Am Eng: bathing suit)"], ["unc (uncountable)"]
+  "examples": [{"text": "The walls are an attractive colour.", "pos": null}],
+  "phrases": null,                     // 子短语/搭配: "(be) bad for somebody", "bright pink"
+  "topics": ["Clothes and Accessories"],   // 来自 Appendix 2 主题表
+  "origin": "main_list",               // main_list=字母序主表 / topic_list=仅出现在主题表
+  "no_pos": true,                      // 仅少数: PDF 原文无词性标注 (如 "film star")
+  "new_at_b1": false,                  // 仅 B1 文件: 该词不在 A2 词表 = B1 新增
+  "source": {"file": "...", "url": "...", "publisher": "Cambridge English / UCLES", "licence": "official-free"},
+  "gloss": {                           // P3.5 增补: 中文释义 + IPA 音标
+    "zh": "n. 苹果",                    // 带词性前缀的中文释义
+    "phonetic": "/ˈæpl/",              // 英式 IPA, 统一包裹为 /.../
+    "source": "ecdict",                // ★ 来源标注: "ecdict"=ECDICT 开源词典 / "llm"=LLM 生成
+    "licence": "CC-BY-SA-4.0",         // ecdict 对应 CC-BY-SA-4.0; llm 对应 llm-generated
+    "phonetic_source": "llm"           // 可选: 仅当音标由 LLM 补、释义来自 ECDICT 时出现
+  }
+}
+```
+
+## 三、下游 agent 使用须知
+
+1. **背单词优先序**（推荐）：A2 全部 → B1 中 `new_at_b1=true` 的 1517 词 → topic_list 补充词。两级共约 3257 去重词条。
+2. **例句仍是稀缺资源**：官方词表仅约 11% 词条带例句（A2 170 / B1 218 条），主要标注多义词的分用法。**中文释义与音标已通过 P3.5 增补进 `gloss` 字段**（见 §三·五），但例句本身未生成。
+3. **主题背诵**：`topics` 字段支持按场景刷词（如先攻 Food and Drink）；同一词可属多个主题。
+4. **词性坑**：`be` 的词性是 `["aux v","v"]`（PDF 原文缩写 "av"，按剑桥缩写表解释为助动词）；不可数名词注记在 `pos_notes`（如 rubbish）。
+5. **topic_only 词**（A2 13 / B1 117）：官方明确为"考生还应知道的补充词"（如 get dressed、climate change），无例句无主表词性，但 `pos` 可能来自主题表标注。
+6. **变体词头**：`trainers`（topic）与 `trainer`（主表）已按复数回退匹配；但 A2 主表无 trainer，故 A2 里 trainers 仍是独立补充词。
+7. **word-sets.json**：基数/序数/星期/月份/季节/大洲为确定性生成集；国家·语言·民族开放集合未枚举（官方用 etc.），需要时另行补表。
+8. **许可**：全部 `official-free`，可进部署门户；但 PDF 内的 "Additions in green"（2025 新增词标记）在文本层无颜色信息，未能结构化。
+
+## 三·五、P3.5 增补说明（中文释义 + IPA 音标）
+
+官方 PDF 无中文释义、无音标，本步补齐 `gloss` 字段（每条记录已写入，无遗漏）。
+
+**来源标注（用户要求：区分「原词条」与「LLM 增补」）**
+- `gloss.source = "ecdict"`：中文释义来自 **ECDICT** 开源英汉词典（CC-BY-SA-4.0），词典级数据，覆盖约 98%。
+- `gloss.source = "llm"`：ECDICT 未命中的残余词头（多为短语/复合词/剑桥多义分项，如 `deal with`、`smart (clever)`、`social media`），由 **LLM 生成**中文释义 + 音标（`licence: llm-generated`）。共 98 条记录（83 个去重词头）。
+- `gloss.phonetic_source = "llm"`（可选）：仅 97 条「释义来自 ECDICT、但 ECDICT 原条目缺音标」的常用单词，音标由 LLM 补、释义 provenance 不变。
+
+**覆盖统计（4902 条）**
+| 指标 | 数值 |
+|---|---|
+| `gloss` 覆盖率 | 100%（4902/4902） |
+| `source=ecdict` | 4804 条 |
+| `source=llm` | 98 条 |
+| 有音标 | 4576 条（93.3%） |
+| 无音标 | 326 条（6.7%，**全部为短语/多词**，如 `a few`、`alarm clock`） |
+
+**为什么短语留空音标**：学习者已掌握组成单词的发音，短语整体音标优先级低；若后续需要可单独补（g2p 模型或 LLM）。
+
+**管线与复现**
+```bash
+cd english-site
+python scripts/enrich_gloss.py match   # ECDICT 匹配 -> _gloss_cache/ecdict.json
+python scripts/make_llm.py             # LLM 增补 83 词 -> _gloss_cache/llm.json
+python scripts/enrich_gloss.py merge   # 合并写回 *.jsonl
+python scripts/finalize_vocab.py       # 音标统一 /.../ + 补 71 常用单词音标
+```
+> ECDICT 原始数据：`assets/vocab/ecdict/`（StarDict 格式，67MB，从 github release 经 gh-proxy 镜像下载）。清洗规则见 `enrich_gloss.py: clean_entry`（剥离词频/考试标签、西里尔 ә→ə、ASCII 撇号→IPA 重音符等）。
+
+## 四、已知局限（诚实披露）
+
+- ~~官方 PDF 无中文释义、无音标~~ → 已在 P3.5 通过 ECDICT + LLM 增补进 `gloss` 字段（见 §三·五）；仅 326 条短语（6.7%）因 ECDICT 无音标而留空，释义仍在。
+- `film star`、`zebra` 等 2 个 B1 词条原文无词性（`no_pos=true`），非解析丢失。
+- `driving/driver's licence` 等极少数跨行词头的地区变体注记可能不完整（已抽查主流 5 例均正确）。
+- B1 词表基于 2018 © 版本（2020 起生效），A2 基于 2025-08 版，两版之间 A2 的新增词（若 B1 版本更旧）不影响 `new_at_b1` 的判断方向，但个别词的归属可能滞后。
+
+## 五、复现
+
+```bash
+cd english-site
+python scripts/parse_vocab.py    # 输出到 data/vocab/, 隔离区 0 条, id 唯一校验通过
+```
+
+## 六、素材勘误（相对调研报告）
+
+- `assets/vocab/b1-preliminary-vocabulary-booklet.pdf`（8MB）**不是词表**，是《B1 for Schools 词汇教学活动手册》（教师用），保留作教学设计参考。
+- B1 官方词表直链 `cambridgeenglish.org/Images/506887-*.pdf` 已 404，改用 **cambridgeenglish.cn 域名**下载成功（文件已存 `assets/vocab/b1-preliminary-2020-vocabulary-list.pdf`）。
--- a/structured/exams/ket/vocab/a2-key-vocab.jsonl
+++ b/structured/exams/ket/vocab/a2-key-vocab.jsonl
--- a/structured/exams/ket/vocab/word-sets.json
+++ b/structured/exams/ket/vocab/word-sets.json
@@ -0,0 +1,201 @@
+{
+  "note": "Appendix 1 Word sets 为开放集合(官方原文用 etc.), 此处按规则生成确定词集。国家/语言/民族名开放集合未枚举, 需另行补表。",
+  "levels": [
+    "A2",
+    "B1"
+  ],
+  "sets": [
+    {
+      "name": "Cardinal numbers (one to one hundred + hundred/thousand)",
+      "words": [
+        "eight",
+        "eighteen",
+        "eighty",
+        "eighty-eight",
+        "eighty-five",
+        "eighty-four",
+        "eighty-nine",
+        "eighty-one",
+        "eighty-seven",
+        "eighty-six",
+        "eighty-three",
+        "eighty-two",
+        "eleven",
+        "fifteen",
+        "fifty",
+        "fifty-eight",
+        "fifty-five",
+        "fifty-four",
+        "fifty-nine",
+        "fifty-one",
+        "fifty-seven",
+        "fifty-six",
+        "fifty-three",
+        "fifty-two",
+        "five",
+        "forty",
+        "forty-eight",
+        "forty-five",
+        "forty-four",
+        "forty-nine",
+        "forty-one",
+        "forty-seven",
+        "forty-six",
+        "forty-three",
+        "forty-two",
+        "four",
+        "fourteen",
+        "hundred",
+        "nine",
+        "nineteen",
+        "ninety",
+        "ninety-eight",
+        "ninety-five",
+        "ninety-four",
+        "ninety-nine",
+        "ninety-one",
+        "ninety-seven",
+        "ninety-six",
+        "ninety-three",
+        "ninety-two",
+        "one",
+        "seven",
+        "seventeen",
+        "seventy",
+        "seventy-eight",
+        "seventy-five",
+        "seventy-four",
+        "seventy-nine",
+        "seventy-one",
+        "seventy-seven",
+        "seventy-six",
+        "seventy-three",
+        "seventy-two",
+        "six",
+        "sixteen",
+        "sixty",
+        "sixty-eight",
+        "sixty-five",
+        "sixty-four",
+        "sixty-nine",
+        "sixty-one",
+        "sixty-seven",
+        "sixty-six",
+        "sixty-three",
+        "sixty-two",
+        "ten",
+        "thirteen",
+        "thirty",
+        "thirty-eight",
+        "thirty-five",
+        "thirty-four",
+        "thirty-nine",
+        "thirty-one",
+        "thirty-seven",
+        "thirty-six",
+        "thirty-three",
+        "thirty-two",
+        "thousand",
+        "three",
+        "twelve",
+        "twenty",
+        "twenty-eight",
+        "twenty-five",
+        "twenty-four",
+        "twenty-nine",
+        "twenty-one",
+        "twenty-seven",
+        "twenty-six",
+        "twenty-three",
+        "twenty-two",
+        "two"
+      ]
+    },
+    {
+      "name": "Ordinal numbers (first to thirty-first)",
+      "words": [
+        "first",
+        "second",
+        "third",
+        "fourth",
+        "fifth",
+        "sixth",
+        "seventh",
+        "eighth",
+        "ninth",
+        "tenth",
+        "eleventh",
+        "twelfth",
+        "thirteenth",
+        "fourteenth",
+        "fifteenth",
+        "sixteenth",
+        "seventeenth",
+        "eighteenth",
+        "nineteenth",
+        "twentyth",
+        "twenty-first",
+        "twenty-second",
+        "twenty-third",
+        "twenty-fourth",
+        "twenty-fifth",
+        "twenty-sixth",
+        "twenty-seventh",
+        "twenty-eighth",
+        "twenty-ninth",
+        "thirtieth",
+        "thirty-first"
+      ]
+    },
+    {
+      "name": "Days of the week",
+      "words": [
+        "Monday",
+        "Tuesday",
+        "Wednesday",
+        "Thursday",
+        "Friday",
+        "Saturday",
+        "Sunday"
+      ]
+    },
+    {
+      "name": "Months of the year",
+      "words": [
+        "January",
+        "February",
+        "March",
+        "April",
+        "May",
+        "June",
+        "July",
+        "August",
+        "September",
+        "October",
+        "November",
+        "December"
+      ]
+    },
+    {
+      "name": "Seasons of the year",
+      "words": [
+        "spring",
+        "summer",
+        "autumn",
+        "winter"
+      ]
+    },
+    {
+      "name": "Continents",
+      "words": [
+        "Africa",
+        "Antarctica",
+        "Asia",
+        "Australia",
+        "Europe",
+        "North America",
+        "South America"
+      ]
+    }
+  ]
+}