euparliamentmonitor 0.9.28 → 1.0.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "euparliamentmonitor",
-  "version": "0.9.28",
+  "version": "1.0.0",
   "type": "module",
   "description": "European Parliament Intelligence Platform - Monitor political activity with systematic transparency",
   "main": "scripts/index.js",
@@ -167,7 +167,7 @@
     "clean-css": "^5.3.3",
     "d3": "7.9.0",
     "esbuild": "0.28.0",
-    "eslint": "10.4.0",
+    "eslint": "10.4.1",
     "eslint-config-prettier": "10.1.8",
     "eslint-plugin-jsdoc": "63.0.0",
     "eslint-plugin-security": "4.0.0",
@@ -179,7 +179,7 @@
     "husky": "9.1.7",
     "jscpd": "4.2.4",
     "knip": "^6.7.0",
-    "lint-staged": "17.0.5",
+    "lint-staged": "17.0.6",
     "mermaid": "11.15.0",
     "papaparse": "5.5.3",
     "prettier": "3.8.3",

package/scripts/aggregator/html/localize-body.d.ts

@@ -22,11 +22,32 @@ export declare function localizeArticleBody(bodyHtml: string, lang: LanguageCode
  * @returns Modified string, or `haystack` unchanged when `needle` is absent
  */
 export declare function replaceFirstStringIn(haystack: string, needle: string, replacement: string): string;
+/**
+ * Locate the cut point that ends the Executive Brief body — the start of
+ * the next top-level boundary heading after `afterHeading`. A boundary is
+ * any `<h2>` whose `id` either starts with the canonical `section-` prefix
+ * or exactly matches one of {@link EXECUTIVE_BRIEF_BOUNDARY_ID_MARKERS}
+ * (Reader Guide / Tradecraft / Analysis Index / Supplementary appendices).
+ *
+ * Critically, this only matches **top-level** section anchors — never the
+ * brief's own internal `<h2>` sub-headings (`## BLUF`, `## 60-Second Read`,
+ * …), which carry slugified ids without the `section-` prefix. That is why
+ * we cannot simply look for the next `<h2`.
+ *
+ * Uses `indexOf`/`lastIndexOf` exclusively (no regex) to stay within
+ * CodeQL's safe-regex envelope.
+ *
+ * @param html - Full article body HTML
+ * @param afterHeading - Index immediately after the Executive Brief `</h2>`
+ * @returns Index of the next boundary `<h2`, or `-1` when the Executive
+ *          Brief is the last block in the body.
+ */
+export declare function findExecutiveBriefSectionCut(html: string, afterHeading: number): number;
 /**
  * Replace the **inner body** of the Executive Brief section (the
  * `<h2 id="section-executive-brief">…</h2>` heading and everything that
- * follows it up to — but not including — the next `<h2 id="section-…">`
- * sibling) with the supplied replacement HTML. The Executive Brief
+ * follows it up to — but not including — the next top-level boundary
+ * heading) with the supplied replacement HTML. The Executive Brief
  * heading itself is preserved by emitting it inline ahead of the
  * replacement, so the in-page anchor (`#section-executive-brief`) and
  * the table-of-contents link continue to work.
@@ -39,8 +60,11 @@ export declare function replaceFirstStringIn(haystack: string, needle: string, r
  * `render-one.writeLanguageVariant`.
  *
  * Implementation uses `indexOf`/slice exclusively to stay within
- * CodeQL's safe-regex envelope. Returns `html` unchanged when the
- * Executive Brief heading is absent or malformed.
+ * CodeQL's safe-regex envelope. The replacement spans from the heading to
+ * the next top-level boundary (see {@link findExecutiveBriefSectionCut});
+ * when the Executive Brief is the last block in the body the replacement
+ * extends to end-of-body. Returns `html` unchanged only when the Executive
+ * Brief heading is absent or malformed.
  *
  * @param html - Full article body HTML
  * @param localizedHeading - Localized text for the Executive Brief H2

package/scripts/aggregator/html/localize-body.js

@@ -12,6 +12,23 @@ import { TRADECRAFT_HEADING_LABELS, TRADECRAFT_INTRO_LABELS, TRADECRAFT_METHODOL
 import { escapeHTML } from '../../utils/file-utils.js';
 import { TRADECRAFT_SECTION_ID, MANIFEST_SECTION_ID, SUPPLEMENTARY_SECTION_ID, } from '../artifact-order.js';
 import { KEY_TAKEAWAYS_SECTION_ID } from '../key-takeaways.js';
+import { READER_GUIDE_SECTION_ID } from '../reader-guide-constants.js';
+/**
+ * Top-level section anchors that mark the **end** of the Executive Brief
+ * body. Canonical analysis sections are matched by the shared
+ * `id="section-…"` prefix (see {@link findExecutiveBriefSectionCut});
+ * the appendix and reader-guide sections below carry bespoke ids that do
+ * **not** share that prefix, so they are matched explicitly. Including
+ * them ensures the localized brief splice also fires on sparse runs where
+ * the Executive Brief is the last canonical section and only appendix
+ * blocks follow it.
+ */
+const EXECUTIVE_BRIEF_BOUNDARY_ID_MARKERS = [
+    `id="${READER_GUIDE_SECTION_ID}"`,
+    `id="${TRADECRAFT_SECTION_ID}"`,
+    `id="${MANIFEST_SECTION_ID}"`,
+    `id="${SUPPLEMENTARY_SECTION_ID}"`,
+];
 /**
  * Localize the Tradecraft References and Analysis Index sections in the
  * rendered article body HTML. Replaces English headings, introductions,
@@ -102,11 +119,48 @@ export function replaceFirstStringIn(haystack, needle, replacement) {
         return haystack;
     return haystack.slice(0, idx) + replacement + haystack.slice(idx + needle.length);
 }
+/**
+ * Locate the cut point that ends the Executive Brief body — the start of
+ * the next top-level boundary heading after `afterHeading`. A boundary is
+ * any `<h2>` whose `id` either starts with the canonical `section-` prefix
+ * or exactly matches one of {@link EXECUTIVE_BRIEF_BOUNDARY_ID_MARKERS}
+ * (Reader Guide / Tradecraft / Analysis Index / Supplementary appendices).
+ *
+ * Critically, this only matches **top-level** section anchors — never the
+ * brief's own internal `<h2>` sub-headings (`## BLUF`, `## 60-Second Read`,
+ * …), which carry slugified ids without the `section-` prefix. That is why
+ * we cannot simply look for the next `<h2`.
+ *
+ * Uses `indexOf`/`lastIndexOf` exclusively (no regex) to stay within
+ * CodeQL's safe-regex envelope.
+ *
+ * @param html - Full article body HTML
+ * @param afterHeading - Index immediately after the Executive Brief `</h2>`
+ * @returns Index of the next boundary `<h2`, or `-1` when the Executive
+ *          Brief is the last block in the body.
+ */
+export function findExecutiveBriefSectionCut(html, afterHeading) {
+    let best = -1;
+    const consider = (markerIdx) => {
+        if (markerIdx === -1)
+            return;
+        const h2 = html.lastIndexOf('<h2', markerIdx);
+        if (h2 === -1 || h2 < afterHeading)
+            return;
+        if (best === -1 || h2 < best)
+            best = h2;
+    };
+    consider(html.indexOf('id="section-', afterHeading));
+    for (const marker of EXECUTIVE_BRIEF_BOUNDARY_ID_MARKERS) {
+        consider(html.indexOf(marker, afterHeading));
+    }
+    return best;
+}
 /**
  * Replace the **inner body** of the Executive Brief section (the
  * `<h2 id="section-executive-brief">…</h2>` heading and everything that
- * follows it up to — but not including — the next `<h2 id="section-…">`
- * sibling) with the supplied replacement HTML. The Executive Brief
+ * follows it up to — but not including — the next top-level boundary
+ * heading) with the supplied replacement HTML. The Executive Brief
  * heading itself is preserved by emitting it inline ahead of the
  * replacement, so the in-page anchor (`#section-executive-brief`) and
  * the table-of-contents link continue to work.
@@ -119,8 +173,11 @@ export function replaceFirstStringIn(haystack, needle, replacement) {
  * `render-one.writeLanguageVariant`.
  *
  * Implementation uses `indexOf`/slice exclusively to stay within
- * CodeQL's safe-regex envelope. Returns `html` unchanged when the
- * Executive Brief heading is absent or malformed.
+ * CodeQL's safe-regex envelope. The replacement spans from the heading to
+ * the next top-level boundary (see {@link findExecutiveBriefSectionCut});
+ * when the Executive Brief is the last block in the body the replacement
+ * extends to end-of-body. Returns `html` unchanged only when the Executive
+ * Brief heading is absent or malformed.
  *
  * @param html - Full article body HTML
  * @param localizedHeading - Localized text for the Executive Brief H2
@@ -147,23 +204,24 @@ export function replaceExecutiveBriefSection(html, localizedHeading, replacement
     if (h2CloseTagIdx === -1)
         return html;
     const afterHeading = h2CloseTagIdx + '</h2>'.length;
-    // Find the next `<h2 id="section-...">` boundary — the start of the
-    // following article section. If there is no further section heading
-    // we conservatively bail out (replacing through end-of-body would
-    // also drop appendix content like Reader Guide / Key Takeaways).
-    const nextSectionId = html.indexOf('id="section-', afterHeading);
-    if (nextSectionId === -1)
-        return html;
-    const nextH2 = html.lastIndexOf('<h2', nextSectionId);
-    if (nextH2 === -1 || nextH2 <= afterHeading)
-        return html;
-    // Find the start of the line containing the next `<h2` so we don't
-    // strip leading whitespace from the next section. We look at most
-    // one newline back.
-    let cutEnd = nextH2;
-    const prevNewline = html.lastIndexOf('\n', nextH2 - 1);
-    if (prevNewline !== -1 && prevNewline >= afterHeading) {
-        cutEnd = prevNewline + 1;
+    // Find the next top-level boundary heading — the start of the following
+    // article section or appendix. When none exists the Executive Brief is
+    // the last block, so we replace through end-of-body. This guarantees the
+    // localized brief is spliced even on sparse runs (previously the splice
+    // bailed and non-English readers were stranded on the English brief).
+    const nextH2 = findExecutiveBriefSectionCut(html, afterHeading);
+    let cutEnd;
+    if (nextH2 === -1) {
+        cutEnd = html.length;
+    }
+    else {
+        // Start of the line containing the next `<h2` so we don't strip
+        // leading whitespace from the next section.
+        cutEnd = nextH2;
+        const prevNewline = html.lastIndexOf('\n', nextH2 - 1);
+        if (prevNewline !== -1 && prevNewline >= afterHeading) {
+            cutEnd = prevNewline + 1;
+        }
     }
     const newHeading = `<h2 id="section-executive-brief">${escapeHTML(localizedHeading)}</h2>\n`;
     const trimmedReplacement = replacementBodyHtml.endsWith('\n')

package/scripts/aggregator/html/shell.js

@@ -19,6 +19,7 @@ import { escapeHTML } from '../../utils/file-utils.js';
 import { buildResponsiveIconLinks, buildResponsiveSocialImageMeta, buildSiteFooter, buildSiteHeader, buildPageBanner, } from '../../templates/section-builders.js';
 import { getPoliticalIntelligenceFilename } from '../../generators/political-intelligence.js';
 import { getSitemapFilename } from '../../generators/sitemap/index.js';
+import { buildRssAlternateLink } from '../../templates/sections/rss-discovery.js';
 import { truncateHeadline, getTitleSeparator, buildPageTitle, getLocalizedArticleType, getLocalizedArticleTypePlain, } from './headline.js';
 import { clampForBudget } from '../metadata/seo-budgets.js';
 import { getArticleFilename, buildArticleHreflangLinks, buildLanguageSwitcher, } from './hreflang.js';
@@ -352,7 +353,7 @@ ${keywordsMeta}  <meta name="robots" content="index, follow, max-snippet:-1, max
   <meta property="article:publisher" content="https://hack23.com">
   <link rel="canonical" href="${canonicalUrl}">
 ${hreflangLinks}
-  <link rel="alternate" type="application/rss+xml" title="EU Parliament Monitor RSS" href="${BASE_URL}/rss.xml">
+  ${buildRssAlternateLink(safeLang, `${BASE_URL}/`)}
   <link rel="preconnect" href="https://hack23.com" crossorigin>
   <meta property="og:type" content="article">
   <meta property="og:title" content="${escapeHTML(ogTitleClamped)}">

package/scripts/aggregator/metadata/artifact-category-heading.js

@@ -153,6 +153,13 @@ export const ARTIFACT_CATEGORY_PREFIXES = [
     'voting patterns',
     'weekly outlook',
     'wildcards blackswans',
+    // CJK localized category prefixes (translations of "executive briefing")
+    'エグゼクティブ・ブリーフィング',
+    'エグゼクティブブリーフィング',
+    'エグゼクティブ・ブリーフ',
+    '행정 브리핑',
+    '执行简报',
+    '執行簡報',
 ];
 /**
  * Match a single calendar month name (English) with optional `-uary` /
@@ -211,7 +218,7 @@ function normaliseCategoryHeading(raw) {
     return stripInlineMarkdown(raw)
         .trim()
         .toLowerCase()
-        .replace(/^[^a-z0-9]+/, '')
+        .replace(/^[^a-z0-9\p{L}]+/u, '')
         .replace(/\s+/g, ' ');
 }
 /**

package/scripts/aggregator/metadata/heading-rules.js

@@ -158,6 +158,17 @@ const BARE_INSTITUTIONAL_HEADINGS = [
     'briefing',
     'intelligence brief',
     'intelligence briefing',
+    // CJK / localized translations of generic headings
+    'エグゼクティブ・ブリーフィング',
+    'エグゼクティブブリーフィング',
+    'エグゼクティブ・ブリーフ',
+    'ブリーフィング',
+    '행정 브리핑',
+    '브리핑',
+    '执行简报',
+    '简报',
+    '執行簡報',
+    '簡報',
 ];
 /**
  * Return `true` when the heading is one of {@link BARE_INSTITUTIONAL_HEADINGS}

package/scripts/aggregator/metadata/seo-budgets.js

@@ -160,15 +160,18 @@ export function clampForBudget(text, lang, surface) {
         if (cleaned.length >= softMin)
             return cleaned;
     }
-    // Whitespace-aware fallback. Chinese and Japanese text often has no
-    // ASCII spaces, so skip this step for them and fall straight through
-    // to the hard cut. Korean is the exception — it uses inter-word spaces.
-    if (family !== 'cjk' || lang === 'ko') {
-        const lastSpace = window.lastIndexOf(' ');
-        if (lastSpace >= softMin) {
-            const safe = trimTrailingSeparators(window.slice(0, lastSpace));
-            return `${safe}…`;
-        }
+    // Whitespace-aware fallback. Runs for every script: an ASCII space
+    // past the soft minimum is a safe break that drops a partial trailing
+    // segment whole rather than slicing it mid-token. Chinese and Japanese
+    // prose has no inter-word spaces, so `lastIndexOf(' ')` returns -1 and
+    // this is a no-op for them — but composed SEO snippets join clauses
+    // (body, dateline, reader label) with ASCII spaces, so honouring that
+    // boundary prevents hard-cutting the reader label mid-word. Korean
+    // uses inter-word spaces natively and benefits the same way.
+    const lastSpace = window.lastIndexOf(' ');
+    if (lastSpace >= softMin) {
+        const safe = trimTrailingSeparators(window.slice(0, lastSpace));
+        return `${safe}…`;
     }
     const hardCut = trimTrailingSeparators(window);
     return `${hardCut}…`;

package/scripts/aggregator/reader-friendly-transform.js

@@ -46,7 +46,7 @@ const ADMIRALTY_LABELS = {
 export function applyReaderFriendlyTransform(html) {
     const state = createInitialState(html);
     const withGlossary = injectReaderGlossary(html);
-    const parts = withGlossary.split(/(<[^>]+>)/g);
+    const parts = withGlossary.split(/(<[^<>]+>)/g);
     for (let i = 0; i < parts.length; i++) {
         const part = parts[i] ?? '';
         if (part.startsWith('<')) {

package/scripts/generators/news-indexes/backfill-hreflang.d.ts

@@ -0,0 +1,13 @@
+/**
+ * Backfill hreflang alternate links for all article HTML files.
+ *
+ * Handles three cases:
+ * 1. Articles with no hreflang links at all → inject the full block before `</head>`
+ * 2. Articles with relative hreflang URLs → replace with absolute URLs
+ * 3. Articles already correct → skip
+ *
+ * @param filenames - News article filenames
+ * @returns Number of HTML files updated
+ */
+export declare function backfillArticleHreflang(filenames: readonly string[]): number;
+//# sourceMappingURL=backfill-hreflang.d.ts.map

package/scripts/generators/news-indexes/backfill-hreflang.js

@@ -0,0 +1,112 @@
+// SPDX-FileCopyrightText: 2024-2026 Hack23 AB
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * @module Generators/NewsIndexes/BackfillHreflang
+ * @description Hreflang alternate-link backfill for article HTML files.
+ * Extracted from `backfill.ts` to keep source files ≤600 lines.
+ */
+import path from 'path';
+import fs from 'fs';
+import { NEWS_DIR, BASE_URL } from '../../constants/config.js';
+import { ALL_LANGUAGES } from '../../constants/languages.js';
+import { parseArticleFilename, atomicWrite } from '../../utils/file-utils.js';
+/**
+ * Read an article HTML file, returning an empty string when unavailable.
+ *
+ * @param filepath - Absolute HTML file path
+ * @returns File content or empty string
+ */
+function readArticleHtml(filepath) {
+    try {
+        return path.isAbsolute(filepath) ? fs.readFileSync(filepath, 'utf8') : '';
+    }
+    catch {
+        return '';
+    }
+}
+/**
+ * Build hreflang `<link rel="alternate">` tags for an article slug.
+ * Produces one tag per supported language plus an `x-default` pointing at
+ * the English variant, all using absolute URLs.
+ *
+ * @param articleSlug - Slug without language suffix (e.g. `2026-02-24-propositions`)
+ * @returns Newline-joined `<link>` tags
+ */
+function buildArticleHreflang(articleSlug) {
+    const entries = ALL_LANGUAGES.map((code) => `  <link rel="alternate" hreflang="${code}" href="${BASE_URL}/news/${articleSlug}-${code}.html">`);
+    entries.push(`  <link rel="alternate" hreflang="x-default" href="${BASE_URL}/news/${articleSlug}-en.html">`);
+    return entries.join('\n');
+}
+/**
+ * Inject hreflang links into an article that has none.
+ *
+ * @param html - Article HTML content
+ * @param hreflangBlock - Pre-built hreflang link block
+ * @returns Updated HTML, or original if no change needed
+ */
+function injectHreflangLinks(html, hreflangBlock) {
+    return html.replace(/(<\/head>)/u, `${hreflangBlock}\n$1`);
+}
+/**
+ * Replace existing relative hreflang links with absolute URLs.
+ *
+ * @param html - Article HTML content
+ * @param hreflangBlock - Pre-built hreflang link block with absolute URLs
+ * @returns Updated HTML, or original if no change needed
+ */
+function fixRelativeHreflangLinks(html, hreflangBlock) {
+    const stripped = html.replace(/\s*<link\s+rel="alternate"\s+hreflang="[^"]*"\s+href="[^"]*">\n?/gu, '');
+    return stripped.replace(/(<\/head>)/u, `${hreflangBlock}\n$1`);
+}
+/**
+ * Backfill hreflang alternate links for all article HTML files.
+ *
+ * Handles three cases:
+ * 1. Articles with no hreflang links at all → inject the full block before `</head>`
+ * 2. Articles with relative hreflang URLs → replace with absolute URLs
+ * 3. Articles already correct → skip
+ *
+ * @param filenames - News article filenames
+ * @returns Number of HTML files updated
+ */
+export function backfillArticleHreflang(filenames) {
+    let updated = 0;
+    for (const filename of filenames) {
+        if (backfillOneArticleHreflang(filename))
+            updated++;
+    }
+    return updated;
+}
+/**
+ * Backfill hreflang for a single article file.
+ *
+ * @param filename - News article filename
+ * @returns True when the file was updated
+ */
+function backfillOneArticleHreflang(filename) {
+    const parsed = parseArticleFilename(filename);
+    if (!parsed)
+        return false;
+    const filepath = path.join(NEWS_DIR, filename);
+    const html = readArticleHtml(filepath);
+    if (!html)
+        return false;
+    const articleSlug = `${parsed.date}-${parsed.slug}`;
+    const hreflangBlock = buildArticleHreflang(articleSlug);
+    const hasHreflang = /<link\s+rel="alternate"\s+hreflang="/u.test(html);
+    let next;
+    if (!hasHreflang) {
+        next = injectHreflangLinks(html, hreflangBlock);
+    }
+    else {
+        const hasRelative = /<link\s+rel="alternate"\s+hreflang="[^"]*"\s+href="(?!https?:\/\/)/u.test(html);
+        if (!hasRelative)
+            return false;
+        next = fixRelativeHreflangLinks(html, hreflangBlock);
+    }
+    if (next === html)
+        return false;
+    atomicWrite(filepath, next);
+    return true;
+}
+//# sourceMappingURL=backfill-hreflang.js.map

package/scripts/generators/news-indexes/backfill-reader-label.d.ts

@@ -0,0 +1,47 @@
+import type { LanguageCode } from '../../types/index.js';
+/**
+ * Remove a trailing **truncated** copy of the localized reader label
+ * (`SEO_CONTEXT_LABELS[lang].reader`) from a candidate description.
+ *
+ * Earlier backfill passes appended the reader label and then clamped the
+ * whole buffer to the per-script `metaDescription` budget, hard-cutting
+ * the label mid-word (e.g. zh `…政策后果的读` instead of `…政策后果的读者`,
+ * ja `…追跡する読`, ko dangling `…추적하는.`). Those mangled fragments were
+ * persisted to `<meta description>` and survive a plain prefix/date-label
+ * strip, so re-feeding them to the resolver re-emits the broken tail.
+ *
+ * A trailing copy that matches the label **in full** is left intact — it
+ * is a complete, reader-facing clause we want to preserve. Only a partial
+ * (truncated) prefix of the label is dropped, leaving the clean body for
+ * the resolver to re-enrich with a budget-aware (whole-label-or-nothing)
+ * reader clause.
+ *
+ * @param description - Candidate description (prefix/date-label removed)
+ * @param langCode - Article language code
+ * @returns Description with any truncated trailing reader label removed
+ */
+export declare function stripTruncatedReaderLabel(description: string, langCode: LanguageCode): string;
+/**
+ * Locate a trailing **truncated** copy of the localized reader label and
+ * return the index at which the description body ends (i.e. where the
+ * partial label begins). Returns -1 when no partial label is present or
+ * when the label is present in full (a complete clause we keep).
+ *
+ * @param text - Trimmed candidate description
+ * @param langCode - Article language code
+ * @returns Cut index for the partial label, or -1 when none applies
+ */
+export declare function findTruncatedReaderLabelCut(text: string, langCode: LanguageCode): number;
+/**
+ * Detect whether a legacy `<meta description>` ends with a **truncated**
+ * reader label once its dateline prefix and redundant date-label clause
+ * are removed. Long, unique legacy descriptions otherwise bypass
+ * `shouldBackfillDescription`, leaving a persisted mid-word cut
+ * (e.g. zh `…政策后果的读`, ja `…追跡する読`, ko `…추적하는.`) in place.
+ *
+ * @param body - Stripped description body (prefix/date-label removed)
+ * @param langCode - Article language code
+ * @returns True when a truncated reader label remains in the body
+ */
+export declare function hasTruncatedReaderLabelInBody(body: string, langCode: LanguageCode): boolean;
+//# sourceMappingURL=backfill-reader-label.d.ts.map

package/scripts/generators/news-indexes/backfill-reader-label.js

@@ -0,0 +1,86 @@
+// SPDX-FileCopyrightText: 2024-2026 Hack23 AB
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * @module Generators/NewsIndexes/BackfillReaderLabel
+ * @description Truncated reader-label detection and stripping helpers,
+ * extracted from `backfill.ts` to keep source files ≤600 lines.
+ */
+import { getLocalizedString } from '../../constants/languages.js';
+import { SEO_CONTEXT_LABELS } from '../../aggregator/metadata/template-fallback.js';
+/**
+ * Remove a trailing **truncated** copy of the localized reader label
+ * (`SEO_CONTEXT_LABELS[lang].reader`) from a candidate description.
+ *
+ * Earlier backfill passes appended the reader label and then clamped the
+ * whole buffer to the per-script `metaDescription` budget, hard-cutting
+ * the label mid-word (e.g. zh `…政策后果的读` instead of `…政策后果的读者`,
+ * ja `…追跡する読`, ko dangling `…추적하는.`). Those mangled fragments were
+ * persisted to `<meta description>` and survive a plain prefix/date-label
+ * strip, so re-feeding them to the resolver re-emits the broken tail.
+ *
+ * A trailing copy that matches the label **in full** is left intact — it
+ * is a complete, reader-facing clause we want to preserve. Only a partial
+ * (truncated) prefix of the label is dropped, leaving the clean body for
+ * the resolver to re-enrich with a budget-aware (whole-label-or-nothing)
+ * reader clause.
+ *
+ * @param description - Candidate description (prefix/date-label removed)
+ * @param langCode - Article language code
+ * @returns Description with any truncated trailing reader label removed
+ */
+export function stripTruncatedReaderLabel(description, langCode) {
+    const text = description.trim();
+    const cut = findTruncatedReaderLabelCut(text, langCode);
+    if (cut < 0)
+        return text;
+    return text
+        .replace(/[.。！？!?…]+$/u, '')
+        .slice(0, cut)
+        .replace(/[\s,;:—\-–·。、]+$/u, '')
+        .trim();
+}
+/**
+ * Locate a trailing **truncated** copy of the localized reader label and
+ * return the index at which the description body ends (i.e. where the
+ * partial label begins). Returns -1 when no partial label is present or
+ * when the label is present in full (a complete clause we keep).
+ *
+ * @param text - Trimmed candidate description
+ * @param langCode - Article language code
+ * @returns Cut index for the partial label, or -1 when none applies
+ */
+export function findTruncatedReaderLabelCut(text, langCode) {
+    const labels = getLocalizedString(SEO_CONTEXT_LABELS, langCode);
+    const reader = (labels.reader ?? '').trim();
+    // Require a reasonably long label so we never strip on a coincidental
+    // short suffix match; real labels are 40+ chars (Latin) / 11+ (CJK).
+    if (reader.length < 8 || text.length < 8)
+        return -1;
+    // Tolerate a terminator the resolver/healer appended after the cut.
+    const core = text.replace(/[.。！？!?…]+$/u, '');
+    const maxK = Math.min(core.length, reader.length);
+    for (let k = maxK; k >= 8; k -= 1) {
+        if (core.slice(core.length - k) === reader.slice(0, k)) {
+            // Full label present at the tail — keep it (not a truncation).
+            if (k === reader.length)
+                return -1;
+            return core.length - k;
+        }
+    }
+    return -1;
+}
+/**
+ * Detect whether a legacy `<meta description>` ends with a **truncated**
+ * reader label once its dateline prefix and redundant date-label clause
+ * are removed. Long, unique legacy descriptions otherwise bypass
+ * `shouldBackfillDescription`, leaving a persisted mid-word cut
+ * (e.g. zh `…政策后果的读`, ja `…追跡する読`, ko `…추적하는.`) in place.
+ *
+ * @param body - Stripped description body (prefix/date-label removed)
+ * @param langCode - Article language code
+ * @returns True when a truncated reader label remains in the body
+ */
+export function hasTruncatedReaderLabelInBody(body, langCode) {
+    return findTruncatedReaderLabelCut(body, langCode) >= 0;
+}
+//# sourceMappingURL=backfill-reader-label.js.map

package/scripts/generators/news-indexes/backfill.d.ts

@@ -60,14 +60,26 @@ export declare function buildLegacyBackfillDescription(date: string, slug: strin
     readonly forceContextPrefix?: boolean;
 }): string;
 /**
- * Apply SEO meta tag replacements to a complete article HTML document.
+ * Strip the legacy dateline prefix **and** the redundant localized
+ * date-label clause from a candidate description, returning the
+ * reader-facing body in isolation. Used to clean a previously-backfilled
+ * `<meta description>` before it is re-fed to the per-language SEO
+ * resolver — without this, the resolver re-clamps the prefixed buffer
+ * against the CJK metaDescription budget and truncates the reader label
+ * mid-clause (live regression in `news/2026-04-26-week-ahead-ko.html`,
+ * a dangling "추적하는." participle).
  *
+ * @param date - Article date (ISO YYYY-MM-DD)
+ * @param slug - Article slug
+ * @param lang - Article language code
+ * @param description - Candidate description (possibly already prefixed)
+ * @returns Reader-facing body with prefix + date label removed
+ */
+export declare function stripLegacyBackfillContext(date: string, slug: string, lang: string, description: string): string;
+/**
+ * Apply SEO meta tag replacements to a complete article HTML document.
  * Exported for the regression test in
- * `test/unit/news-indexes-jsonld-description-regex.test.js`, which
- * locks in the JSON-LD description regex against the duplicate-tail
- * bug (the legacy `"description":"[^"]*"` pattern terminated at the
- * first JSON-escaped quote `\"` and left the previous description's
- * tail in place, accumulating duplicates on every prebuild run).
+ * `test/unit/news-indexes-jsonld-description-regex.test.js`.
  *
  * @param html - Existing article HTML
  * @param description - Backfilled meta description
@@ -75,16 +87,5 @@ export declare function buildLegacyBackfillDescription(date: string, slug: strin
  * @returns Updated HTML
  */
 export declare function applyArticleSeoBackfill(html: string, description: string, keywords: readonly string[]): string;
-/**
- * Backfill hreflang alternate links for all article HTML files.
- *
- * Handles three cases:
- * 1. Articles with no hreflang links at all → inject the full block before `</head>`
- * 2. Articles with relative hreflang URLs → replace with absolute URLs
- * 3. Articles already correct → skip
- *
- * @param filenames - News article filenames
- * @returns Number of HTML files updated
- */
-export declare function backfillArticleHreflang(filenames: readonly string[]): number;
+export { backfillArticleHreflang } from './backfill-hreflang.js';
 //# sourceMappingURL=backfill.d.ts.map