dsh-cait 1.1.3 → 1.1.6

package/CHANGELOG.md

@@ -4,6 +4,145 @@ All notable changes to CAIT are documented here. The project follows
 [Semantic Versioning](https://semver.org/) and the format is based on
 [Keep a Changelog](https://keepachangelog.com/).
 
+## [1.1.6] - 2026-11-09
+
+Review-workbench release: find and replace, a command palette, segment notes,
+draft undo/redo, segment merge/split/lock, proper-name term discovery, and an
+optional AI review pass that suggests fixes without writing to the document.
+
+### Added
+
+- **Find and replace.** Preview every match in source/target, apply in one step,
+  and undo the last replacement. Confirmed segments are downgraded to review when
+  their target changes; TM entries are left untouched.
+- **Command palette and shortcuts.** `Ctrl/Cmd+K` opens a command palette;
+  `Ctrl/Cmd+F` opens find and replace.
+- **Segment notes.** Per-segment review comments with add, edit, and delete,
+  persisted in the document JSON.
+- **Draft undo/redo.** Per-segment editing history with `Ctrl/Cmd+Z`,
+  `Ctrl/Cmd+Shift+Z`, and `Ctrl/Cmd+Y`.
+- **Merge, split, and lock.** Merge adjacent text segments, split a segment at a
+  source offset, and lock segments against edits, confirmation, replacement,
+  pre-translation, propagation, and clearing.
+- **Term discovery for one-off names.** Candidates now include quoted terms,
+  capitalized names, and Chinese proper names that occur once, in addition to
+  repeated n-grams. Candidates carry a type, confidence, and evidence; they are
+  never added to the glossary automatically.
+- **AI review.** `reviewDocument` asks the configured DSH route for structured
+  issues (`severity`, `code`, `message`, `suggestion`) and the QA tab can jump to
+  a segment or fill a suggestion into the draft. `cat_review` exposes the same
+  pass to agents; it never writes the document.
+- **Separate review route.** Settings can give AI review its own provider and
+  model instead of following the pre-translation route. The default still
+  inherits the MT route, so existing configurations keep working.
+
+### Changed
+
+- Store schema v7 adds `segment.notes` and `segment.locked`; v6 data migrates in
+  place.
+- The QA tab now separates mechanical checks (placeholders, numbers, terms,
+  length, residue) from model review suggestions.
+
+### Fixed
+
+- Batch MT maps responses by stable segment id and rejects partial or renumbered
+  output instead of shifting later translations by one segment.
+
+## [1.1.5] - 2026-11-08
+
+Project resources, delivery output, and TM upkeep: each document can bind its own
+TM and glossary libraries, the workbench can price a job and hand off a bilingual
+or XLIFF review package, and TM libraries can be merged, deduplicated, or pruned.
+
+### Added
+
+- **Per-document resources.** A document can inherit the global mounts or bind its
+  own TM/glossary libraries and default write TM. Writes are checked against the
+  document scope (`NO_TM_BOUND`, `TM_NOT_MOUNTED`, `TM_NOT_BOUND`), and deleting a
+  library removes dangling document references.
+- **Word count and leverage analysis.** `analyzeDocument` reports segments, words,
+  characters, translated/confirmed counts, duplicates, ICE/exact/fuzzy/new buckets,
+  match rate, and leverage. The workbench shows the breakdown and downloads it as
+  CSV or JSON.
+- **Bilingual review DOCX.** Export a three-column Source/Target/Status table for
+  reviewers who do not use CAIT.
+- **XLIFF 1.2 review packages.** Export carries the segment uid, target state, and
+  a note; re-import applies only segments matched by uid, XLIFF id, or a unique
+  source, refuses source mismatches, and reports unmatched entries.
+- **TM maintenance.** Preview and confirm merge, deduplicate, and prune operations;
+  merges preserve the target library and optionally delete the source.
+
+### Changed
+
+- Confirmations write to the document's default write TM when one is bound.
+- TM lookup, glossary hits, QA term checks, and pre-translation all honor the
+  document's effective resource scope.
+
+## [1.1.4] - 2026-11-08
+
+Brief and context: CAIT now carries a per-document translation brief and the
+neighbouring source sentences into pre-translation and agent work, and lets a
+reviewer undo a confirmation without losing the translation-memory record.
+
+### Added
+
+- **Document brief.** Six optional fields — domain, audience, tone, notes,
+  do-not-translate terms, and a style sample — are stored with the document,
+  editable from the workbench, and injected into single and batch MT prompts.
+  Agents read and write it with `cat_get_brief` and `cat_set_brief`.
+- **Context for agents.** `cat_get_context` returns the brief, the current
+  segment, its previous/next source segments, and paragraph information, so an
+  agent can resolve pronouns and omitted subjects before translating.
+- **Context-aware batch pre-translation.** Pre-translation now sends up to eight
+  segments per request with the brief and neighbouring source sentences, then
+  falls back to one-segment calls for anything the model omits. A partial batch
+  no longer aborts the document.
+- **Terminology status.** Glossary entries can be marked `forbidden` (never use
+  this rendering) or `preferred` (must use it when the source term appears).
+  The flags survive CSV/JSON/TBX export, CSV import, the glossary UI, and
+  `cat_add_terms`.
+- **Risk-graded QA.** Findings are graded high / medium / low; forbidden terms
+  and length drift are new checks. The report returns `risks` and `riskCounts`,
+  and the rail can filter to segments that need attention.
+- **QA panel and fixes.** A QA tab lists risk counts and findings with jump,
+  copy-source, clear-target, and insert-term actions. A concordance tab and
+  `cat_concordance` search every mounted TM for a word or phrase.
+- **Undo confirmation.** `cat_unconfirm_segment` and the editor button revoke a
+  confirmation, keep the text as a draft, and leave TM untouched. Editing a
+  confirmed segment now revokes its confirmation automatically; confirming it
+  again updates the same TM entry.
+- **Stable segment identity.** Every segment gets a `uid`; the store schema moves
+  to version 5 and migrates version 4 snapshots in place.
+- **Bulk pre-translation cleanup.** The rail can clear every unconfirmed
+  pre-translated or auto-propagated target in the document after a confirmation
+  dialog that states the count and the MT/TM/auto breakdown. Confirmed segments,
+  manual drafts, and saved TM entries are left untouched.
+
+### Changed
+
+- Pre-translation uses batched MT requests with per-segment fallback instead of
+  one request per segment.
+- `getQaReport` also reports `risks` and `riskCounts`; `counts` includes
+  `TERM_FORBIDDEN` and `LENGTH_DRIFT`.
+- `segmentView` includes `uid`; document summaries include `brief`; glossary CSV
+  exports use seven columns including `forbidden` and `preferred`.
+
+### Fixed
+
+- Batch pre-translation now asks for stable segment ids and rejects ambiguous
+  positional output; a dropped, merged, or renumbered entry falls back to
+  one-segment calls instead of shifting later translations onto the wrong
+  segment.
+- `cat_concordance` renders the previous/next context of a match instead of
+  assuming an array.
+- TM and glossary toolbars now use the import/export icon in the direction the
+  button actually moves data; the document toolbar follows the same convention.
+- The library toolbar has an explicit **default write** selector again, and the
+  scope checkboxes no longer silently choose which library receives new entries,
+  imports, exports, or clears.
+- Unconfirming a segment that was just auto-unconfirmed by saving an edit is
+  treated as idempotent instead of showing a `NOT_CONFIRMED` error.
+
 ## [1.1.3] - 2026-11-07
 
 Alignment, repetition and terminology: CAIT can now build a translation memory
@@ -41,6 +180,10 @@ terminology with evidence instead of guessing.
   and never block confirming or exporting.
 - `proposeTranslation` writes only a `suggestion` plus its reason, never the
   target: an agent can propose, the user keeps the last word.
+- The fuzzy-match threshold now defaults to **70%** instead of 55%: below that a
+  match usually shares only function words, so showing it costs review time and
+  saves none. Settings explains this and offers a **Reset** button that appears
+  only while the slider sits away from the default.
 
 ### Fixed
 

package/README.ja.md

@@ -17,12 +17,12 @@
 <p align="center">
   <a href="https://www.npmjs.com/package/dsh-cait"><img alt="npm バージョン" src="https://img.shields.io/npm/v/dsh-cait.svg" /></a>
   <img alt="DSH 0.2.x" src="https://img.shields.io/badge/DSH-0.2.x-4d6bfe.svg" />
-  <img alt="テスト 29 スイート" src="https://img.shields.io/badge/tests-29%20suites-brightgreen.svg" />
+  <img alt="テスト 41 スイート" src="https://img.shields.io/badge/tests-41%20suites-brightgreen.svg" />
   <img alt="API キー不要" src="https://img.shields.io/badge/API%20key-not%20required-6b7280.svg" />
 </p>
 
 CAIT は DSH Web UI と DSH デスクトップに CAT ワークベンチを追加します。左サイドバーの
-エントリから開き、ワークベンチと 15 のエージェントツールが 1 つのストアを共有するため、
+エントリから開き、ワークベンチと 21 のエージェントツールが 1 つのストアを共有するため、
 ソースファイルのインポートから訳文のエクスポートまでを一通りこなせます。
 
 > **DSH `0.2.2-rc.1` で動作確認済み。** CAIT は DSH 自身のモデルルートと認証情報を
@@ -38,7 +38,7 @@ CAIT は DSH Web UI と DSH デスクトップに CAT ワークベンチを追
 | **用語を意識した編集** | 複数の用語集、ソース言語による絞り込み、インライン強調、用語のクリック挿入、用語ごとの大文字小文字の区別。 |
 | **文書ワークフロー** | タブ、閉じる / 開き直し、最近の文書の整理、同一内容の再インポート再利用、文書ごとの進捗。 |
 | **実用的な形式対応** | TXT、Markdown、DOCX、EPUB、TMX、CSV、SRT/VTT、PO、XLIFF 1.2/2.0/SDLXLIFF、HTML、JSON、XLSX、PPTX、RTF。 |
-| **エージェント対応** | 10 のツールが開いている文書とストアを共有するので、エージェントと並行して翻訳できます。 |
+| **エージェント対応** | 21 のツールが開いている文書とストアを共有するので、エージェントと並行して翻訳できます。 |
 
 ## クイックスタート
 
@@ -56,6 +56,32 @@ Host のリビジョンをポーリングするため、エージェントによ
 
 ## 機能
 
+### 1.1.6 の新機能
+
+- **検索と置換**：原文/訳文のプレビュー、一括適用、直前の置換の取り消し。確定済みセグメントの訳文を変更すると校閲待ちに戻ります。
+- **コマンドパレット**（`Ctrl/Cmd+K`）と `Ctrl/Cmd+F` の検索置換。
+- **セグメントメモ**と、セグメント内の**ドラフト undo/redo**（`Ctrl/Cmd+Z`、`Ctrl/Cmd+Shift+Z`、`Ctrl/Cmd+Y`）。
+- テキストセグメントの**マージ・分割・ロック**。マージ/分割は当面 TXT と Markdown のみで、Office の段落マッピングは安全に書き戻せません。
+- **一度しか出ない固有名詞の用語発見**：引用符内の用語、文中の大文字名、中国語の人名/地名/造物名を候補に含め、種類・信頼度・原文の根拠を付けます。用語集への自動追加はしません。
+- **AI 校閲**：QA タブから現在の DSH モデルに構造化された提案を求め、該当セグメントへ移動したり訳文へ反映したりできます。Agent は `cat_review` を呼べますが、どちらも文書を直接書き換えません。校閲は既定で翻訳モデルに従い、設定で専用のプロバイダーとモデルを指定できます。
+
+### 1.1.5 の新機能
+
+- 文書ごとに**グローバルな TM/用語集のマウントを継承するか、独自のライブラリをバインド**でき、既定の書き込み TM も指定できます。検索、用語ヒット、QA、事前翻訳、確定はすべてそのスコープに従います。
+- **語数とレバレッジ分析**：セグメント、語数、文字数、重複、ICE/完全/ファジー/新規の内訳、マッチ率、レバレッジを表示し、CSV/JSON でダウンロードできます。
+- CAIT を使わない校閲者向けに**バイリンガル校閲 DOCX**（原文/訳文/状態）を出力します。
+- **XLIFF 1.2 校閲パッケージ**を出力・再取り込みできます。uid、XLIFF id、一意な原文で照合し、原文不一致は拒否、未一致件数を報告します。
+- **TM メンテナンス**：マージ、重複排除、整理をプレビューしてから確定します。
+
+### 1.1.4 の新機能
+
+- 文書ごとに**翻訳ブリーフ**（領域、読者、トーン、メモ、訳禁用語、スタイル例）を持て、MT プロンプトに注入されます。Agent は `cat_get_brief` で読み取れます。
+- 事前翻訳は**1 リクエスト最大 8 セグメント**のバッチになり、ブリーフと隣接原文を送ります。モデルが返さなかったセグメントは 1 件ずつフォールバックします。
+- 用語集エントリに **forbidden**（使用禁止の訳）と **preferred**（必ず使う訳）を設定でき、CSV/JSON/TBX 出力と CSV 入力で保持されます。
+- QA は **high / medium / low** に分類され、禁止用語と長さドリフトの検査が加わりました。QA タブからジャンプ、原文コピー、訳文クリア、必須用語の挿入ができます。Concordance はマウント中の全 TM を検索します。
+- 確定済みセグメントを**確定解除**でき、訳文と TM エントリは残ります。確定済みセグメントを編集すると自動的に `translated` に戻り、再確定で同じ TM エントリを更新します。
+- セグメントレールから**未確定の事前翻訳を一括クリア**できます。実行前に件数と MT/TM/自動伝播の内訳を示し、確定済み・手動下書き・保存済み TM は対象外であることを明示します。
+
 ### 翻訳メモリ
 
 - id、名前、マウント状態、既定の書き込みライブラリを持つ複数の TM ライブラリ。
@@ -67,7 +93,7 @@ Host のリビジョンをポーリングするため、エージェントによ
   - `near_exact` は数字・句読点・空白を無視したキーで比較し、スコアのしきい値には
     依存しません。
   - `high_fuzzy` は `>= 0.85`、`medium_fuzzy` は `0.75–0.85`、`low_fuzzy` は
-    `0.75` 未満（設定可能な下限、既定値 `0.55` より上）です。
+    `0.75` 未満（設定可能な下限、既定値 `0.70` より上）です。
 - ファジースコアは字句ベースです。文字バイグラムの Dice 72% + トークンの Jaccard 28%。
   ラテン文字は単語単位、CJK は文字単位で扱います。ステミング、同義語、意味的マッチングは
   行いません。
@@ -247,7 +273,7 @@ DSH デスクトップ（または `plugin_manager` ツールがある任意の
 plugin_manager install_bundle target: dsh-cait
 ```
 
-バージョンを固定するときは `dsh-cait@1.1.3`、最新版は `dsh-cait@latest` を指定します。
+バージョンを固定するときは `dsh-cait@1.1.6`、最新版は `dsh-cait@latest` を指定します。
 マネージャが registry のパッケージを解決し、bundle patch を適用してプラグインを有効化します。
 
 CLI 管理のプロファイル:
@@ -349,6 +375,7 @@ dsh web --patch C:\absolute\path\to\dsh-CAIT\_dev\cait.patch.yml
 | `cat_consistency_report` | 訳文が揺れている同一原文をグループ化して一覧 |
 | `cat_term_candidates` | 用語候補の短いリスト: 頻度、例文、メモリにある既存の訳語 |
 | `cat_add_terms` | 選んだ用語を用語集へまとめて登録 |
+| `cat_review` | モデルに構造化された校閲提案を求める。文書は書き換えない |
 
 そのためエージェントは UI を操作せずに文書全体を下書きできます: `cat_pre_translate` を
 `from: next` で `next` が `null` になるまで繰り返し、`cat_get_segments` で確認し、
@@ -364,7 +391,7 @@ dsh web --patch C:\absolute\path\to\dsh-CAIT\_dev\cait.patch.yml
 flowchart LR
   UI["DSH Web UI<br/>(sidebar + workbench)"] --> Client["client.js<br/>generated Web bundle"]
   Client --> API["/cait/api/rpc<br/>same-origin JSON"]
-  Tools["15 agent tools"] --> Store["CatStore"]
+  Tools["21 agent tools"] --> Store["CatStore"]
   API --> Store
   Store --> Snapshot[("store.json<br/>+ sources/")]
 ```
@@ -451,7 +478,7 @@ node _dev/test-mt.mjs
 
 インポート / エクスポートのフィクスチャは `_dev/sample.docx`、`_dev/sample-table.docx`、
 `_dev/sample-format.docx` です。対応する `python _dev/make-*.py` スクリプトで再生成
-できます。29 スイートすべてが通るはずです。
+できます。41 スイートすべてが通るはずです。
 
 ### リリース
 

package/README.md

@@ -6,7 +6,7 @@
 
 <p align="center">
   <strong>Computer-aided translation, inside DeepSeek Harness.</strong><br/>
-  Translation memory · glossary · segment editor · MT assistance · document tabs · Office / subtitle / localization formats
+  Translation memory 路 glossary 路 segment editor 路 MT assistance 路 document tabs 路 Office / subtitle / localization formats
 </p>
 <p align="center">
   <a href="README.zh-CN.md">简体中文</a> · <a href="README.ja.md">日本語</a> · <a href="README.md">English</a>
@@ -16,12 +16,12 @@
 <p align="center">
   <a href="https://www.npmjs.com/package/dsh-cait"><img alt="npm version" src="https://img.shields.io/npm/v/dsh-cait.svg" /></a>
   <img alt="DSH 0.2.x" src="https://img.shields.io/badge/DSH-0.2.x-4d6bfe.svg" />
-  <img alt="Tests: 29 suites" src="https://img.shields.io/badge/tests-29%20suites-brightgreen.svg" />
+  <img alt="Tests: 41 suites" src="https://img.shields.io/badge/tests-41%20suites-brightgreen.svg" />
   <img alt="No API key required" src="https://img.shields.io/badge/API%20key-not%20required-6b7280.svg" />
 </p>
 
 CAIT adds a CAT workbench to the DSH Web UI and Desktop. It opens from a
-left-sidebar entry, keeps one shared store behind the workbench and seven agent
+left-sidebar entry, keeps one shared store behind the workbench and twenty agent
 tools, and covers the practical path from importing a source file to exporting a
 translated document.
 
@@ -38,7 +38,7 @@ translated document.
 | **Glossary-aware editing** | Multiple term bases, source-language filtering, inline highlighting, click-to-insert terms, and per-term case sensitivity. |
 | **A document workflow** | Tabs, close/reopen, recent-document cleanup, duplicate-import reuse, and progress per document. |
 | **Real format coverage** | TXT, Markdown, DOCX, EPUB, TMX, CSV, SRT/VTT, PO, XLIFF 1.2/2.0/SDLXLIFF, HTML, JSON, XLSX, PPTX, and RTF. |
-| **Agent-ready** | Seven tools share the open document and store, so an agent can translate alongside you. |
+| **Agent-ready** | Twenty-one tools share the open document and store, so an agent can translate alongside you. |
 
 ## Quick start
 
@@ -57,6 +57,59 @@ manual reload.
 
 ## Features
 
+### New in 1.1.6
+
+- **Find and replace** across source/target with a preview, one-step apply, and
+  undo; confirmed segments return to review when their target changes.
+- **Command palette** (`Ctrl/Cmd+K`) and `Ctrl/Cmd+F` for find and replace.
+- **Segment notes** for review comments, plus per-segment **draft undo/redo**.
+- **Merge, split, and lock** text segments. Merge/split are limited to TXT and
+  Markdown until Office paragraph mapping can be written back safely.
+- **Term discovery for one-off names.** Quoted terms, capitalized names, and
+  Chinese proper names that occur once are offered as candidates with evidence,
+  type, and confidence; nothing is added to the glossary automatically.
+- **AI review.** The QA tab can ask the configured DSH route for structured
+  suggestions, jump to a segment, and fill a suggestion into the draft. Agents can
+  call `cat_review`; neither path writes the document. Review follows the
+  translation model by default, and Settings can give it a separate provider and
+  model.
+
+### New in 1.1.5
+
+- Each document can **inherit the global TM/glossary mounts or bind its own
+  libraries**, with a default write TM per document. Lookup, glossary hits, QA,
+  pre-translation, and confirmations all honor that scope.
+- **Word count and leverage analysis** reports segments, words, characters,
+  duplicates, ICE/exact/fuzzy/new buckets, match rate, and leverage, with CSV and
+  JSON download.
+- Export a **bilingual review DOCX** (Source/Target/Status) for reviewers who do
+  not use CAIT.
+- Export and re-import an **XLIFF 1.2 review package**. Re-import matches by uid,
+  XLIFF id, or a unique source, refuses source mismatches, and reports unmatched
+  segments.
+- **TM maintenance** previews and confirms merge, deduplicate, and prune actions.
+
+### New in 1.1.4
+
+- Each document can carry a **translation brief** — domain, audience, tone,
+  notes, do-not-translate terms, and a style sample — which is injected into MT
+  prompts and readable by agents through `cat_get_brief`.
+- Pre-translation now sends **up to eight segments per request** with the brief
+  and neighbouring source sentences, then falls back to one segment at a time
+  for anything the model leaves out.
+- Glossary entries can be **forbidden** (never use this rendering) or
+  **preferred** (must use it when the source term appears), and the flags survive
+  CSV/JSON/TBX export and CSV import.
+- QA findings are graded **high / medium / low**, and the QA tab can jump to a
+  finding, copy the source, clear the target, or insert the required term. A
+  concordance tab searches every mounted TM for a word or phrase.
+- A confirmed segment can be **unconfirmed** without deleting its text or its TM
+  entry; editing a confirmed segment revokes the confirmation automatically.
+- The segment rail can **clear all unconfirmed pre-translated targets** in one
+  action. A confirmation dialog states the count and the MT/TM/auto breakdown,
+  and confirms that confirmed segments, manual drafts, and saved TM entries are
+  left untouched.
+
 ### Translation memory
 
 - Multiple TM libraries with id, name, mount state, and a default write library.
@@ -68,7 +121,7 @@ manual reload.
   - `near_exact` compares a key that ignores digits, punctuation, and whitespace;
     it does not depend on a score cutoff.
   - `high_fuzzy` is `>= 0.85`, `medium_fuzzy` is `0.75–0.85`, `low_fuzzy` is
-    below `0.75` (above the configurable floor, default `0.55`).
+    below `0.75` (above the configurable floor, default `0.70`).
 - Fuzzy scoring is lexical: 72% character-bigram Dice + 28% token Jaccard.
   Latin text uses words; CJK text uses individual characters. There is no
   stemming, synonym, or semantic matching.
@@ -92,9 +145,10 @@ manual reload.
   only decides what the workbench searches and writes.
 - **Terminology mining**: *Mine terms from the document* in the glossary toolbar
   lists candidates with occurrence counts, sample wording from memory, and a
-  marker when the term is already known. Counting finds the candidates and ranks
-  them (Latin n-grams, repeated CJK substrings filtered by boundary entropy);
-  a model only judges a shortlist, which is the part models are good at.
+  marker when the term is already known. Counting finds and ranks the candidates
+  (Latin n-grams, repeated CJK substrings filtered by boundary entropy); you
+  confirm the wording and add them. A model-assisted shortlist that can include
+  single-occurrence proper nouns is planned for 1.1.6.
 
 ### Library workspace
 
@@ -136,8 +190,13 @@ separate files in any supported format and any pair of languages.
   size of 20/50/100/200 and a box to jump straight to a segment number. Jumping
   clears the filter when the target segment would otherwise be hidden.
 - **Pre-translate** fills the whole document from exact TM matches and machine
-  translation in small blocks, shows `done/total`, and can be stopped between
-  blocks. Confirmed and already-translated segments are never overwritten.
+  translation in batches of up to eight segments, carrying the document brief
+  and neighbouring source sentences, and falls back to one segment at a time
+  when a batch response is partial. It shows `done/total` and can be stopped
+  between blocks. Confirmed and already-translated segments are never overwritten.
+- **Translation brief** lives beside the document: domain, audience, tone, notes,
+  do-not-translate terms, and a style sample. It is editable from the segment
+  header and used by both pre-translation and agent translation.
 - **Stacked review**: source, the machine/agent baseline, and the editable target
   are laid out top to bottom, so each block gets the full pane width.
   **Confirm all** accepts every segment awaiting review after one confirmation
@@ -149,10 +208,16 @@ separate files in any supported format and any pair of languages.
   unconfirmed copy with the same source (`origin: auto`), editing a draft keeps
   them in sync, and *Fill repeated segments* does the whole document at once.
   Confirmed segments are never touched.
-- Advisory QA markers in the rail and an export-time summary: missing
-  placeholders/tags, mismatched numbers, unused glossary terms, empty targets,
-  an untouched copy of the source, and one source with several translations.
+- Advisory QA markers in the rail and an export-time summary, graded **high /
+  medium / low**: missing placeholders/tags, mismatched numbers, unused or
+  forbidden glossary terms, empty targets, an untouched copy of the source,
+  length drift, and one source with several translations. The QA tab lists
+  findings and offers jump, copy-source, clear-target, and insert-term actions.
   Nothing is blocked and nothing is stored.
+- **Undo confirmation.** A confirmed segment can be unconfirmed back to
+  `translated`; its text and TM entry stay in place. Editing a confirmed segment
+  revokes the confirmation automatically, and confirming it again updates the
+  same TM entry.
 - Document tabs; closing a tab does not delete the document.
 - **Open existing document** reopens a closed document.
 - Interface language switch in **Settings → Interface**: *Follow DSH language*
@@ -194,7 +259,10 @@ CAIT does not own an endpoint or key. The selector is built from DSH:
 - `ctx.llm.listModels(provider)` supplies model ids; CAIT selects the first model
   by default.
 - `ctx.llm.stream({ provider, model, messages, temperature, purpose: 'cait-mt' })`
-  performs the call; CAIT concatenates `text-delta` chunks.
+  performs the call; CAIT concatenates `text-delta` chunks. Batch pre-translation
+  uses `purpose: 'cait-mt-batch'` and asks for a JSON array keyed by position;
+  both paths include the document brief, the neighbouring source sentences, and
+  the usable/forbidden glossary rules in the prompt.
 
 If no usable route exists, the selected route disappears, or it has no models,
 CAIT falls back to offline glossary substitution and says so in the UI and in the
@@ -205,21 +273,21 @@ provider/model/code/status; the client receives only `MT_PROVIDER_FAILED`.
 
 | Format | Import | Export | Notes |
 | --- | :---: | :---: | --- |
-| TXT / Markdown | ✅ | ✅ | Paragraph and sentence segmentation |
-| CSV | ✅ | ✅ | Comma, semicolon, tab, or pipe delimiter auto-detected; source/target column by header name or 1-based number; existing bilingual export |
-| DOCX | ✅ | ✅ | Body paragraphs; table blocks are excluded from import. Export patches the original package, keeping styles, numbering, images, headers, tables, and layout. |
-| EPUB | ✅ | ✅ | EPUB 3 spine order; XHTML text blocks with inline markup kept as tokens and patched back into the retained package. DRM-protected, fixed-layout, and scripted EPUBs are not supported. |
-| TMX 1.4 | ✅ | ✅ | Common `<tu>/<tuv>/<seg>` subset; inline markup flattened with warning; context kept in `<prop type="x-context">` |
-| SRT / WebVTT | ✅ | ✅ | One cue per segment, timings/settings/identifier retained; semantic tags kept; invalid ranges rejected |
-| PO | ✅ | ✅ | Contexts, plural forms, translator comments, obsolete `#~` entries |
-| XLIFF 1.2 | ✅ | ✅ | Inline tags become `{n}` tokens; missing tokens block confirmation |
-| XLIFF 2.0 | ✅ | ✅ | Basic `<unit><segment><source>/<target>` subset |
-| SDLXLIFF | ✅ | ✅ | Basic XLIFF 1.2 path; SDL-specific extensions are not round-tripped |
-| HTML | ✅ | ✅ | Text blocks and inline tags preserved; document CSS/layout is not rebuilt |
-| JSON | ✅ | ✅ | String-path subset; object/array structure preserved via a template; keys and numbers are not translated |
-| XLSX | ✅ | ✅ | First worksheet; `Source`/`Target` headers or columns A/B; export is a new two-column workbook |
-| PPTX | ✅ | ✅* | Source-preserving export; requires the imported `.pptx` package. Body text only; notes/charts/SmartArt not imported. |
-| RTF | ✅ | ✅ | Basic paragraphs, `\uN`, `\'hh`, control words; original styling not round-tripped |
+| TXT / Markdown | 鉁?| 鉁?| Paragraph and sentence segmentation |
+| CSV | 鉁?| 鉁?| Comma, semicolon, tab, or pipe delimiter auto-detected; source/target column by header name or 1-based number; existing bilingual export |
+| DOCX | 鉁?| 鉁?| Body paragraphs; table blocks are excluded from import. Export patches the original package, keeping styles, numbering, images, headers, tables, and layout. |
+| EPUB | 鉁?| 鉁?| EPUB 3 spine order; XHTML text blocks with inline markup kept as tokens and patched back into the retained package. DRM-protected, fixed-layout, and scripted EPUBs are not supported. |
+| TMX 1.4 | 鉁?| 鉁?| Common `<tu>/<tuv>/<seg>` subset; inline markup flattened with warning; context kept in `<prop type="x-context">` |
+| SRT / WebVTT | 鉁?| 鉁?| One cue per segment, timings/settings/identifier retained; semantic tags kept; invalid ranges rejected |
+| PO | 鉁?| 鉁?| Contexts, plural forms, translator comments, obsolete `#~` entries |
+| XLIFF 1.2 | 鉁?| 鉁?| Inline tags become `{n}` tokens; missing tokens block confirmation |
+| XLIFF 2.0 | 鉁?| 鉁?| Basic `<unit><segment><source>/<target>` subset |
+| SDLXLIFF | 鉁?| 鉁?| Basic XLIFF 1.2 path; SDL-specific extensions are not round-tripped |
+| HTML | 鉁?| 鉁?| Text blocks and inline tags preserved; document CSS/layout is not rebuilt |
+| JSON | 鉁?| 鉁?| String-path subset; object/array structure preserved via a template; keys and numbers are not translated |
+| XLSX | 鉁?| 鉁?| First worksheet; `Source`/`Target` headers or columns A/B; export is a new two-column workbook |
+| PPTX | 鉁?| 鉁? | Source-preserving export; requires the imported `.pptx` package. Body text only; notes/charts/SmartArt not imported. |
+| RTF | 鉁?| 鉁?| Basic paragraphs, `\uN`, `\'hh`, control words; original styling not round-tripped |
 
 \* PPTX export fails with `PPTX_SOURCE_REQUIRED` when the retained source package
 is missing rather than silently producing an unformatted deck.
@@ -263,7 +331,7 @@ In DSH Desktop (or any profile with the `plugin_manager` tool):
 plugin_manager install_bundle target: dsh-cait
 ```
 
-Pin a release with `dsh-cait@1.1.3`, or ask for the newest one with
+Pin a release with `dsh-cait@1.1.6`, or ask for the newest one with
 `dsh-cait@latest`. The manager resolves the registry package, applies the bundle
 patch, and activates the plugin — no hand-written profile rows.
 
@@ -365,8 +433,14 @@ confirmed becomes `translated` with `origin: manual`.
 | `cat_propose_translation` | Leave a suggestion plus a reason without touching the confirmed target |
 | `cat_align` | Align two texts and report high / to-review / doubtful pairs |
 | `cat_consistency_report` | Group repeated sources that carry different translations |
-| `cat_term_candidates` | Mining shortlist: frequency, sample sentences, existing memory wording |
+| `cat_term_candidates` | Mining shortlist, including one-off proper names, with type/confidence and evidence |
+| `cat_get_brief` | Read the document brief before translating |
+| `cat_set_brief` | Update brief fields, or clear the whole brief |
+| `cat_get_context` | Brief, current segment, previous/next sources, and paragraph around a segment |
+| `cat_unconfirm_segment` | Revoke a confirmation without deleting the text or TM entry |
+| `cat_concordance` | Search every mounted TM for a word or phrase in source or target |
 | `cat_add_terms` | Add accepted terminology to a library in one batch |
+| `cat_review` | Ask the model for structured review suggestions; never writes the document |
 
 An agent can therefore draft a whole document without clicking in the UI:
 `cat_pre_translate` with `from: next` until `next` is `null`, then
@@ -383,7 +457,7 @@ document the user has open.
 flowchart LR
   UI["DSH Web UI<br/>(sidebar + workbench)"] --> Client["client.js<br/>generated Web bundle"]
   Client --> API["/cait/api/rpc<br/>same-origin JSON"]
-  Tools["15 agent tools"] --> Store["CatStore"]
+  Tools["21 agent tools"] --> Store["CatStore"]
   API --> Store
   Store --> Snapshot[("store.json<br/>+ sources/")]
 ```
@@ -400,48 +474,48 @@ flowchart LR
 
 ```text
 cait/
-├─ package.json
-├─ cordis.patch.yml
-├─ index.js                 # Host half
-├─ client.js                # generated Web bundle
-├─ client/                  # helpers, styles, RPC, components, locales, entry
-├─ scripts/build-client.mjs
-├─ icon.svg
-├─ locale/{en,zh,ja}.json
-├─ NOTICE
-├─ README.md
-├─ README.zh-CN.md
-├─ README.ja.md
-├─ CHANGELOG.md
-├─ lib/
-│  ├─ store.js              # CatStore facade + document/segment core
-│  ├─ errors.js
-│  ├─ store-utils.js
-│  ├─ libraries.js          # TM/glossary registries + mounts
-│  ├─ library-view.js       # Library paging/search + export entry points
-│  ├─ library-export.js     # Glossary TBX / CSV / JSON builders
-│  ├─ tm.js
-│  ├─ glossary.js
-│  ├─ persistence.js
-│  ├─ import-export.js
-│  ├─ formats.js            # CSV/SRT/VTT/PO/XLIFF/HTML/JSON/RTF
-│  ├─ html.js               # Shared HTML/XHTML run scanner + patcher
-│  ├─ epub.js               # EPUB package read / spine import / patch export
-│  ├─ preprocess.js         # Import-time source cleaning stage
-│  ├─ pre-translate.js      # Block pre-translation methods
-│  ├─ qa.js                 # Derived QA report (never persisted)
-│  ├─ xlsx.js
-│  ├─ pptx.js
-│  ├─ segmenter.js
-│  ├─ similarity.js
-│  ├─ tmx.js
-│  ├─ docx.js
-│  ├─ markdown.js
-│  ├─ xml.js
-│  ├─ mt.js
-│  ├─ tools.js
-│  └─ api.js
-└─ _dev/                    # tests, fixtures, local patch overlay
+鈹溾攢 package.json
+鈹溾攢 cordis.patch.yml
+鈹溾攢 index.js                 # Host half
+鈹溾攢 client.js                # generated Web bundle
+鈹溾攢 client/                  # helpers, styles, RPC, components, locales, entry
+鈹溾攢 scripts/build-client.mjs
+鈹溾攢 icon.svg
+鈹溾攢 locale/{en,zh,ja}.json
+鈹溾攢 NOTICE
+鈹溾攢 README.md
+鈹溾攢 README.zh-CN.md
+鈹溾攢 README.ja.md
+鈹溾攢 CHANGELOG.md
+鈹溾攢 lib/
+鈹? 鈹溾攢 store.js              # CatStore facade + document/segment core
+鈹? 鈹溾攢 errors.js
+鈹? 鈹溾攢 store-utils.js
+鈹? 鈹溾攢 libraries.js          # TM/glossary registries + mounts
+鈹? 鈹溾攢 library-view.js       # Library paging/search + export entry points
+鈹? 鈹溾攢 library-export.js     # Glossary TBX / CSV / JSON builders
+鈹? 鈹溾攢 tm.js
+鈹? 鈹溾攢 glossary.js
+鈹? 鈹溾攢 persistence.js
+鈹? 鈹溾攢 import-export.js
+鈹? 鈹溾攢 formats.js            # CSV/SRT/VTT/PO/XLIFF/HTML/JSON/RTF
+鈹? 鈹溾攢 html.js               # Shared HTML/XHTML run scanner + patcher
+鈹? 鈹溾攢 epub.js               # EPUB package read / spine import / patch export
+鈹? 鈹溾攢 preprocess.js         # Import-time source cleaning stage
+鈹? 鈹溾攢 pre-translate.js      # Block pre-translation methods
+鈹? 鈹溾攢 qa.js                 # Derived QA report (never persisted)
+鈹? 鈹溾攢 xlsx.js
+鈹? 鈹溾攢 pptx.js
+鈹? 鈹溾攢 segmenter.js
+鈹? 鈹溾攢 similarity.js
+鈹? 鈹溾攢 tmx.js
+鈹? 鈹溾攢 docx.js
+鈹? 鈹溾攢 markdown.js
+鈹? 鈹溾攢 xml.js
+鈹? 鈹溾攢 mt.js
+鈹? 鈹溾攢 tools.js
+鈹? 鈹斺攢 api.js
+鈹斺攢 _dev/                    # tests, fixtures, local patch overlay
 ```
 
 </details>
@@ -477,13 +551,18 @@ node _dev/test-qa.mjs
 node _dev/test-library-crud.mjs
 node _dev/test-library-view.mjs
 node _dev/test-pagination.mjs
+node _dev/test-replace.mjs
+node _dev/test-notes.mjs
+node _dev/test-lock.mjs
+node _dev/test-segment-edit.mjs
+node _dev/test-review.mjs
 node _dev/test-theme-tokens.mjs
 node _dev/test-e2e-1.1.mjs
 ```
 
 The import/export fixtures are `_dev/sample.docx`, `_dev/sample-table.docx`, and
 `_dev/sample-format.docx`; regenerate them with the matching
-`python _dev/make-*.py` scripts. All 29 suites should pass.
+`python _dev/make-*.py` scripts. All 36 suites should pass.
 
 ### Releasing
 
@@ -543,15 +622,19 @@ GitHub with a release.
 - One translation per source text per context: a repeated sentence keeps a
   separate variant for each surrounding pair, capped at 8 variants per key, and
   reuses the generic entry when the neighbours differ.
-- Context is the two neighbouring source sentences only; document, customer, or
-  project metadata is not part of the match.
+- TM context is the two neighbouring source sentences only; the document brief
+  and project metadata are used by MT prompts but are not part of TM scoring.
 - TBX and JSON glossary exports are one-way; only TMX and CSV import back.
 - The workbench UI ships in Chinese, English, and Japanese, switchable in
   Settings; this README ships in English, Simplified Chinese, and Japanese, and
   the English version is the source of truth.
 
-Planned next: image/vision input through a multimodal DSH route,
-style-preserving XLSX output, and XLIFF `<alt-trans>` / notes support.
+Planned next: per-document TM/glossary binding, word-count and leverage analysis,
+bilingual DOCX review output, and an XLIFF review package (1.1.5); then
+find/replace, segment merge/split/lock, comments, local undo (1.1.6), plus a
+term-extraction overhaul that stops ranking by frequency and can surface
+single-occurrence proper nouns and fiction terms, and an optional LLM review
+pass (1.1.6).
 
 ## Changelog and versioning
 
@@ -564,8 +647,13 @@ and the PDF policy, `v1.1.0` adds agent-driven pre-translation, the
 three-state segment model, stacked review, and advisory QA, `v1.1.1` adds
 the library workspace with TMX/TBX/CSV/JSON export, segment paging, and the
 panel-to-library jump, `v1.1.2` adds context-aware memory with 101% (ICE)
-matches, and `v1.1.3` adds two-file alignment, automatic propagation of
-repeated segments, terminology mining, and the proposal primitive agents use.
+matches, `v1.1.3` adds two-file alignment, automatic propagation of repeated
+segments, terminology mining, and the proposal primitive agents use, and
+`v1.1.4` adds the document brief, context-aware batch pre-translation,
+forbidden/preferred terminology, risk-graded QA with one-click fixes,
+concordance, and undo confirmation, and `v1.1.5` adds per-document TM/glossary
+scopes, word-count and leverage analysis, bilingual DOCX and XLIFF review
+packages, and TM maintenance.
 The project
 was developed as `0.1.0` and renumbered for the GitHub release.