@intlayer/engine 9.5.10 → 9.5.12

package/dist/assets/installSkills/skills/bundle-optimization.md

@@ -0,0 +1,123 @@
+---
+name: intlayer-bundle-optimization
+description: Keeps Intlayer dictionaries small in production bundles (purge, minify, import modes). Use when the user asks to "reduce bundle size", "optimize dictionaries", fix "cannot be purged or minified" / "Opaque field" build warnings, or configure `build.optimize`, `build.purge`, `build.minify` or `importMode`.
+metadata:
+  author: Intlayer
+  url: https://intlayer.org
+  license: Apache-2.0
+  category: productivity
+  tags: [i18n, performance, bundle]
+  documentation: https://intlayer.org/doc/concept/bundle-optimization
+  support: contact@intlayer.org
+---
+
+# Intlayer Bundle Optimization
+
+[Doc](https://intlayer.org/doc/concept/bundle-optimization.md)
+
+## Configuration
+
+```typescript fileName="intlayer.config.ts"
+import type { IntlayerConfig } from "intlayer";
+
+const config: IntlayerConfig = {
+  dictionary: {
+    importMode: "dynamic", // 'static' (default) | 'dynamic' | 'fetch'
+  },
+  build: {
+    optimize: undefined, // auto: enabled in production builds only
+    purge: true, // drop fields never read in source code
+    minify: true, // rename field keys to short aliases (title → a)
+    chunkGrouping: true, // dynamic mode: one dictionary chunk per split boundary
+    dictionariesPreload: true, // dynamic mode: fetch content with its chunk
+  },
+};
+
+export default config;
+```
+
+- `purge` and `minify` do nothing when `optimize` is `false`.
+- `minify` skips key renaming while `editor.enabled` is `true`, and for `importMode: 'fetch'` dictionaries.
+- `importMode: 'dynamic'` ships only the current locale's JSON; it can also be set per dictionary.
+
+## Dictionaries Follow Their Chunk
+
+The build optimization rewrites each `useIntlayer('key')` / `getIntlayer('key')` into a direct import of that dictionary, then empties the global dictionary registry:
+
+- A dictionary is bundled with the component that reads it, so a lazy-loaded route only ships its own content.
+
+With `importMode: 'dynamic'`, `vite-intlayer` adds two more build plugins (build only, not in dev):
+
+- `intlayerChunk` (`build.chunkGrouping`, default `true`) merges the per-dictionary, per-locale chunks by the code-split boundary that uses them (`React.lazy`, split route). Each boundary loads its content in one request per locale. Dictionaries used by several boundaries go to a shared chunk.
+- `intlayerPreload` (`build.dictionariesPreload`, default `true`) starts the dictionary fetch when its chunk is loaded, instead of when the component renders, so navigation does not flash a loading state.
+
+Keep dictionaries scoped to their component or page. A dictionary read from many routes ends up in a shared chunk.
+
+## Write Code the Analyzer Can Follow
+
+Purge and minify rely on a static analysis of every `useIntlayer` / `getIntlayer` call. When a read cannot be followed, the build keeps the whole field or dictionary and warns:
+
+- `Dictionary <key> cannot be purged or minified`: the content object escapes.
+- `Dictionary <key> partially minified. Opaque field: '<path>'`: a field is read with a dynamic key.
+
+### Read fields by name
+
+```tsx
+// ✅ Destructure or use dot access
+const { title, description } = useIntlayer("post");
+```
+
+### Use `select()` instead of dynamic keys
+
+```tsx
+// ❌ Opaque field: 'statuses' is kept whole, keys not minified
+const { statuses } = useIntlayer("post");
+<p>{statuses[post.status]}</p>;
+
+// ✅ post.content.ts: publishStatus: select({ draft: "…", published: "…" })
+const { publishStatus } = useIntlayer("post");
+<p>{publishStatus(post.status)}</p>;
+```
+
+Same for numbers (`enu()`), booleans (`cond()`) and genders (`gender()`).
+
+### Call `useIntlayer` in each component
+
+```tsx
+// ❌ Passing content as a prop: the whole dictionary is kept
+const Post = () => {
+  const content = useIntlayer("post");
+  return <PostFooter content={content} />;
+};
+
+// ✅ Each component reads its own fields; the dictionary is loaded once
+const PostFooter = () => {
+  const { footer } = useIntlayer("post");
+  return <footer>{footer}</footer>;
+};
+```
+
+### Don't let the content object escape
+
+Spreading (`{ ...content }`), iterating (`Object.entries(content)`) or passing it to a helper (`format(content)`) keeps the whole dictionary.
+
+## Accepted Exceptions
+
+Some framework APIs need the content as a value, for example a TanStack Start route that loads it in `loader` and reads it in `head`:
+
+```tsx
+loader: async ({ params }) => ({
+  content: await getIntlayerAsync("admin-metadata", params.locale),
+}),
+head: ({ loaderData }) => ({
+  meta: [{ title: loaderData?.content.title }],
+}),
+```
+
+The analyzer cannot follow `loaderData`, so this dictionary stays whole and the warning is expected. Keep such dictionaries dedicated and small (e.g. a `*-metadata` dictionary holding only the page metadata) so the rest of the content is still optimized.
+
+## References
+
+- [Bundle Optimization](https://intlayer.org/doc/concept/bundle-optimization.md)
+- [Select](https://intlayer.org/doc/concept/content/select.md)
+- [Configuration](https://intlayer.org/doc/concept/configuration.md)

package/dist/assets/installSkills/skills/compat.md

@@ -80,8 +80,4 @@ Once the app runs on Intlayer, components can be moved progressively to the nati
 - [Lingui](https://intlayer.org/doc/compatibility/lingui.md)
 - [NuxtJS I18n](https://intlayer.org/doc/compatibility/nuxtjs-i18n.md)
 - [NGX Translate](https://intlayer.org/doc/compatibility/ngx-translate.md)
-- [Transloco](https://intlayer.org/doc/compatibility/transloco.md)
 - [Svelte I18n](https://intlayer.org/doc/compatibility/svelte-i18n.md)
-- [Next Translate](https://intlayer.org/doc/compatibility/next-translate.md)
-- [Polyglot.js](https://intlayer.org/doc/compatibility/polyglot.md)
-- [i18n-js](https://intlayer.org/doc/compatibility/i18n-js.md)

package/dist/assets/installSkills/skills/content.md

@@ -210,7 +210,7 @@ const publishStatus = select({
 // Usage: publishStatus(post.status)
 ```
 
-> Prefer `select()` over indexing a plain object (`content[status]`): dynamic property access prevents the compiler from pruning and minifying the content.
+> Prefer `select()` over indexing a plain object (`content.statuses[status]`): dynamic property access prevents the compiler from pruning and minifying the content, and the build warns `Opaque field`.
 
 ### Choosing a node by discriminant
 
@@ -304,6 +304,7 @@ import {
   html,
   md,
   nest,
+  select,
   t,
   type Dictionary,
 } from "intlayer";
@@ -312,7 +313,7 @@ const content = {
   key: "test",
   title: "Test component content",
   description:
-    "Content declarations for the Test component, including examples of plurals, conditions, gender-specific messages, dynamic insertions, markdown, file-based content and nested dictionaries used for demonstration and testing purposes.",
+    "Content declarations for the Test component, including examples of plurals, conditions, gender-specific messages, string-based selections, dynamic insertions, markdown, file-based content and nested dictionaries used for demonstration and testing purposes.",
   content: {
     baseContent: "Intlayer", // Content that no need to be i18n
     welcomeMessage: t({
@@ -342,6 +343,11 @@ const content = {
       female: "my content for female users",
       fallback: "my content when gender is not specified", // Optional but avoid undefined type
     }),
+    mySelect: select({
+      draft: "my content when the status is draft",
+      published: "my content when the status is published",
+      fallback: "my content for any other status", // Optional but avoid undefined type
+    }),
     myInsertion: insert(
       "Hello, my name is {{name}} and I am {{age}} years old!"
     ),

package/dist/assets/installSkills/skills/next-js.md

@@ -90,8 +90,8 @@ The [Intlayer Compiler](https://intlayer.org/doc/compiler.md) can extract all yo
 - [Next.js 14](https://intlayer.org/doc/environment/nextjs/14.md)
 - [Next.js 15](https://intlayer.org/doc/environment/nextjs/15.md)
 - [Next.js with Page Router](https://intlayer.org/doc/environment/nextjs/next-with-page-router.md)
-- [Intlayer with next-intl](https://intlayer.org/doc/next-intl.md)
-- [Intlayer with next-i18next](https://intlayer.org/doc/next-i18next.md)
+- [Intlayer with next-intl](https://intlayer.org/doc/migration/next-intl.md)
+- [Intlayer with next-i18next](https://intlayer.org/doc/migration/next-i18next.md)
 
 ### Concepts
 

package/dist/assets/installSkills/skills/usage.md

@@ -19,7 +19,7 @@ To use Intlayer effectively:
 1.  **Retrieve Locales**: Check `intlayer.config.{ts,js,json,json5,jsonc,cjs,mjs}`, `.intlayerrc` to see the configured locales.
 
 2.  **Declare Content**:
-    We recommend creating one content declaration file per component, located alongside the component file. This keeps translations close to the code.
+    We recommend creating one content declaration file per component or per section of your application, located alongside the component file. This keeps translations close to the code.
 
 3.  **Consume Content**: Use the provided hooks and functions to access your content.
     - [Intlayer Exports](https://intlayer.org/doc/packages/intlayer/exports.md)

package/dist/cjs/build.cjs

@@ -19,11 +19,13 @@ const require_loadDictionaries_loadDictionaries = require('./loadDictionaries/lo
 const require_loadDictionaries_loadLocalDictionaries = require('./loadDictionaries/loadLocalDictionaries.cjs');
 const require_writeConfiguration_index = require('./writeConfiguration/index.cjs');
 const require_prepareIntlayer = require('./prepareIntlayer.cjs');
+const require_prepareIntlayerServer = require('./prepareIntlayerServer.cjs');
 const require_writeContentDeclaration_detectExportedComponentName = require('./writeContentDeclaration/detectExportedComponentName.cjs');
 const require_writeContentDeclaration_transformJSFile = require('./writeContentDeclaration/transformJSFile.cjs');
 const require_writeContentDeclaration_writeJSFile = require('./writeContentDeclaration/writeJSFile.cjs');
 const require_writeContentDeclaration_writeContentDeclaration = require('./writeContentDeclaration/writeContentDeclaration.cjs');
 
+exports.INTLAYER_WATCH_ENV_VAR = require_prepareIntlayerServer.INTLAYER_WATCH_ENV_VAR;
 exports.buildDictionary = require_buildIntlayerDictionary_buildIntlayerDictionary.buildDictionary;
 exports.cleanOutputDir = require_cleanOutputDir.cleanOutputDir;
 exports.createDictionaryEntryPoint = require_createDictionaryEntryPoint_createDictionaryEntryPoint.createDictionaryEntryPoint;
@@ -46,6 +48,8 @@ exports.getBuiltDynamicDictionariesPath = require_createDictionaryEntryPoint_get
 exports.getBuiltFetchDictionariesPath = require_createDictionaryEntryPoint_getBuiltFetchDictionariesPath.getBuiltFetchDictionariesPath;
 exports.getBuiltRemoteDictionariesPath = require_createDictionaryEntryPoint_getBuiltRemoteDictionariesPath.getBuiltRemoteDictionariesPath;
 exports.getBuiltUnmergedDictionariesPath = require_createDictionaryEntryPoint_getBuiltUnmergedDictionariesPath.getBuiltUnmergedDictionariesPath;
+exports.getIsDevWatcherCommand = require_prepareIntlayerServer.getIsDevWatcherCommand;
+exports.getIsServerDevelopment = require_prepareIntlayerServer.getIsServerDevelopment;
 exports.getTypeName = require_createType_createModuleAugmentation.getTypeName;
 exports.isCachedConfigurationUpToDate = require_writeConfiguration_index.isCachedConfigurationUpToDate;
 exports.loadContentDeclaration = require_loadDictionaries_loadContentDeclaration.loadContentDeclaration;
@@ -54,6 +58,7 @@ exports.loadDictionaries = require_loadDictionaries_loadDictionaries.loadDiction
 exports.loadLocalDictionaries = require_loadDictionaries_loadLocalDictionaries.loadLocalDictionaries;
 exports.loadRemoteDictionaries = require_loadDictionaries_loadRemoteDictionaries.loadRemoteDictionaries;
 exports.prepareIntlayer = require_prepareIntlayer.prepareIntlayer;
+exports.prepareIntlayerServer = require_prepareIntlayerServer.prepareIntlayerServer;
 exports.processContentDeclaration = require_buildIntlayerDictionary_processContentDeclaration.processContentDeclaration;
 exports.readDictionariesFromDisk = require_utils_readDictionariesFromDisk.readDictionariesFromDisk;
 exports.transformJSFile = require_writeContentDeclaration_transformJSFile.transformJSFile;

package/dist/cjs/cli.cjs

@@ -33,6 +33,7 @@ exports.detectExportedComponentName = require_writeContentDeclaration_detectExpo
 exports.detectFormatCommand = require_detectFormatCommand.detectFormatCommand;
 exports.detectPackageManager = require_init_utils_packageManager.detectPackageManager;
 exports.enableEditorInConfig = require_init_cms.enableEditorInConfig;
+exports.findLockFileDir = require_init_utils_packageManager.findLockFileDir;
 exports.getContentDeclarationFileTemplate = require_getContentDeclarationFileTemplate_getContentDeclarationFileTemplate.getContentDeclarationFileTemplate;
 exports.getDefaultApplicationURL = require_init_index.getDefaultApplicationURL;
 exports.getDocumentationUrl = require_init_documentationRouter.getDocumentationUrl;

package/dist/cjs/docReview/alignBlocks.cjs

@@ -63,6 +63,7 @@ const alignBaseAndTargetBlocks = (baseBlocks, targetBlocks) => {
 	const computeMatchScore = (baseIndex, targetIndex) => {
 		const baseBlock = baseBlocks[baseIndex];
 		const targetBlock = targetBlocks[targetIndex];
+		if (!baseBlock || !targetBlock) return STRUCTURAL_MISMATCH_PENALTY;
 		const hasComparableHeadingDepths = baseBlock.headingDepth !== null && targetBlock.headingDepth !== null;
 		if (hasComparableHeadingDepths && baseBlock.headingDepth !== targetBlock.headingDepth) return STRUCTURAL_MISMATCH_PENALTY;
 		const baseBodyLength = measureBodyLength(baseBlock);
@@ -76,49 +77,65 @@ const alignBaseAndTargetBlocks = (baseBlocks, targetBlocks) => {
 		return headingDepthBonus + typeBonus + lengthBonus + anchorSimilarity * 8;
 	};
 	for (let i = 1; i <= baseLength; i += 1) {
-		scoreMatrix[i][0] = scoreMatrix[i - 1][0] + GAP_PENALTY;
-		traceMatrix[i][0] = "up";
+		const row = scoreMatrix[i];
+		const prevRow = scoreMatrix[i - 1];
+		const traceRow = traceMatrix[i];
+		if (row && prevRow) row[0] = (prevRow[0] ?? 0) + GAP_PENALTY;
+		if (traceRow) traceRow[0] = "up";
 	}
+	const firstRow = scoreMatrix[0];
+	const firstTraceRow = traceMatrix[0];
 	for (let j = 1; j <= targetLength; j += 1) {
-		scoreMatrix[0][j] = scoreMatrix[0][j - 1] + GAP_PENALTY;
-		traceMatrix[0][j] = "left";
+		if (firstRow) firstRow[j] = (firstRow[j - 1] ?? 0) + GAP_PENALTY;
+		if (firstTraceRow) firstTraceRow[j] = "left";
 	}
-	for (let i = 1; i <= baseLength; i += 1) for (let j = 1; j <= targetLength; j += 1) {
-		const match = scoreMatrix[i - 1][j - 1] + computeMatchScore(i - 1, j - 1);
-		const deleteGap = scoreMatrix[i - 1][j] + GAP_PENALTY;
-		const insertGap = scoreMatrix[i][j - 1] + GAP_PENALTY;
-		const best = Math.max(match, deleteGap, insertGap);
-		scoreMatrix[i][j] = best;
-		traceMatrix[i][j] = best === match ? "diagonal" : best === deleteGap ? "up" : "left";
+	for (let i = 1; i <= baseLength; i += 1) {
+		const currentRow = scoreMatrix[i];
+		const prevRow = scoreMatrix[i - 1];
+		const currentTraceRow = traceMatrix[i];
+		if (!currentRow || !prevRow || !currentTraceRow) continue;
+		for (let j = 1; j <= targetLength; j += 1) {
+			const match = (prevRow[j - 1] ?? 0) + computeMatchScore(i - 1, j - 1);
+			const deleteGap = (prevRow[j] ?? 0) + GAP_PENALTY;
+			const insertGap = (currentRow[j - 1] ?? 0) + GAP_PENALTY;
+			const best = Math.max(match, deleteGap, insertGap);
+			currentRow[j] = best;
+			currentTraceRow[j] = best === match ? "diagonal" : best === deleteGap ? "up" : "left";
+		}
 	}
 	const result = [];
 	let i = baseLength;
 	let j = targetLength;
-	while (i > 0 || j > 0) if (i > 0 && j > 0 && traceMatrix[i][j] === "diagonal") {
-		const baseIndex = i - 1;
-		const targetIndex = j - 1;
-		const similarityScore = require_docReview_computeSimilarity.computeJaccardSimilarity(baseBlocks[baseIndex].anchorText, targetBlocks[targetIndex].anchorText, 3);
-		result.unshift({
-			baseIndex,
-			targetIndex,
-			similarityScore
-		});
-		i -= 1;
-		j -= 1;
-	} else if (i > 0 && (j === 0 || traceMatrix[i][j] === "up")) {
-		result.unshift({
-			baseIndex: i - 1,
-			targetIndex: null,
-			similarityScore: 0
-		});
-		i -= 1;
-	} else if (j > 0 && (i === 0 || traceMatrix[i][j] === "left")) {
-		result.unshift({
-			baseIndex: -1,
-			targetIndex: j - 1,
-			similarityScore: 0
-		});
-		j -= 1;
+	while (i > 0 || j > 0) {
+		const direction = traceMatrix[i]?.[j];
+		if (i > 0 && j > 0 && direction === "diagonal") {
+			const baseIndex = i - 1;
+			const targetIndex = j - 1;
+			const baseAnchor = baseBlocks[baseIndex]?.anchorText ?? "";
+			const targetAnchor = targetBlocks[targetIndex]?.anchorText ?? "";
+			const similarityScore = require_docReview_computeSimilarity.computeJaccardSimilarity(baseAnchor, targetAnchor, 3);
+			result.unshift({
+				baseIndex,
+				targetIndex,
+				similarityScore
+			});
+			i -= 1;
+			j -= 1;
+		} else if (i > 0 && (j === 0 || direction === "up")) {
+			result.unshift({
+				baseIndex: i - 1,
+				targetIndex: null,
+				similarityScore: 0
+			});
+			i -= 1;
+		} else if (j > 0 && (i === 0 || direction === "left")) {
+			result.unshift({
+				baseIndex: -1,
+				targetIndex: j - 1,
+				similarityScore: 0
+			});
+			j -= 1;
+		}
 	}
 	return result;
 };

package/dist/cjs/docReview/alignBlocks.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"alignBlocks.cjs","names":["computeJaccardSimilarity"],"sources":["../../../src/docReview/alignBlocks.ts"],"sourcesContent":["import { computeJaccardSimilarity } from './computeSimilarity';\nimport type { AlignmentPair, FingerprintedBlock } from './types';\n\n/** Cost of leaving a block unaligned (an insertion or a deletion). */\nconst GAP_PENALTY = -2;\n\n/**\n * Score of a pair that structural evidence rules out.\n *\n * Strictly below the cost of two gaps (`2 * GAP_PENALTY`) so the aligner always\n * prefers reporting an insertion plus a deletion over such a pair. Every other\n * score is positive, which means that without an explicit veto the aligner would\n * rather pair two unrelated blocks than leave them unaligned — and a bogus pair\n * is planned as `reuse`, which keeps the stale translation verbatim *and*\n * inserts the freshly translated base block next to it, producing the duplicated\n * headings this penalty exists to prevent.\n */\nconst STRUCTURAL_MISMATCH_PENALTY = GAP_PENALTY * 2 - 1;\n\n/** Reward for two blocks opened by a heading of the very same depth. */\nconst HEADING_DEPTH_MATCH_BONUS = 3;\n\n/** Minimum length ratio for the (small) \"comparable size\" reward. */\nconst COMPARABLE_LENGTH_RATIO = 0.75;\n\n/**\n * Body length from which a section is considered to carry content of its own.\n *\n * Kept above a single sentence so a translator's short lead-in paragraph is not\n * mistaken for a whole section body.\n */\nconst SUBSTANTIAL_BODY_LENGTH = 80;\n\n/**\n * Length of the body a block carries below its opening heading.\n *\n * A section that only holds its heading (because subsections carry all of its\n * content) is structurally different from one that holds a full body, and that\n * difference survives translation.\n *\n * @param block - The block to measure.\n * @returns The trimmed length of everything below the opening heading line.\n */\nconst measureBodyLength = (block: FingerprintedBlock): number => {\n  if (block.headingDepth === null) return block.content.trim().length;\n\n  const [, ...bodyLines] = block.content.split('\\n');\n\n  return bodyLines.join('\\n').trim().length;\n};\n\n/**\n * Align the blocks of a base document with the blocks of its translation using a\n * Needleman–Wunsch global alignment over heading depth, anchor similarity and\n * block type.\n *\n * Because prose differs across languages, the score is weighted toward the\n * structural signals — heading depth first, then the anchor (digits and symbols)\n * — rather than the words themselves.\n *\n * @param baseBlocks - Blocks of the base (source) document.\n * @param targetBlocks - Blocks of the target (translated) document.\n * @returns The ordered list of alignment pairs, including insertions and deletions.\n */\nexport const alignBaseAndTargetBlocks = (\n  baseBlocks: FingerprintedBlock[],\n  targetBlocks: FingerprintedBlock[]\n): AlignmentPair[] => {\n  const baseLength = baseBlocks.length;\n  const targetLength = targetBlocks.length;\n\n  const scoreMatrix: number[][] = Array.from({ length: baseLength + 1 }, () =>\n    Array.from({ length: targetLength + 1 }, () => 0)\n  );\n  const traceMatrix: ('diagonal' | 'up' | 'left')[][] = Array.from(\n    { length: baseLength + 1 },\n    () => Array.from({ length: targetLength + 1 }, () => 'diagonal')\n  );\n\n  const computeMatchScore = (\n    baseIndex: number,\n    targetIndex: number\n  ): number => {\n    const baseBlock = baseBlocks[baseIndex]!;\n    const targetBlock = targetBlocks[targetIndex]!;\n\n    // Translating a document never changes its heading depths, so two headings\n    // of different depths cannot be counterparts, however similar their content.\n    const hasComparableHeadingDepths =\n      baseBlock.headingDepth !== null && targetBlock.headingDepth !== null;\n\n    if (\n      hasComparableHeadingDepths &&\n      baseBlock.headingDepth !== targetBlock.headingDepth\n    ) {\n      return STRUCTURAL_MISMATCH_PENALTY;\n    }\n\n    // A section reduced to its bare heading (its content living in subsections)\n    // is not the translation of a section holding a full body. Pairing them\n    // reuses that whole body while the base heading is translated again right\n    // next to it — which is exactly how a duplicated heading appears.\n    const baseBodyLength = measureBodyLength(baseBlock);\n    const targetBodyLength = measureBodyLength(targetBlock);\n    const isBodyPresenceMismatched =\n      (baseBodyLength === 0 && targetBodyLength >= SUBSTANTIAL_BODY_LENGTH) ||\n      (targetBodyLength === 0 && baseBodyLength >= SUBSTANTIAL_BODY_LENGTH);\n\n    if (isBodyPresenceMismatched) return STRUCTURAL_MISMATCH_PENALTY;\n\n    const lengthRatio =\n      Math.min(baseBlock.content.length, targetBlock.content.length) /\n      Math.max(baseBlock.content.length, targetBlock.content.length);\n\n    const headingDepthBonus = hasComparableHeadingDepths\n      ? HEADING_DEPTH_MATCH_BONUS\n      : 0;\n    const typeBonus = baseBlock.type === targetBlock.type ? 2 : 0;\n    const anchorSimilarity = computeJaccardSimilarity(\n      baseBlock.anchorText,\n      targetBlock.anchorText,\n      3\n    );\n    const lengthBonus = lengthRatio > COMPARABLE_LENGTH_RATIO ? 1 : 0;\n\n    // weighted toward the structural signals (heading depth, then anchor)\n    return headingDepthBonus + typeBonus + lengthBonus + anchorSimilarity * 8;\n  };\n\n  // initialize first row and column\n  for (let i = 1; i <= baseLength; i += 1) {\n    scoreMatrix[i][0] = scoreMatrix[i - 1][0] + GAP_PENALTY;\n    traceMatrix[i][0] = 'up';\n  }\n  for (let j = 1; j <= targetLength; j += 1) {\n    scoreMatrix[0][j] = scoreMatrix[0][j - 1] + GAP_PENALTY;\n    traceMatrix[0][j] = 'left';\n  }\n\n  // fill\n  for (let i = 1; i <= baseLength; i += 1) {\n    for (let j = 1; j <= targetLength; j += 1) {\n      const match = scoreMatrix[i - 1][j - 1] + computeMatchScore(i - 1, j - 1);\n      const deleteGap = scoreMatrix[i - 1][j] + GAP_PENALTY;\n      const insertGap = scoreMatrix[i][j - 1] + GAP_PENALTY;\n\n      const best = Math.max(match, deleteGap, insertGap);\n      scoreMatrix[i][j] = best;\n      traceMatrix[i][j] =\n        best === match ? 'diagonal' : best === deleteGap ? 'up' : 'left';\n    }\n  }\n\n  // traceback\n  const result: AlignmentPair[] = [];\n  let i = baseLength;\n  let j = targetLength;\n  while (i > 0 || j > 0) {\n    if (i > 0 && j > 0 && traceMatrix[i][j] === 'diagonal') {\n      const baseIndex = i - 1;\n      const targetIndex = j - 1;\n      const similarityScore = computeJaccardSimilarity(\n        baseBlocks[baseIndex].anchorText,\n        targetBlocks[targetIndex].anchorText,\n        3\n      );\n      result.unshift({ baseIndex, targetIndex, similarityScore });\n      i -= 1;\n      j -= 1;\n    } else if (i > 0 && (j === 0 || traceMatrix[i][j] === 'up')) {\n      result.unshift({\n        baseIndex: i - 1,\n        targetIndex: null,\n        similarityScore: 0,\n      });\n      i -= 1;\n    } else if (j > 0 && (i === 0 || traceMatrix[i][j] === 'left')) {\n      // target block has no corresponding base block (deleted)\n      result.unshift({\n        baseIndex: -1,\n        targetIndex: j - 1,\n        similarityScore: 0,\n      });\n      j -= 1;\n    }\n  }\n  return result;\n};\n"],"mappings":";;;;;AAIA,MAAM,cAAc;;;;;;;;;;;;AAapB,MAAM,8BAA8B;;AAGpC,MAAM,4BAA4B;;AAGlC,MAAM,0BAA0B;;;;;;;AAQhC,MAAM,0BAA0B;;;;;;;;;;;AAYhC,MAAM,qBAAqB,UAAsC;CAC/D,IAAI,MAAM,iBAAiB,MAAM,OAAO,MAAM,QAAQ,KAAK,CAAC,CAAC;CAE7D,MAAM,GAAG,GAAG,aAAa,MAAM,QAAQ,MAAM,IAAI;CAEjD,OAAO,UAAU,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC;AACrC;;;;;;;;;;;;;;AAeA,MAAa,4BACX,YACA,iBACoB;CACpB,MAAM,aAAa,WAAW;CAC9B,MAAM,eAAe,aAAa;CAElC,MAAM,cAA0B,MAAM,KAAK,EAAE,QAAQ,aAAa,EAAE,SAClE,MAAM,KAAK,EAAE,QAAQ,eAAe,EAAE,SAAS,CAAC,CAClD;CACA,MAAM,cAAgD,MAAM,KAC1D,EAAE,QAAQ,aAAa,EAAE,SACnB,MAAM,KAAK,EAAE,QAAQ,eAAe,EAAE,SAAS,UAAU,CACjE;CAEA,MAAM,qBACJ,WACA,gBACW;EACX,MAAM,YAAY,WAAW;EAC7B,MAAM,cAAc,aAAa;EAIjC,MAAM,6BACJ,UAAU,iBAAiB,QAAQ,YAAY,iBAAiB;EAElE,IACE,8BACA,UAAU,iBAAiB,YAAY,cAEvC,OAAO;EAOT,MAAM,iBAAiB,kBAAkB,SAAS;EAClD,MAAM,mBAAmB,kBAAkB,WAAW;EAKtD,IAHG,mBAAmB,KAAK,oBAAoB,2BAC5C,qBAAqB,KAAK,kBAAkB,yBAEjB,OAAO;EAErC,MAAM,cACJ,KAAK,IAAI,UAAU,QAAQ,QAAQ,YAAY,QAAQ,MAAM,IAC7D,KAAK,IAAI,UAAU,QAAQ,QAAQ,YAAY,QAAQ,MAAM;EAE/D,MAAM,oBAAoB,6BACtB,4BACA;EACJ,MAAM,YAAY,UAAU,SAAS,YAAY,OAAO,IAAI;EAC5D,MAAM,mBAAmBA,6DACvB,UAAU,YACV,YAAY,YACZ,CACF;EACA,MAAM,cAAc,cAAc,0BAA0B,IAAI;EAGhE,OAAO,oBAAoB,YAAY,cAAc,mBAAmB;CAC1E;CAGA,KAAK,IAAI,IAAI,GAAG,KAAK,YAAY,KAAK,GAAG;EACvC,YAAY,EAAE,CAAC,KAAK,YAAY,IAAI,EAAE,CAAC,KAAK;EAC5C,YAAY,EAAE,CAAC,KAAK;CACtB;CACA,KAAK,IAAI,IAAI,GAAG,KAAK,cAAc,KAAK,GAAG;EACzC,YAAY,EAAE,CAAC,KAAK,YAAY,EAAE,CAAC,IAAI,KAAK;EAC5C,YAAY,EAAE,CAAC,KAAK;CACtB;CAGA,KAAK,IAAI,IAAI,GAAG,KAAK,YAAY,KAAK,GACpC,KAAK,IAAI,IAAI,GAAG,KAAK,cAAc,KAAK,GAAG;EACzC,MAAM,QAAQ,YAAY,IAAI,EAAE,CAAC,IAAI,KAAK,kBAAkB,IAAI,GAAG,IAAI,CAAC;EACxE,MAAM,YAAY,YAAY,IAAI,EAAE,CAAC,KAAK;EAC1C,MAAM,YAAY,YAAY,EAAE,CAAC,IAAI,KAAK;EAE1C,MAAM,OAAO,KAAK,IAAI,OAAO,WAAW,SAAS;EACjD,YAAY,EAAE,CAAC,KAAK;EACpB,YAAY,EAAE,CAAC,KACb,SAAS,QAAQ,aAAa,SAAS,YAAY,OAAO;CAC9D;CAIF,MAAM,SAA0B,CAAC;CACjC,IAAI,IAAI;CACR,IAAI,IAAI;CACR,OAAO,IAAI,KAAK,IAAI,GAClB,IAAI,IAAI,KAAK,IAAI,KAAK,YAAY,EAAE,CAAC,OAAO,YAAY;EACtD,MAAM,YAAY,IAAI;EACtB,MAAM,cAAc,IAAI;EACxB,MAAM,kBAAkBA,6DACtB,WAAW,UAAU,CAAC,YACtB,aAAa,YAAY,CAAC,YAC1B,CACF;EACA,OAAO,QAAQ;GAAE;GAAW;GAAa;EAAgB,CAAC;EAC1D,KAAK;EACL,KAAK;CACP,OAAO,IAAI,IAAI,MAAM,MAAM,KAAK,YAAY,EAAE,CAAC,OAAO,OAAO;EAC3D,OAAO,QAAQ;GACb,WAAW,IAAI;GACf,aAAa;GACb,iBAAiB;EACnB,CAAC;EACD,KAAK;CACP,OAAO,IAAI,IAAI,MAAM,MAAM,KAAK,YAAY,EAAE,CAAC,OAAO,SAAS;EAE7D,OAAO,QAAQ;GACb,WAAW;GACX,aAAa,IAAI;GACjB,iBAAiB;EACnB,CAAC;EACD,KAAK;CACP;CAEF,OAAO;AACT"}
+{"version":3,"file":"alignBlocks.cjs","names":["computeJaccardSimilarity"],"sources":["../../../src/docReview/alignBlocks.ts"],"sourcesContent":["import { computeJaccardSimilarity } from './computeSimilarity';\nimport type { AlignmentPair, FingerprintedBlock } from './types';\n\n/** Cost of leaving a block unaligned (an insertion or a deletion). */\nconst GAP_PENALTY = -2;\n\n/**\n * Score of a pair that structural evidence rules out.\n *\n * Strictly below the cost of two gaps (`2 * GAP_PENALTY`) so the aligner always\n * prefers reporting an insertion plus a deletion over such a pair. Every other\n * score is positive, which means that without an explicit veto the aligner would\n * rather pair two unrelated blocks than leave them unaligned — and a bogus pair\n * is planned as `reuse`, which keeps the stale translation verbatim *and*\n * inserts the freshly translated base block next to it, producing the duplicated\n * headings this penalty exists to prevent.\n */\nconst STRUCTURAL_MISMATCH_PENALTY = GAP_PENALTY * 2 - 1;\n\n/** Reward for two blocks opened by a heading of the very same depth. */\nconst HEADING_DEPTH_MATCH_BONUS = 3;\n\n/** Minimum length ratio for the (small) \"comparable size\" reward. */\nconst COMPARABLE_LENGTH_RATIO = 0.75;\n\n/**\n * Body length from which a section is considered to carry content of its own.\n *\n * Kept above a single sentence so a translator's short lead-in paragraph is not\n * mistaken for a whole section body.\n */\nconst SUBSTANTIAL_BODY_LENGTH = 80;\n\n/**\n * Length of the body a block carries below its opening heading.\n *\n * A section that only holds its heading (because subsections carry all of its\n * content) is structurally different from one that holds a full body, and that\n * difference survives translation.\n *\n * @param block - The block to measure.\n * @returns The trimmed length of everything below the opening heading line.\n */\nconst measureBodyLength = (block: FingerprintedBlock): number => {\n  if (block.headingDepth === null) return block.content.trim().length;\n\n  const [, ...bodyLines] = block.content.split('\\n');\n\n  return bodyLines.join('\\n').trim().length;\n};\n\n/**\n * Align the blocks of a base document with the blocks of its translation using a\n * Needleman–Wunsch global alignment over heading depth, anchor similarity and\n * block type.\n *\n * Because prose differs across languages, the score is weighted toward the\n * structural signals — heading depth first, then the anchor (digits and symbols)\n * — rather than the words themselves.\n *\n * @param baseBlocks - Blocks of the base (source) document.\n * @param targetBlocks - Blocks of the target (translated) document.\n * @returns The ordered list of alignment pairs, including insertions and deletions.\n */\nexport const alignBaseAndTargetBlocks = (\n  baseBlocks: FingerprintedBlock[],\n  targetBlocks: FingerprintedBlock[]\n): AlignmentPair[] => {\n  const baseLength = baseBlocks.length;\n  const targetLength = targetBlocks.length;\n\n  const scoreMatrix: number[][] = Array.from({ length: baseLength + 1 }, () =>\n    Array.from({ length: targetLength + 1 }, () => 0)\n  );\n  const traceMatrix: ('diagonal' | 'up' | 'left')[][] = Array.from(\n    { length: baseLength + 1 },\n    () => Array.from({ length: targetLength + 1 }, () => 'diagonal')\n  );\n\n  const computeMatchScore = (\n    baseIndex: number,\n    targetIndex: number\n  ): number => {\n    const baseBlock = baseBlocks[baseIndex];\n    const targetBlock = targetBlocks[targetIndex];\n\n    if (!baseBlock || !targetBlock) return STRUCTURAL_MISMATCH_PENALTY;\n\n    // Translating a document never changes its heading depths, so two headings\n    // of different depths cannot be counterparts, however similar their content.\n    const hasComparableHeadingDepths =\n      baseBlock.headingDepth !== null && targetBlock.headingDepth !== null;\n\n    if (\n      hasComparableHeadingDepths &&\n      baseBlock.headingDepth !== targetBlock.headingDepth\n    ) {\n      return STRUCTURAL_MISMATCH_PENALTY;\n    }\n\n    // A section reduced to its bare heading (its content living in subsections)\n    // is not the translation of a section holding a full body. Pairing them\n    // reuses that whole body while the base heading is translated again right\n    // next to it — which is exactly how a duplicated heading appears.\n    const baseBodyLength = measureBodyLength(baseBlock);\n    const targetBodyLength = measureBodyLength(targetBlock);\n    const isBodyPresenceMismatched =\n      (baseBodyLength === 0 && targetBodyLength >= SUBSTANTIAL_BODY_LENGTH) ||\n      (targetBodyLength === 0 && baseBodyLength >= SUBSTANTIAL_BODY_LENGTH);\n\n    if (isBodyPresenceMismatched) return STRUCTURAL_MISMATCH_PENALTY;\n\n    const lengthRatio =\n      Math.min(baseBlock.content.length, targetBlock.content.length) /\n      Math.max(baseBlock.content.length, targetBlock.content.length);\n\n    const headingDepthBonus = hasComparableHeadingDepths\n      ? HEADING_DEPTH_MATCH_BONUS\n      : 0;\n    const typeBonus = baseBlock.type === targetBlock.type ? 2 : 0;\n    const anchorSimilarity = computeJaccardSimilarity(\n      baseBlock.anchorText,\n      targetBlock.anchorText,\n      3\n    );\n    const lengthBonus = lengthRatio > COMPARABLE_LENGTH_RATIO ? 1 : 0;\n\n    // weighted toward the structural signals (heading depth, then anchor)\n    return headingDepthBonus + typeBonus + lengthBonus + anchorSimilarity * 8;\n  };\n\n  // initialize first row and column\n  for (let i = 1; i <= baseLength; i += 1) {\n    const row = scoreMatrix[i];\n    const prevRow = scoreMatrix[i - 1];\n    const traceRow = traceMatrix[i];\n\n    if (row && prevRow) {\n      row[0] = (prevRow[0] ?? 0) + GAP_PENALTY;\n    }\n    if (traceRow) {\n      traceRow[0] = 'up';\n    }\n  }\n  const firstRow = scoreMatrix[0];\n  const firstTraceRow = traceMatrix[0];\n\n  for (let j = 1; j <= targetLength; j += 1) {\n    if (firstRow) {\n      firstRow[j] = (firstRow[j - 1] ?? 0) + GAP_PENALTY;\n    }\n    if (firstTraceRow) {\n      firstTraceRow[j] = 'left';\n    }\n  }\n\n  // fill\n  for (let i = 1; i <= baseLength; i += 1) {\n    const currentRow = scoreMatrix[i];\n    const prevRow = scoreMatrix[i - 1];\n    const currentTraceRow = traceMatrix[i];\n\n    if (!currentRow || !prevRow || !currentTraceRow) continue;\n\n    for (let j = 1; j <= targetLength; j += 1) {\n      const match = (prevRow[j - 1] ?? 0) + computeMatchScore(i - 1, j - 1);\n      const deleteGap = (prevRow[j] ?? 0) + GAP_PENALTY;\n      const insertGap = (currentRow[j - 1] ?? 0) + GAP_PENALTY;\n\n      const best = Math.max(match, deleteGap, insertGap);\n      currentRow[j] = best;\n      currentTraceRow[j] =\n        best === match ? 'diagonal' : best === deleteGap ? 'up' : 'left';\n    }\n  }\n\n  // traceback\n  const result: AlignmentPair[] = [];\n  let i = baseLength;\n  let j = targetLength;\n  while (i > 0 || j > 0) {\n    const direction = traceMatrix[i]?.[j];\n    if (i > 0 && j > 0 && direction === 'diagonal') {\n      const baseIndex = i - 1;\n      const targetIndex = j - 1;\n      const baseAnchor = baseBlocks[baseIndex]?.anchorText ?? '';\n      const targetAnchor = targetBlocks[targetIndex]?.anchorText ?? '';\n      const similarityScore = computeJaccardSimilarity(\n        baseAnchor,\n        targetAnchor,\n        3\n      );\n      result.unshift({ baseIndex, targetIndex, similarityScore });\n      i -= 1;\n      j -= 1;\n    } else if (i > 0 && (j === 0 || direction === 'up')) {\n      result.unshift({\n        baseIndex: i - 1,\n        targetIndex: null,\n        similarityScore: 0,\n      });\n      i -= 1;\n    } else if (j > 0 && (i === 0 || direction === 'left')) {\n      // target block has no corresponding base block (deleted)\n      result.unshift({\n        baseIndex: -1,\n        targetIndex: j - 1,\n        similarityScore: 0,\n      });\n      j -= 1;\n    }\n  }\n  return result;\n};\n"],"mappings":";;;;;AAIA,MAAM,cAAc;;;;;;;;;;;;AAapB,MAAM,8BAA8B;;AAGpC,MAAM,4BAA4B;;AAGlC,MAAM,0BAA0B;;;;;;;AAQhC,MAAM,0BAA0B;;;;;;;;;;;AAYhC,MAAM,qBAAqB,UAAsC;CAC/D,IAAI,MAAM,iBAAiB,MAAM,OAAO,MAAM,QAAQ,KAAK,CAAC,CAAC;CAE7D,MAAM,GAAG,GAAG,aAAa,MAAM,QAAQ,MAAM,IAAI;CAEjD,OAAO,UAAU,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC;AACrC;;;;;;;;;;;;;;AAeA,MAAa,4BACX,YACA,iBACoB;CACpB,MAAM,aAAa,WAAW;CAC9B,MAAM,eAAe,aAAa;CAElC,MAAM,cAA0B,MAAM,KAAK,EAAE,QAAQ,aAAa,EAAE,SAClE,MAAM,KAAK,EAAE,QAAQ,eAAe,EAAE,SAAS,CAAC,CAClD;CACA,MAAM,cAAgD,MAAM,KAC1D,EAAE,QAAQ,aAAa,EAAE,SACnB,MAAM,KAAK,EAAE,QAAQ,eAAe,EAAE,SAAS,UAAU,CACjE;CAEA,MAAM,qBACJ,WACA,gBACW;EACX,MAAM,YAAY,WAAW;EAC7B,MAAM,cAAc,aAAa;EAEjC,IAAI,CAAC,aAAa,CAAC,aAAa,OAAO;EAIvC,MAAM,6BACJ,UAAU,iBAAiB,QAAQ,YAAY,iBAAiB;EAElE,IACE,8BACA,UAAU,iBAAiB,YAAY,cAEvC,OAAO;EAOT,MAAM,iBAAiB,kBAAkB,SAAS;EAClD,MAAM,mBAAmB,kBAAkB,WAAW;EAKtD,IAHG,mBAAmB,KAAK,oBAAoB,2BAC5C,qBAAqB,KAAK,kBAAkB,yBAEjB,OAAO;EAErC,MAAM,cACJ,KAAK,IAAI,UAAU,QAAQ,QAAQ,YAAY,QAAQ,MAAM,IAC7D,KAAK,IAAI,UAAU,QAAQ,QAAQ,YAAY,QAAQ,MAAM;EAE/D,MAAM,oBAAoB,6BACtB,4BACA;EACJ,MAAM,YAAY,UAAU,SAAS,YAAY,OAAO,IAAI;EAC5D,MAAM,mBAAmBA,6DACvB,UAAU,YACV,YAAY,YACZ,CACF;EACA,MAAM,cAAc,cAAc,0BAA0B,IAAI;EAGhE,OAAO,oBAAoB,YAAY,cAAc,mBAAmB;CAC1E;CAGA,KAAK,IAAI,IAAI,GAAG,KAAK,YAAY,KAAK,GAAG;EACvC,MAAM,MAAM,YAAY;EACxB,MAAM,UAAU,YAAY,IAAI;EAChC,MAAM,WAAW,YAAY;EAE7B,IAAI,OAAO,SACT,IAAI,MAAM,QAAQ,MAAM,KAAK;EAE/B,IAAI,UACF,SAAS,KAAK;CAElB;CACA,MAAM,WAAW,YAAY;CAC7B,MAAM,gBAAgB,YAAY;CAElC,KAAK,IAAI,IAAI,GAAG,KAAK,cAAc,KAAK,GAAG;EACzC,IAAI,UACF,SAAS,MAAM,SAAS,IAAI,MAAM,KAAK;EAEzC,IAAI,eACF,cAAc,KAAK;CAEvB;CAGA,KAAK,IAAI,IAAI,GAAG,KAAK,YAAY,KAAK,GAAG;EACvC,MAAM,aAAa,YAAY;EAC/B,MAAM,UAAU,YAAY,IAAI;EAChC,MAAM,kBAAkB,YAAY;EAEpC,IAAI,CAAC,cAAc,CAAC,WAAW,CAAC,iBAAiB;EAEjD,KAAK,IAAI,IAAI,GAAG,KAAK,cAAc,KAAK,GAAG;GACzC,MAAM,SAAS,QAAQ,IAAI,MAAM,KAAK,kBAAkB,IAAI,GAAG,IAAI,CAAC;GACpE,MAAM,aAAa,QAAQ,MAAM,KAAK;GACtC,MAAM,aAAa,WAAW,IAAI,MAAM,KAAK;GAE7C,MAAM,OAAO,KAAK,IAAI,OAAO,WAAW,SAAS;GACjD,WAAW,KAAK;GAChB,gBAAgB,KACd,SAAS,QAAQ,aAAa,SAAS,YAAY,OAAO;EAC9D;CACF;CAGA,MAAM,SAA0B,CAAC;CACjC,IAAI,IAAI;CACR,IAAI,IAAI;CACR,OAAO,IAAI,KAAK,IAAI,GAAG;EACrB,MAAM,YAAY,YAAY,EAAE,GAAG;EACnC,IAAI,IAAI,KAAK,IAAI,KAAK,cAAc,YAAY;GAC9C,MAAM,YAAY,IAAI;GACtB,MAAM,cAAc,IAAI;GACxB,MAAM,aAAa,WAAW,UAAU,EAAE,cAAc;GACxD,MAAM,eAAe,aAAa,YAAY,EAAE,cAAc;GAC9D,MAAM,kBAAkBA,6DACtB,YACA,cACA,CACF;GACA,OAAO,QAAQ;IAAE;IAAW;IAAa;GAAgB,CAAC;GAC1D,KAAK;GACL,KAAK;EACP,OAAO,IAAI,IAAI,MAAM,MAAM,KAAK,cAAc,OAAO;GACnD,OAAO,QAAQ;IACb,WAAW,IAAI;IACf,aAAa;IACb,iBAAiB;GACnB,CAAC;GACD,KAAK;EACP,OAAO,IAAI,IAAI,MAAM,MAAM,KAAK,cAAc,SAAS;GAErD,OAAO,QAAQ;IACb,WAAW;IACX,aAAa,IAAI;IACjB,iBAAiB;GACnB,CAAC;GACD,KAAK;EACP;CACF;CACA,OAAO;AACT"}

package/dist/cjs/docReview/rebuildDocument.cjs

@@ -13,7 +13,8 @@ const identifySegmentsToReview = ({ baseBlocks, targetBlocks, plan }) => {
 	plan.actions.forEach((action, actionIndex) => {
 		if (action.kind === "review") {
 			const baseBlock = baseBlocks[action.baseIndex];
-			const targetBlockText = action.targetIndex !== null ? targetBlocks[action.targetIndex].content : null;
+			if (!baseBlock) return;
+			const targetBlockText = action.targetIndex !== null ? targetBlocks[action.targetIndex]?.content ?? null : null;
 			segmentsToReview.push({
 				baseBlock,
 				targetBlockText,
@@ -21,6 +22,7 @@ const identifySegmentsToReview = ({ baseBlocks, targetBlocks, plan }) => {
 			});
 		} else if (action.kind === "insert_new") {
 			const baseBlock = baseBlocks[action.baseIndex];
+			if (!baseBlock) return;
 			segmentsToReview.push({
 				baseBlock,
 				targetBlockText: null,

package/dist/cjs/docReview/rebuildDocument.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"rebuildDocument.cjs","names":[],"sources":["../../../src/docReview/rebuildDocument.ts"],"sourcesContent":["import type { AlignmentPlan, FingerprintedBlock } from './types';\n\n/**\n * A block that needs to be translated or re-translated by an external consumer\n * (an AI client, a human, or an agent).\n */\nexport type SegmentToReview = {\n  /** The base block to translate. */\n  baseBlock: FingerprintedBlock;\n  /** Existing target translation, or `null` when the block is new. */\n  targetBlockText: string | null;\n  /** Index of the originating action within {@link AlignmentPlan.actions}. */\n  actionIndex: number;\n};\n\nexport type RebuildInput = {\n  baseBlocks: FingerprintedBlock[];\n  targetBlocks: FingerprintedBlock[];\n  plan: AlignmentPlan;\n};\n\nexport type RebuildResult = {\n  segmentsToReview: SegmentToReview[];\n};\n\n/**\n * Analyze the alignment plan and return only the segments that need\n * review/translation. Does not generate output text - that is done by\n * {@link mergeReviewedSegments} once the translations are available.\n *\n * @param input - The base/target blocks and the alignment plan.\n * @returns The list of segments that require translation.\n */\nexport const identifySegmentsToReview = ({\n  baseBlocks,\n  targetBlocks,\n  plan,\n}: RebuildInput): RebuildResult => {\n  const segmentsToReview: SegmentToReview[] = [];\n\n  plan.actions.forEach((action, actionIndex) => {\n    if (action.kind === 'review') {\n      const baseBlock = baseBlocks[action.baseIndex];\n      const targetBlockText =\n        action.targetIndex !== null\n          ? targetBlocks[action.targetIndex].content\n          : null;\n\n      segmentsToReview.push({ baseBlock, targetBlockText, actionIndex });\n    } else if (action.kind === 'insert_new') {\n      const baseBlock = baseBlocks[action.baseIndex];\n\n      segmentsToReview.push({\n        baseBlock,\n        targetBlockText: null,\n        actionIndex,\n      });\n    }\n  });\n\n  return { segmentsToReview };\n};\n\n/** Markdown separates two blocks with a blank line. */\nconst endsWithBlockBoundary = (text: string): boolean =>\n  /\\n[ \\t]*\\n$/.test(text);\n\n/**\n * The newlines missing at the end of `text` for it to close a markdown block.\n *\n * @param text - The document built so far.\n * @returns `''`, `'\\n'` or `'\\n\\n'` depending on how the text already ends.\n */\nconst buildBlockSeparator = (text: string): string => {\n  if (text.length === 0 || endsWithBlockBoundary(text)) return '';\n\n  return text.endsWith('\\n') ? '\\n' : '\\n\\n';\n};\n\n/**\n * Merge reviewed translations back into the final document following the\n * alignment plan, reusing untouched target blocks as-is.\n *\n * @param plan - The alignment plan.\n * @param targetBlocks - Blocks of the existing target document.\n * @param reviewedSegments - Map of action index to its reviewed translation.\n * @returns The rebuilt target document.\n */\nexport const mergeReviewedSegments = (\n  plan: AlignmentPlan,\n  targetBlocks: FingerprintedBlock[],\n  reviewedSegments: Map<number, string>\n): string => {\n  let mergedText = '';\n  let previousPartWasGenerated = false;\n\n  /**\n   * Append one part, keeping a blank line around the content this run produced.\n   *\n   * Blocks own the blank lines that trail them, so two verbatim blocks already\n   * separate themselves and are appended untouched — a document with nothing to\n   * change merges back byte for byte. Generated content is the exception: the\n   * block it lands after may be the one that used to end the document, which\n   * carries no trailing blank line, and the two would be glued into a single\n   * markdown block (a heading swallowed by the paragraph above it).\n   */\n  const appendPart = (content: string, isGenerated: boolean): void => {\n    if (content.length === 0) return;\n\n    if (isGenerated || previousPartWasGenerated) {\n      mergedText += buildBlockSeparator(mergedText);\n    }\n\n    mergedText += content;\n    previousPartWasGenerated = isGenerated;\n  };\n\n  plan.actions.forEach((action, actionIndex) => {\n    if (action.kind === 'reuse') {\n      appendPart(targetBlocks[action.targetIndex]!.content, false);\n    } else if (action.kind === 'review' || action.kind === 'insert_new') {\n      const reviewedContent = reviewedSegments.get(actionIndex);\n\n      if (reviewedContent !== undefined) {\n        appendPart(reviewedContent, true);\n      } else {\n        // Fallback: if review failed, use existing or blank\n        if (action.kind === 'review' && action.targetIndex !== null) {\n          appendPart(targetBlocks[action.targetIndex]!.content, false);\n        } else {\n          appendPart('\\n', false);\n        }\n      }\n    } else if (action.kind === 'delete') {\n      const reviewedContent = reviewedSegments.get(actionIndex);\n      if (reviewedContent !== undefined) {\n        // Caller explicitly resolved this block: empty string = actually delete,\n        // non-empty string = replacement content.\n        appendPart(reviewedContent, true);\n      } else {\n        // Default: keep verbatim. A target block with no base counterpart may\n        // just be a section the aligner could not follow (reordering, split\n        // prose) — keeping it prevents accidental data loss in log/read-only mode.\n        appendPart(targetBlocks[action.targetIndex]!.content, false);\n      }\n    }\n  });\n\n  return mergedText;\n};\n"],"mappings":";;;;;;;;;;AAiCA,MAAa,4BAA4B,EACvC,YACA,cACA,WACiC;CACjC,MAAM,mBAAsC,CAAC;CAE7C,KAAK,QAAQ,SAAS,QAAQ,gBAAgB;EAC5C,IAAI,OAAO,SAAS,UAAU;GAC5B,MAAM,YAAY,WAAW,OAAO;GACpC,MAAM,kBACJ,OAAO,gBAAgB,OACnB,aAAa,OAAO,YAAY,CAAC,UACjC;GAEN,iBAAiB,KAAK;IAAE;IAAW;IAAiB;GAAY,CAAC;EACnE,OAAO,IAAI,OAAO,SAAS,cAAc;GACvC,MAAM,YAAY,WAAW,OAAO;GAEpC,iBAAiB,KAAK;IACpB;IACA,iBAAiB;IACjB;GACF,CAAC;EACH;CACF,CAAC;CAED,OAAO,EAAE,iBAAiB;AAC5B;;AAGA,MAAM,yBAAyB,SAC7B,cAAc,KAAK,IAAI;;;;;;;AAQzB,MAAM,uBAAuB,SAAyB;CACpD,IAAI,KAAK,WAAW,KAAK,sBAAsB,IAAI,GAAG,OAAO;CAE7D,OAAO,KAAK,SAAS,IAAI,IAAI,OAAO;AACtC;;;;;;;;;;AAWA,MAAa,yBACX,MACA,cACA,qBACW;CACX,IAAI,aAAa;CACjB,IAAI,2BAA2B;;;;;;;;;;;CAY/B,MAAM,cAAc,SAAiB,gBAA+B;EAClE,IAAI,QAAQ,WAAW,GAAG;EAE1B,IAAI,eAAe,0BACjB,cAAc,oBAAoB,UAAU;EAG9C,cAAc;EACd,2BAA2B;CAC7B;CAEA,KAAK,QAAQ,SAAS,QAAQ,gBAAgB;EAC5C,IAAI,OAAO,SAAS,SAClB,WAAW,aAAa,OAAO,YAAY,CAAE,SAAS,KAAK;OACtD,IAAI,OAAO,SAAS,YAAY,OAAO,SAAS,cAAc;GACnE,MAAM,kBAAkB,iBAAiB,IAAI,WAAW;GAExD,IAAI,oBAAoB,QACtB,WAAW,iBAAiB,IAAI;QAGhC,IAAI,OAAO,SAAS,YAAY,OAAO,gBAAgB,MACrD,WAAW,aAAa,OAAO,YAAY,CAAE,SAAS,KAAK;QAE3D,WAAW,MAAM,KAAK;EAG5B,OAAO,IAAI,OAAO,SAAS,UAAU;GACnC,MAAM,kBAAkB,iBAAiB,IAAI,WAAW;GACxD,IAAI,oBAAoB,QAGtB,WAAW,iBAAiB,IAAI;QAKhC,WAAW,aAAa,OAAO,YAAY,CAAE,SAAS,KAAK;EAE/D;CACF,CAAC;CAED,OAAO;AACT"}
+{"version":3,"file":"rebuildDocument.cjs","names":[],"sources":["../../../src/docReview/rebuildDocument.ts"],"sourcesContent":["import type { AlignmentPlan, FingerprintedBlock } from './types';\n\n/**\n * A block that needs to be translated or re-translated by an external consumer\n * (an AI client, a human, or an agent).\n */\nexport type SegmentToReview = {\n  /** The base block to translate. */\n  baseBlock: FingerprintedBlock;\n  /** Existing target translation, or `null` when the block is new. */\n  targetBlockText: string | null;\n  /** Index of the originating action within {@link AlignmentPlan.actions}. */\n  actionIndex: number;\n};\n\nexport type RebuildInput = {\n  baseBlocks: FingerprintedBlock[];\n  targetBlocks: FingerprintedBlock[];\n  plan: AlignmentPlan;\n};\n\nexport type RebuildResult = {\n  segmentsToReview: SegmentToReview[];\n};\n\n/**\n * Analyze the alignment plan and return only the segments that need\n * review/translation. Does not generate output text - that is done by\n * {@link mergeReviewedSegments} once the translations are available.\n *\n * @param input - The base/target blocks and the alignment plan.\n * @returns The list of segments that require translation.\n */\nexport const identifySegmentsToReview = ({\n  baseBlocks,\n  targetBlocks,\n  plan,\n}: RebuildInput): RebuildResult => {\n  const segmentsToReview: SegmentToReview[] = [];\n\n  plan.actions.forEach((action, actionIndex) => {\n    if (action.kind === 'review') {\n      const baseBlock = baseBlocks[action.baseIndex];\n\n      if (!baseBlock) return;\n\n      const targetBlockText =\n        action.targetIndex !== null\n          ? (targetBlocks[action.targetIndex]?.content ?? null)\n          : null;\n\n      segmentsToReview.push({ baseBlock, targetBlockText, actionIndex });\n    } else if (action.kind === 'insert_new') {\n      const baseBlock = baseBlocks[action.baseIndex];\n\n      if (!baseBlock) return;\n\n      segmentsToReview.push({\n        baseBlock,\n        targetBlockText: null,\n        actionIndex,\n      });\n    }\n  });\n\n  return { segmentsToReview };\n};\n\n/** Markdown separates two blocks with a blank line. */\nconst endsWithBlockBoundary = (text: string): boolean =>\n  /\\n[ \\t]*\\n$/.test(text);\n\n/**\n * The newlines missing at the end of `text` for it to close a markdown block.\n *\n * @param text - The document built so far.\n * @returns `''`, `'\\n'` or `'\\n\\n'` depending on how the text already ends.\n */\nconst buildBlockSeparator = (text: string): string => {\n  if (text.length === 0 || endsWithBlockBoundary(text)) return '';\n\n  return text.endsWith('\\n') ? '\\n' : '\\n\\n';\n};\n\n/**\n * Merge reviewed translations back into the final document following the\n * alignment plan, reusing untouched target blocks as-is.\n *\n * @param plan - The alignment plan.\n * @param targetBlocks - Blocks of the existing target document.\n * @param reviewedSegments - Map of action index to its reviewed translation.\n * @returns The rebuilt target document.\n */\nexport const mergeReviewedSegments = (\n  plan: AlignmentPlan,\n  targetBlocks: FingerprintedBlock[],\n  reviewedSegments: Map<number, string>\n): string => {\n  let mergedText = '';\n  let previousPartWasGenerated = false;\n\n  /**\n   * Append one part, keeping a blank line around the content this run produced.\n   *\n   * Blocks own the blank lines that trail them, so two verbatim blocks already\n   * separate themselves and are appended untouched — a document with nothing to\n   * change merges back byte for byte. Generated content is the exception: the\n   * block it lands after may be the one that used to end the document, which\n   * carries no trailing blank line, and the two would be glued into a single\n   * markdown block (a heading swallowed by the paragraph above it).\n   */\n  const appendPart = (content: string, isGenerated: boolean): void => {\n    if (content.length === 0) return;\n\n    if (isGenerated || previousPartWasGenerated) {\n      mergedText += buildBlockSeparator(mergedText);\n    }\n\n    mergedText += content;\n    previousPartWasGenerated = isGenerated;\n  };\n\n  plan.actions.forEach((action, actionIndex) => {\n    if (action.kind === 'reuse') {\n      appendPart(targetBlocks[action.targetIndex]!.content, false);\n    } else if (action.kind === 'review' || action.kind === 'insert_new') {\n      const reviewedContent = reviewedSegments.get(actionIndex);\n\n      if (reviewedContent !== undefined) {\n        appendPart(reviewedContent, true);\n      } else {\n        // Fallback: if review failed, use existing or blank\n        if (action.kind === 'review' && action.targetIndex !== null) {\n          appendPart(targetBlocks[action.targetIndex]!.content, false);\n        } else {\n          appendPart('\\n', false);\n        }\n      }\n    } else if (action.kind === 'delete') {\n      const reviewedContent = reviewedSegments.get(actionIndex);\n      if (reviewedContent !== undefined) {\n        // Caller explicitly resolved this block: empty string = actually delete,\n        // non-empty string = replacement content.\n        appendPart(reviewedContent, true);\n      } else {\n        // Default: keep verbatim. A target block with no base counterpart may\n        // just be a section the aligner could not follow (reordering, split\n        // prose) — keeping it prevents accidental data loss in log/read-only mode.\n        appendPart(targetBlocks[action.targetIndex]!.content, false);\n      }\n    }\n  });\n\n  return mergedText;\n};\n"],"mappings":";;;;;;;;;;AAiCA,MAAa,4BAA4B,EACvC,YACA,cACA,WACiC;CACjC,MAAM,mBAAsC,CAAC;CAE7C,KAAK,QAAQ,SAAS,QAAQ,gBAAgB;EAC5C,IAAI,OAAO,SAAS,UAAU;GAC5B,MAAM,YAAY,WAAW,OAAO;GAEpC,IAAI,CAAC,WAAW;GAEhB,MAAM,kBACJ,OAAO,gBAAgB,OAClB,aAAa,OAAO,YAAY,EAAE,WAAW,OAC9C;GAEN,iBAAiB,KAAK;IAAE;IAAW;IAAiB;GAAY,CAAC;EACnE,OAAO,IAAI,OAAO,SAAS,cAAc;GACvC,MAAM,YAAY,WAAW,OAAO;GAEpC,IAAI,CAAC,WAAW;GAEhB,iBAAiB,KAAK;IACpB;IACA,iBAAiB;IACjB;GACF,CAAC;EACH;CACF,CAAC;CAED,OAAO,EAAE,iBAAiB;AAC5B;;AAGA,MAAM,yBAAyB,SAC7B,cAAc,KAAK,IAAI;;;;;;;AAQzB,MAAM,uBAAuB,SAAyB;CACpD,IAAI,KAAK,WAAW,KAAK,sBAAsB,IAAI,GAAG,OAAO;CAE7D,OAAO,KAAK,SAAS,IAAI,IAAI,OAAO;AACtC;;;;;;;;;;AAWA,MAAa,yBACX,MACA,cACA,qBACW;CACX,IAAI,aAAa;CACjB,IAAI,2BAA2B;;;;;;;;;;;CAY/B,MAAM,cAAc,SAAiB,gBAA+B;EAClE,IAAI,QAAQ,WAAW,GAAG;EAE1B,IAAI,eAAe,0BACjB,cAAc,oBAAoB,UAAU;EAG9C,cAAc;EACd,2BAA2B;CAC7B;CAEA,KAAK,QAAQ,SAAS,QAAQ,gBAAgB;EAC5C,IAAI,OAAO,SAAS,SAClB,WAAW,aAAa,OAAO,YAAY,CAAE,SAAS,KAAK;OACtD,IAAI,OAAO,SAAS,YAAY,OAAO,SAAS,cAAc;GACnE,MAAM,kBAAkB,iBAAiB,IAAI,WAAW;GAExD,IAAI,oBAAoB,QACtB,WAAW,iBAAiB,IAAI;QAGhC,IAAI,OAAO,SAAS,YAAY,OAAO,gBAAgB,MACrD,WAAW,aAAa,OAAO,YAAY,CAAE,SAAS,KAAK;QAE3D,WAAW,MAAM,KAAK;EAG5B,OAAO,IAAI,OAAO,SAAS,UAAU;GACnC,MAAM,kBAAkB,iBAAiB,IAAI,WAAW;GACxD,IAAI,oBAAoB,QAGtB,WAAW,iBAAiB,IAAI;QAKhC,WAAW,aAAa,OAAO,YAAY,CAAE,SAAS,KAAK;EAE/D;CACF,CAAC;CAED,OAAO;AACT"}

package/dist/cjs/docReview/segmentDocument.cjs

@@ -1,8 +1,8 @@
 Object.defineProperty(exports, Symbol.toStringTag, { value: 'Module' });
 //#region src/docReview/segmentDocument.ts
 const HEADING_PATTERN = /^\s*(#{1,6})\s+/;
-const isBlankLine = (line) => line.trim().length === 0;
-const isFencedCodeDelimiter = (line) => /^\s*```/.test(line);
+const isBlankLine = (line) => !line || line.trim().length === 0;
+const isFencedCodeDelimiter = (line) => Boolean(line && /^\s*```/.test(line));
 /**
 * Read the depth of an ATX markdown heading (`#` → 1, `######` → 6).
 *
@@ -10,10 +10,11 @@ const isFencedCodeDelimiter = (line) => /^\s*```/.test(line);
 * @returns The heading depth, or `null` when the line is not a heading.
 */
 const parseHeadingDepth = (line) => {
+	if (!line) return null;
 	return HEADING_PATTERN.exec(line)?.[1]?.length ?? null;
 };
 const isHeading = (line) => parseHeadingDepth(line) !== null;
-const isFrontmatterDelimiter = (line) => /^\s*---\s*$/.test(line);
+const isFrontmatterDelimiter = (line) => Boolean(line && /^\s*---\s*$/.test(line));
 /**
 * Split a markdown document into fine-grained blocks.
 *
@@ -88,7 +89,8 @@ const segmentDocument = (text) => {
 	if (units.length === 0) return [];
 	return units.map((unit, unitIndex) => {
 		const blockStartIndex = unitIndex === 0 ? 0 : unit.startIndex;
-		const blockEndIndex = unitIndex === units.length - 1 ? lineCount - 1 : units[unitIndex + 1].startIndex - 1;
+		const nextUnit = units[unitIndex + 1];
+		const blockEndIndex = unitIndex === units.length - 1 || !nextUnit ? lineCount - 1 : nextUnit.startIndex - 1;
 		const blockLines = lines.slice(blockStartIndex, blockEndIndex + 1);
 		const content = blockEndIndex < lineCount - 1 ? `${blockLines.join("\n")}\n` : blockLines.join("\n");
 		return {
@@ -124,8 +126,9 @@ const segmentSections = (text) => {
 	let currentBlocks = [];
 	const flushSection = () => {
 		if (currentBlocks.length === 0) return;
-		const [firstBlock] = currentBlocks;
+		const firstBlock = currentBlocks[0];
 		const lastBlock = currentBlocks[currentBlocks.length - 1];
+		if (!firstBlock || !lastBlock) return;
 		sections.push({
 			type: firstBlock.type,
 			content: currentBlocks.map((block) => block.content).join(""),

package/dist/cjs/docReview/segmentDocument.cjs.map

@@ -1 +1 @@
-{"version":3,"file":"segmentDocument.cjs","names":[],"sources":["../../../src/docReview/segmentDocument.ts"],"sourcesContent":["import type { Block, BlockType } from './types';\n\nconst HEADING_PATTERN = /^\\s*(#{1,6})\\s+/;\n\nconst isBlankLine = (line: string): boolean => line.trim().length === 0;\nconst isFencedCodeDelimiter = (line: string): boolean => /^\\s*```/.test(line);\n\n/**\n * Read the depth of an ATX markdown heading (`#` → 1, `######` → 6).\n *\n * @param line - The line to inspect.\n * @returns The heading depth, or `null` when the line is not a heading.\n */\nconst parseHeadingDepth = (line: string): number | null => {\n  const match = HEADING_PATTERN.exec(line);\n\n  return match?.[1]?.length ?? null;\n};\n\nconst isHeading = (line: string): boolean => parseHeadingDepth(line) !== null;\nconst isFrontmatterDelimiter = (line: string): boolean =>\n  /^\\s*---\\s*$/.test(line);\n\n/**\n * A content unit (heading, paragraph, code block or frontmatter) spanning a\n * 0-based, inclusive line range. Blank-line runs are not units of their own;\n * they are folded into the preceding unit as trailing separators (see\n * {@link segmentDocument}) so that the concatenation of every block's content\n * reproduces the document byte-for-byte.\n */\ntype ContentUnit = {\n  type: BlockType;\n  /** Depth of the ATX heading opening the unit, `null` when it is not a heading. */\n  headingDepth: number | null;\n  startIndex: number;\n  endIndex: number;\n};\n\n/**\n * Split a markdown document into fine-grained blocks.\n *\n * Boundaries are drawn at headings, blank lines (paragraph breaks) and fenced\n * code blocks, while frontmatter and the inside of code fences are kept intact.\n * Each blank-line run is appended to the block that precedes it, so the blocks\n * form an exact partition of the document: concatenating every `content` (in\n * order) yields the original text unchanged. This is what lets the block-aware\n * review re-translate only the paragraphs/snippets that actually changed instead\n * of the whole heading section.\n *\n * @param text - The full markdown document.\n * @returns The ordered list of blocks with their 1-based line ranges.\n */\nexport const segmentDocument = (text: string): Block[] => {\n  const lines = text.split('\\n');\n  const lineCount = lines.length;\n\n  // 1. Tokenize into content units, skipping blank-line runs (folded in below).\n  const units: ContentUnit[] = [];\n  let index = 0;\n\n  while (index < lineCount) {\n    const currentLine = lines[index];\n\n    if (isBlankLine(currentLine)) {\n      index += 1;\n      continue;\n    }\n\n    // Frontmatter: only when it opens the document.\n    if (units.length === 0 && isFrontmatterDelimiter(currentLine)) {\n      const startIndex = index;\n      index += 1;\n      while (index < lineCount && !isFrontmatterDelimiter(lines[index])) {\n        index += 1;\n      }\n      // Include the closing delimiter when present.\n      if (index < lineCount) index += 1;\n      units.push({\n        type: 'unknown',\n        headingDepth: null,\n        startIndex,\n        endIndex: index - 1,\n      });\n      continue;\n    }\n\n    // Fenced code block: consumed whole so inner blank lines and `#` lines are\n    // never treated as boundaries.\n    if (isFencedCodeDelimiter(currentLine)) {\n      const startIndex = index;\n      index += 1;\n      while (index < lineCount && !isFencedCodeDelimiter(lines[index])) {\n        index += 1;\n      }\n      // Include the closing fence when present.\n      if (index < lineCount) index += 1;\n      units.push({\n        type: 'code_block',\n        headingDepth: null,\n        startIndex,\n        endIndex: index - 1,\n      });\n      continue;\n    }\n\n    // Heading: a single self-contained line.\n    const headingDepth = parseHeadingDepth(currentLine);\n\n    if (headingDepth !== null) {\n      units.push({\n        type: 'heading',\n        headingDepth,\n        startIndex: index,\n        endIndex: index,\n      });\n      index += 1;\n      continue;\n    }\n\n    // Paragraph: a run of consecutive lines until a blank line, a heading or a\n    // code fence. Tables and tight lists stay together (no blank line between\n    // their rows/items).\n    const startIndex = index;\n    while (\n      index < lineCount &&\n      !isBlankLine(lines[index]) &&\n      !isHeading(lines[index]) &&\n      !isFencedCodeDelimiter(lines[index])\n    ) {\n      index += 1;\n    }\n    units.push({\n      type: 'paragraph',\n      headingDepth: null,\n      startIndex,\n      endIndex: index - 1,\n    });\n  }\n\n  if (units.length === 0) return [];\n\n  // 2. Turn each unit into a block whose line range extends to just before the\n  //    next unit, so the trailing blank-line run is owned by it. The first block\n  //    also absorbs any leading blank lines, and the last block runs to EOF.\n  return units.map((unit, unitIndex): Block => {\n    const blockStartIndex = unitIndex === 0 ? 0 : unit.startIndex;\n    const blockEndIndex =\n      unitIndex === units.length - 1\n        ? lineCount - 1\n        : units[unitIndex + 1].startIndex - 1;\n\n    const blockLines = lines.slice(blockStartIndex, blockEndIndex + 1);\n    // Re-append the boundary newline dropped by `split` for every block but the\n    // one ending at EOF, so concatenating all blocks rebuilds the document.\n    const content =\n      blockEndIndex < lineCount - 1\n        ? `${blockLines.join('\\n')}\\n`\n        : blockLines.join('\\n');\n\n    return {\n      type: unit.type,\n      content,\n      headingDepth: unit.headingDepth,\n      lineStart: blockStartIndex + 1,\n      lineEnd: blockEndIndex + 1,\n    };\n  });\n};\n\n/**\n * Split a markdown document into coarse, heading-anchored sections.\n *\n * Built by grouping the fine blocks of {@link segmentDocument}: frontmatter and\n * each heading open a new section, and the following non-heading blocks are\n * folded into it. Because it only concatenates adjacent fine blocks, the result\n * is still an exact partition of the document (sections concatenate back to the\n * source unchanged).\n *\n * Sections are the robust alignment unit between a base document and its\n * translation — both share the same heading structure, so they align almost\n * perfectly and a translation that splits its prose into a different number of\n * paragraphs never causes a section to be dropped. Fine-grained review happens\n * within a section once it is known to have changed.\n *\n * @param text - The full markdown document.\n * @returns The ordered list of sections with their 1-based line ranges.\n */\nexport const segmentSections = (text: string): Block[] => {\n  const fineBlocks = segmentDocument(text);\n  const sections: Block[] = [];\n  let currentBlocks: Block[] = [];\n\n  const flushSection = (): void => {\n    if (currentBlocks.length === 0) return;\n\n    const [firstBlock] = currentBlocks;\n    const lastBlock = currentBlocks[currentBlocks.length - 1];\n\n    sections.push({\n      type: firstBlock.type,\n      content: currentBlocks.map((block) => block.content).join(''),\n      // A section is identified by the heading that opens it, so it inherits its\n      // depth — this is what keeps a `##` section from aligning with a `###` one.\n      headingDepth: firstBlock.headingDepth,\n      lineStart: firstBlock.lineStart,\n      lineEnd: lastBlock.lineEnd,\n    });\n    currentBlocks = [];\n  };\n\n  for (const block of fineBlocks) {\n    // Frontmatter (a leading `unknown` block) and every heading open a section.\n    const opensSection =\n      block.type === 'heading' ||\n      (block.type === 'unknown' && sections.length === 0);\n\n    if (opensSection) flushSection();\n    currentBlocks.push(block);\n    if (block.type === 'unknown' && sections.length === 0) flushSection();\n  }\n\n  flushSection();\n\n  return sections;\n};\n"],"mappings":";;AAEA,MAAM,kBAAkB;AAExB,MAAM,eAAe,SAA0B,KAAK,KAAK,CAAC,CAAC,WAAW;AACtE,MAAM,yBAAyB,SAA0B,UAAU,KAAK,IAAI;;;;;;;AAQ5E,MAAM,qBAAqB,SAAgC;CAGzD,OAFc,gBAAgB,KAAK,IAExB,CAAC,GAAG,EAAE,EAAE,UAAU;AAC/B;AAEA,MAAM,aAAa,SAA0B,kBAAkB,IAAI,MAAM;AACzE,MAAM,0BAA0B,SAC9B,cAAc,KAAK,IAAI;;;;;;;;;;;;;;;AA+BzB,MAAa,mBAAmB,SAA0B;CACxD,MAAM,QAAQ,KAAK,MAAM,IAAI;CAC7B,MAAM,YAAY,MAAM;CAGxB,MAAM,QAAuB,CAAC;CAC9B,IAAI,QAAQ;CAEZ,OAAO,QAAQ,WAAW;EACxB,MAAM,cAAc,MAAM;EAE1B,IAAI,YAAY,WAAW,GAAG;GAC5B,SAAS;GACT;EACF;EAGA,IAAI,MAAM,WAAW,KAAK,uBAAuB,WAAW,GAAG;GAC7D,MAAM,aAAa;GACnB,SAAS;GACT,OAAO,QAAQ,aAAa,CAAC,uBAAuB,MAAM,MAAM,GAC9D,SAAS;GAGX,IAAI,QAAQ,WAAW,SAAS;GAChC,MAAM,KAAK;IACT,MAAM;IACN,cAAc;IACd;IACA,UAAU,QAAQ;GACpB,CAAC;GACD;EACF;EAIA,IAAI,sBAAsB,WAAW,GAAG;GACtC,MAAM,aAAa;GACnB,SAAS;GACT,OAAO,QAAQ,aAAa,CAAC,sBAAsB,MAAM,MAAM,GAC7D,SAAS;GAGX,IAAI,QAAQ,WAAW,SAAS;GAChC,MAAM,KAAK;IACT,MAAM;IACN,cAAc;IACd;IACA,UAAU,QAAQ;GACpB,CAAC;GACD;EACF;EAGA,MAAM,eAAe,kBAAkB,WAAW;EAElD,IAAI,iBAAiB,MAAM;GACzB,MAAM,KAAK;IACT,MAAM;IACN;IACA,YAAY;IACZ,UAAU;GACZ,CAAC;GACD,SAAS;GACT;EACF;EAKA,MAAM,aAAa;EACnB,OACE,QAAQ,aACR,CAAC,YAAY,MAAM,MAAM,KACzB,CAAC,UAAU,MAAM,MAAM,KACvB,CAAC,sBAAsB,MAAM,MAAM,GAEnC,SAAS;EAEX,MAAM,KAAK;GACT,MAAM;GACN,cAAc;GACd;GACA,UAAU,QAAQ;EACpB,CAAC;CACH;CAEA,IAAI,MAAM,WAAW,GAAG,OAAO,CAAC;CAKhC,OAAO,MAAM,KAAK,MAAM,cAAqB;EAC3C,MAAM,kBAAkB,cAAc,IAAI,IAAI,KAAK;EACnD,MAAM,gBACJ,cAAc,MAAM,SAAS,IACzB,YAAY,IACZ,MAAM,YAAY,EAAE,CAAC,aAAa;EAExC,MAAM,aAAa,MAAM,MAAM,iBAAiB,gBAAgB,CAAC;EAGjE,MAAM,UACJ,gBAAgB,YAAY,IACxB,GAAG,WAAW,KAAK,IAAI,EAAE,MACzB,WAAW,KAAK,IAAI;EAE1B,OAAO;GACL,MAAM,KAAK;GACX;GACA,cAAc,KAAK;GACnB,WAAW,kBAAkB;GAC7B,SAAS,gBAAgB;EAC3B;CACF,CAAC;AACH;;;;;;;;;;;;;;;;;;;AAoBA,MAAa,mBAAmB,SAA0B;CACxD,MAAM,aAAa,gBAAgB,IAAI;CACvC,MAAM,WAAoB,CAAC;CAC3B,IAAI,gBAAyB,CAAC;CAE9B,MAAM,qBAA2B;EAC/B,IAAI,cAAc,WAAW,GAAG;EAEhC,MAAM,CAAC,cAAc;EACrB,MAAM,YAAY,cAAc,cAAc,SAAS;EAEvD,SAAS,KAAK;GACZ,MAAM,WAAW;GACjB,SAAS,cAAc,KAAK,UAAU,MAAM,OAAO,CAAC,CAAC,KAAK,EAAE;GAG5D,cAAc,WAAW;GACzB,WAAW,WAAW;GACtB,SAAS,UAAU;EACrB,CAAC;EACD,gBAAgB,CAAC;CACnB;CAEA,KAAK,MAAM,SAAS,YAAY;EAM9B,IAHE,MAAM,SAAS,aACd,MAAM,SAAS,aAAa,SAAS,WAAW,GAEjC,aAAa;EAC/B,cAAc,KAAK,KAAK;EACxB,IAAI,MAAM,SAAS,aAAa,SAAS,WAAW,GAAG,aAAa;CACtE;CAEA,aAAa;CAEb,OAAO;AACT"}
+{"version":3,"file":"segmentDocument.cjs","names":[],"sources":["../../../src/docReview/segmentDocument.ts"],"sourcesContent":["import type { Block, BlockType } from './types';\n\nconst HEADING_PATTERN = /^\\s*(#{1,6})\\s+/;\n\nconst isBlankLine = (line?: string): boolean =>\n  !line || line.trim().length === 0;\nconst isFencedCodeDelimiter = (line?: string): boolean =>\n  Boolean(line && /^\\s*```/.test(line));\n\n/**\n * Read the depth of an ATX markdown heading (`#` → 1, `######` → 6).\n *\n * @param line - The line to inspect.\n * @returns The heading depth, or `null` when the line is not a heading.\n */\nconst parseHeadingDepth = (line?: string): number | null => {\n  if (!line) return null;\n  const match = HEADING_PATTERN.exec(line);\n\n  return match?.[1]?.length ?? null;\n};\n\nconst isHeading = (line?: string): boolean => parseHeadingDepth(line) !== null;\nconst isFrontmatterDelimiter = (line?: string): boolean =>\n  Boolean(line && /^\\s*---\\s*$/.test(line));\n\n/**\n * A content unit (heading, paragraph, code block or frontmatter) spanning a\n * 0-based, inclusive line range. Blank-line runs are not units of their own;\n * they are folded into the preceding unit as trailing separators (see\n * {@link segmentDocument}) so that the concatenation of every block's content\n * reproduces the document byte-for-byte.\n */\ntype ContentUnit = {\n  type: BlockType;\n  /** Depth of the ATX heading opening the unit, `null` when it is not a heading. */\n  headingDepth: number | null;\n  startIndex: number;\n  endIndex: number;\n};\n\n/**\n * Split a markdown document into fine-grained blocks.\n *\n * Boundaries are drawn at headings, blank lines (paragraph breaks) and fenced\n * code blocks, while frontmatter and the inside of code fences are kept intact.\n * Each blank-line run is appended to the block that precedes it, so the blocks\n * form an exact partition of the document: concatenating every `content` (in\n * order) yields the original text unchanged. This is what lets the block-aware\n * review re-translate only the paragraphs/snippets that actually changed instead\n * of the whole heading section.\n *\n * @param text - The full markdown document.\n * @returns The ordered list of blocks with their 1-based line ranges.\n */\nexport const segmentDocument = (text: string): Block[] => {\n  const lines = text.split('\\n');\n  const lineCount = lines.length;\n\n  // 1. Tokenize into content units, skipping blank-line runs (folded in below).\n  const units: ContentUnit[] = [];\n  let index = 0;\n\n  while (index < lineCount) {\n    const currentLine = lines[index];\n\n    if (isBlankLine(currentLine)) {\n      index += 1;\n      continue;\n    }\n\n    // Frontmatter: only when it opens the document.\n    if (units.length === 0 && isFrontmatterDelimiter(currentLine)) {\n      const startIndex = index;\n      index += 1;\n      while (index < lineCount && !isFrontmatterDelimiter(lines[index])) {\n        index += 1;\n      }\n      // Include the closing delimiter when present.\n      if (index < lineCount) index += 1;\n      units.push({\n        type: 'unknown',\n        headingDepth: null,\n        startIndex,\n        endIndex: index - 1,\n      });\n      continue;\n    }\n\n    // Fenced code block: consumed whole so inner blank lines and `#` lines are\n    // never treated as boundaries.\n    if (isFencedCodeDelimiter(currentLine)) {\n      const startIndex = index;\n      index += 1;\n      while (index < lineCount && !isFencedCodeDelimiter(lines[index])) {\n        index += 1;\n      }\n      // Include the closing fence when present.\n      if (index < lineCount) index += 1;\n      units.push({\n        type: 'code_block',\n        headingDepth: null,\n        startIndex,\n        endIndex: index - 1,\n      });\n      continue;\n    }\n\n    // Heading: a single self-contained line.\n    const headingDepth = parseHeadingDepth(currentLine);\n\n    if (headingDepth !== null) {\n      units.push({\n        type: 'heading',\n        headingDepth,\n        startIndex: index,\n        endIndex: index,\n      });\n      index += 1;\n      continue;\n    }\n\n    // Paragraph: a run of consecutive lines until a blank line, a heading or a\n    // code fence. Tables and tight lists stay together (no blank line between\n    // their rows/items).\n    const startIndex = index;\n    while (\n      index < lineCount &&\n      !isBlankLine(lines[index]) &&\n      !isHeading(lines[index]) &&\n      !isFencedCodeDelimiter(lines[index])\n    ) {\n      index += 1;\n    }\n    units.push({\n      type: 'paragraph',\n      headingDepth: null,\n      startIndex,\n      endIndex: index - 1,\n    });\n  }\n\n  if (units.length === 0) return [];\n\n  // 2. Turn each unit into a block whose line range extends to just before the\n  //    next unit, so the trailing blank-line run is owned by it. The first block\n  //    also absorbs any leading blank lines, and the last block runs to EOF.\n  return units.map((unit, unitIndex): Block => {\n    const blockStartIndex = unitIndex === 0 ? 0 : unit.startIndex;\n    const nextUnit = units[unitIndex + 1];\n    const blockEndIndex =\n      unitIndex === units.length - 1 || !nextUnit\n        ? lineCount - 1\n        : nextUnit.startIndex - 1;\n\n    const blockLines = lines.slice(blockStartIndex, blockEndIndex + 1);\n    // Re-append the boundary newline dropped by `split` for every block but the\n    // one ending at EOF, so concatenating all blocks rebuilds the document.\n    const content =\n      blockEndIndex < lineCount - 1\n        ? `${blockLines.join('\\n')}\\n`\n        : blockLines.join('\\n');\n\n    return {\n      type: unit.type,\n      content,\n      headingDepth: unit.headingDepth,\n      lineStart: blockStartIndex + 1,\n      lineEnd: blockEndIndex + 1,\n    };\n  });\n};\n\n/**\n * Split a markdown document into coarse, heading-anchored sections.\n *\n * Built by grouping the fine blocks of {@link segmentDocument}: frontmatter and\n * each heading open a new section, and the following non-heading blocks are\n * folded into it. Because it only concatenates adjacent fine blocks, the result\n * is still an exact partition of the document (sections concatenate back to the\n * source unchanged).\n *\n * Sections are the robust alignment unit between a base document and its\n * translation — both share the same heading structure, so they align almost\n * perfectly and a translation that splits its prose into a different number of\n * paragraphs never causes a section to be dropped. Fine-grained review happens\n * within a section once it is known to have changed.\n *\n * @param text - The full markdown document.\n * @returns The ordered list of sections with their 1-based line ranges.\n */\nexport const segmentSections = (text: string): Block[] => {\n  const fineBlocks = segmentDocument(text);\n  const sections: Block[] = [];\n  let currentBlocks: Block[] = [];\n\n  const flushSection = (): void => {\n    if (currentBlocks.length === 0) return;\n\n    const firstBlock = currentBlocks[0];\n    const lastBlock = currentBlocks[currentBlocks.length - 1];\n\n    if (!firstBlock || !lastBlock) return;\n\n    sections.push({\n      type: firstBlock.type,\n      content: currentBlocks.map((block) => block.content).join(''),\n      // A section is identified by the heading that opens it, so it inherits its\n      // depth — this is what keeps a `##` section from aligning with a `###` one.\n      headingDepth: firstBlock.headingDepth,\n      lineStart: firstBlock.lineStart,\n      lineEnd: lastBlock.lineEnd,\n    });\n    currentBlocks = [];\n  };\n\n  for (const block of fineBlocks) {\n    // Frontmatter (a leading `unknown` block) and every heading open a section.\n    const opensSection =\n      block.type === 'heading' ||\n      (block.type === 'unknown' && sections.length === 0);\n\n    if (opensSection) flushSection();\n    currentBlocks.push(block);\n    if (block.type === 'unknown' && sections.length === 0) flushSection();\n  }\n\n  flushSection();\n\n  return sections;\n};\n"],"mappings":";;AAEA,MAAM,kBAAkB;AAExB,MAAM,eAAe,SACnB,CAAC,QAAQ,KAAK,KAAK,CAAC,CAAC,WAAW;AAClC,MAAM,yBAAyB,SAC7B,QAAQ,QAAQ,UAAU,KAAK,IAAI,CAAC;;;;;;;AAQtC,MAAM,qBAAqB,SAAiC;CAC1D,IAAI,CAAC,MAAM,OAAO;CAGlB,OAFc,gBAAgB,KAAK,IAExB,CAAC,GAAG,EAAE,EAAE,UAAU;AAC/B;AAEA,MAAM,aAAa,SAA2B,kBAAkB,IAAI,MAAM;AAC1E,MAAM,0BAA0B,SAC9B,QAAQ,QAAQ,cAAc,KAAK,IAAI,CAAC;;;;;;;;;;;;;;;AA+B1C,MAAa,mBAAmB,SAA0B;CACxD,MAAM,QAAQ,KAAK,MAAM,IAAI;CAC7B,MAAM,YAAY,MAAM;CAGxB,MAAM,QAAuB,CAAC;CAC9B,IAAI,QAAQ;CAEZ,OAAO,QAAQ,WAAW;EACxB,MAAM,cAAc,MAAM;EAE1B,IAAI,YAAY,WAAW,GAAG;GAC5B,SAAS;GACT;EACF;EAGA,IAAI,MAAM,WAAW,KAAK,uBAAuB,WAAW,GAAG;GAC7D,MAAM,aAAa;GACnB,SAAS;GACT,OAAO,QAAQ,aAAa,CAAC,uBAAuB,MAAM,MAAM,GAC9D,SAAS;GAGX,IAAI,QAAQ,WAAW,SAAS;GAChC,MAAM,KAAK;IACT,MAAM;IACN,cAAc;IACd;IACA,UAAU,QAAQ;GACpB,CAAC;GACD;EACF;EAIA,IAAI,sBAAsB,WAAW,GAAG;GACtC,MAAM,aAAa;GACnB,SAAS;GACT,OAAO,QAAQ,aAAa,CAAC,sBAAsB,MAAM,MAAM,GAC7D,SAAS;GAGX,IAAI,QAAQ,WAAW,SAAS;GAChC,MAAM,KAAK;IACT,MAAM;IACN,cAAc;IACd;IACA,UAAU,QAAQ;GACpB,CAAC;GACD;EACF;EAGA,MAAM,eAAe,kBAAkB,WAAW;EAElD,IAAI,iBAAiB,MAAM;GACzB,MAAM,KAAK;IACT,MAAM;IACN;IACA,YAAY;IACZ,UAAU;GACZ,CAAC;GACD,SAAS;GACT;EACF;EAKA,MAAM,aAAa;EACnB,OACE,QAAQ,aACR,CAAC,YAAY,MAAM,MAAM,KACzB,CAAC,UAAU,MAAM,MAAM,KACvB,CAAC,sBAAsB,MAAM,MAAM,GAEnC,SAAS;EAEX,MAAM,KAAK;GACT,MAAM;GACN,cAAc;GACd;GACA,UAAU,QAAQ;EACpB,CAAC;CACH;CAEA,IAAI,MAAM,WAAW,GAAG,OAAO,CAAC;CAKhC,OAAO,MAAM,KAAK,MAAM,cAAqB;EAC3C,MAAM,kBAAkB,cAAc,IAAI,IAAI,KAAK;EACnD,MAAM,WAAW,MAAM,YAAY;EACnC,MAAM,gBACJ,cAAc,MAAM,SAAS,KAAK,CAAC,WAC/B,YAAY,IACZ,SAAS,aAAa;EAE5B,MAAM,aAAa,MAAM,MAAM,iBAAiB,gBAAgB,CAAC;EAGjE,MAAM,UACJ,gBAAgB,YAAY,IACxB,GAAG,WAAW,KAAK,IAAI,EAAE,MACzB,WAAW,KAAK,IAAI;EAE1B,OAAO;GACL,MAAM,KAAK;GACX;GACA,cAAc,KAAK;GACnB,WAAW,kBAAkB;GAC7B,SAAS,gBAAgB;EAC3B;CACF,CAAC;AACH;;;;;;;;;;;;;;;;;;;AAoBA,MAAa,mBAAmB,SAA0B;CACxD,MAAM,aAAa,gBAAgB,IAAI;CACvC,MAAM,WAAoB,CAAC;CAC3B,IAAI,gBAAyB,CAAC;CAE9B,MAAM,qBAA2B;EAC/B,IAAI,cAAc,WAAW,GAAG;EAEhC,MAAM,aAAa,cAAc;EACjC,MAAM,YAAY,cAAc,cAAc,SAAS;EAEvD,IAAI,CAAC,cAAc,CAAC,WAAW;EAE/B,SAAS,KAAK;GACZ,MAAM,WAAW;GACjB,SAAS,cAAc,KAAK,UAAU,MAAM,OAAO,CAAC,CAAC,KAAK,EAAE;GAG5D,cAAc,WAAW;GACzB,WAAW,WAAW;GACtB,SAAS,UAAU;EACrB,CAAC;EACD,gBAAgB,CAAC;CACnB;CAEA,KAAK,MAAM,SAAS,YAAY;EAM9B,IAHE,MAAM,SAAS,aACd,MAAM,SAAS,aAAa,SAAS,WAAW,GAEjC,aAAa;EAC/B,cAAc,KAAK,KAAK;EACxB,IAAI,MAAM,SAAS,aAAa,SAAS,WAAW,GAAG,aAAa;CACtE;CAEA,aAAa;CAEb,OAAO;AACT"}

package/dist/cjs/init/frameworkSetup/nextAppRouter/transforms.cjs

@@ -74,7 +74,10 @@ const setHtmlLang = (ast) => {
 			const langAttr = node.attributes?.find((attr) => attr.type === "JSXAttribute" && attr.name?.name === "lang");
 			const localeExpression = b.jsxExpressionContainer(b.identifier("locale"));
 			if (langAttr) langAttr.value = localeExpression;
-			else node.attributes.push(b.jsxAttribute(b.jsxIdentifier("lang"), localeExpression));
+			else {
+				node.attributes = node.attributes ?? [];
+				node.attributes.push(b.jsxAttribute(b.jsxIdentifier("lang"), localeExpression));
+			}
 			return false;
 		}
 		this.traverse(path);