euparliamentmonitor 0.9.22 → 0.9.24

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "euparliamentmonitor",
-  "version": "0.9.22",
+  "version": "0.9.24",
   "type": "module",
   "description": "European Parliament Intelligence Platform - Monitor political activity with systematic transparency",
   "main": "scripts/index.js",
@@ -66,6 +66,7 @@
     "discover:untranslated-briefs": "node scripts/discover-untranslated-briefs.js",
     "validate:translations": "node scripts/validate-brief-translations.js",
     "validate:manifest-seo": "node scripts/validate-manifest-seo.js",
+    "validate:seo": "node scripts/validate-article-seo.js",
     "sync:templates": "node scripts/templates/sync-template-frontmatter.js",
     "sync:templates:check": "node scripts/templates/sync-template-frontmatter.js --check",
     "prior-run-diff": "node scripts/aggregator/prior-run-diff.js",

package/scripts/aggregator/article-metadata.js

@@ -64,7 +64,7 @@
 import { ALL_LANGUAGES } from '../constants/language-core.js';
 import { resolveLocalizedBriefHighlight } from './editorial-brief-resolver.js';
 import { buildTemplateFallback } from './metadata/template-fallback.js';
-import { buildSeoKeywords, composeContextualDescription, composeContextualTitle, manifestOverrideFor, pickFirstNonEmpty, resolveEditorialContent, } from './metadata/resolve-helpers.js';
+import { buildSeoKeywords, composeContextualDescription, composeContextualExtendedDescription, composeContextualTitle, deriveHeadlineFromSummary, hasLeakySeoToken, isUsableResolvedTitle, manifestOverrideFor, pickFirstNonEmpty, resolveEditorialContent, sanitizeDescriptionCandidate, } from './metadata/resolve-helpers.js';
 import { ENRICHMENT_TRIGGER_LENGTH, truncateDescription, truncateExtendedDescription, truncateTitle, } from './metadata/text-utils.js';
 export { shouldSkipDescriptionLine, stripLeadingProseLabel, stripInlineMarkdown, truncateDescription, truncateExtendedDescription, truncateTitle, extractFirstSentence, } from './metadata/text-utils.js';
 export { isArtifactCategoryHeading, stripArtifactCategoryAffix, isGenericHeading, } from './metadata/heading-rules.js';
@@ -109,6 +109,7 @@ export function resolveArticleMetadata(opts) {
     }
     return result;
 }
+const LOCALIZED_BRIEF_SOURCE = 'localized-brief';
 /**
  * Resolve `{title, description, keywords, source}` for one language.
  *
@@ -122,20 +123,74 @@ function resolveOneLanguage(input) {
     const editorial = perLanguage.editorial;
     const contextualTitle = composeContextualTitle(input.template.title, editorial.headline, input.runId);
     const title = pickFirstNonEmpty([manifestTitle, contextualTitle, input.template.title]);
-    const rawDescription = pickFirstNonEmpty([
-        manifestDescription,
-        editorial.summary,
-        input.template.subtitle,
+    const rawDescription = sanitizeDescriptionCandidate(pickFirstNonEmpty([manifestDescription, editorial.summary, input.template.subtitle]));
+    const safeEditorial = {
+        headline: isUsableResolvedTitle(editorial.headline) ? editorial.headline.trim() : '',
+        summary: sanitizeDescriptionCandidate(editorial.summary),
+        extendedSummary: sanitizeDescriptionCandidate(editorial.extendedSummary),
+    };
+    const normalizedRawDescription = rawDescription || sanitizeDescriptionCandidate(input.template.subtitle);
+    const skipEnrichment = perLanguage.source === LOCALIZED_BRIEF_SOURCE && normalizedRawDescription.length > 0;
+    const description = skipEnrichment || normalizedRawDescription.length >= ENRICHMENT_TRIGGER_LENGTH
+        ? normalizedRawDescription
+        : composeContextualDescription(input.lang, normalizedRawDescription, safeEditorial, input.date, input.runId);
+    const clippedTitle = truncateTitle(title).trim();
+    const explicitTitle = manifestTitle && !hasLeakySeoToken(manifestTitle) ? truncateTitle(manifestTitle).trim() : '';
+    const allowShortResolvedTitle = perLanguage.source === LOCALIZED_BRIEF_SOURCE;
+    const resolvedTitleCandidate = clippedTitle &&
+        !hasLeakySeoToken(clippedTitle) &&
+        (allowShortResolvedTitle || isUsableResolvedTitle(clippedTitle))
+        ? clippedTitle
+        : '';
+    const summaryDerivedTitle = deriveHeadlineFromSummary(safeEditorial.summary || normalizedRawDescription);
+    // `truncateTitle` returns '' when an editorial title overruns the
+    // budget with no acceptable clause boundary — fall back to the
+    // localized template title in that case so we never emit an empty
+    // `<title>`. Live regression: 2026-05-22 breaking
+    // `AI Trade Strategy: A Legislative First with Structural…` clipped
+    // to '' after the no-ellipsis guard landed; template fallback
+    // (`Extended Executive Brief — Breaking News`) is preferable to a
+    // blank `<title>`.
+    //
+    // The fallback path passes the template title back through
+    // {@link composeContextualTitle} (with an empty editorial headline)
+    // so `withRunQualifier` re-appends the `— Run N` suffix. Without
+    // this, two same-date / same-articleType runs (republish, hot-fix
+    // re-run) would collapse to byte-identical `<title>` strings, and
+    // the duplicate-title gate in `scripts/validate-article-seo.js`
+    // would (correctly) fail CI.
+    const contextualFallback = composeContextualTitle(input.template.title, '', input.runId);
+    const truncatedTitle = pickFirstNonEmpty([
+        explicitTitle,
+        resolvedTitleCandidate,
+        isUsableResolvedTitle(summaryDerivedTitle, { allowFullSentence: true })
+            ? summaryDerivedTitle
+            : '',
+        truncateTitle(contextualFallback),
+        contextualFallback,
     ]);
-    const description = rawDescription.length >= ENRICHMENT_TRIGGER_LENGTH
-        ? rawDescription
-        : composeContextualDescription(input.lang, rawDescription, editorial, input.date, input.runId);
-    const truncatedTitle = truncateTitle(title);
     const truncatedDescription = truncateDescription(description);
-    const extendedSource = manifestDescription
-        ? manifestDescription
-        : editorial.extendedSummary || rawDescription;
-    const truncatedExtendedDescription = truncateExtendedDescription(extendedSource);
+    const extendedSource = sanitizeDescriptionCandidate(manifestDescription || safeEditorial.extendedSummary || normalizedRawDescription);
+    // Two-tier extended-description resolution:
+    // 1. Direct truncation — preferred when the editorial source paragraph
+    //    is already ≥181 chars (the truncator's gating threshold). This
+    //    yields the highest-fidelity og:description text.
+    // 2. Contextual synthesis — when direct truncation returns '' (source
+    //    was too short), synthesize a longer string by stitching together
+    //    `<source> + Date: YYYY-MM-DD + Context: <editorial> + <reader>`.
+    //    This is the **only** SEO path that surfaces the localized
+    //    "for democratic-accountability readers …" framing (the short
+    //    <meta description> no longer carries it — see comment in
+    //    {@link composeContextualDescription}). The synthesized string is
+    //    re-clamped to the 200–300 char og:description budget.
+    //
+    // Live regression (2026-05): 56 breaking briefs shipped with empty
+    // extendedDescription because their lead paragraph was only 80–150
+    // chars. AI-overview and Discover surfaces dropped them entirely.
+    let truncatedExtendedDescription = truncateExtendedDescription(extendedSource);
+    if (!truncatedExtendedDescription) {
+        truncatedExtendedDescription = composeContextualExtendedDescription(input.lang, extendedSource || normalizedRawDescription, safeEditorial, input.date);
+    }
     const source = manifestTitle || manifestDescription ? 'manifest' : perLanguage.source;
     return {
         title: truncatedTitle,
@@ -163,7 +218,7 @@ function resolvePerLanguageEditorial(input) {
                     summary: localized.summary,
                     extendedSummary: localized.extendedSummary,
                 },
-                source: 'localized-brief',
+                source: LOCALIZED_BRIEF_SOURCE,
             };
         }
     }

package/scripts/aggregator/editorial-brief-resolver.js

@@ -30,6 +30,7 @@
 import fs from 'fs';
 import path from 'path';
 import { extractFirstH1, extractLedeAfterHeading, extractExtendedLedeAfterHeading, extractStrongProseLine, isGenericHeading, stripArtifactCategoryAffix, truncateTitle, } from './article-metadata.js';
+import { extractBriefingHighlight } from './metadata/briefing-highlight.js';
 /**
  * Run-relative candidate paths for a translated brief, in precedence
  * order. Mirrors the `executive-brief.md` → `extended/executive-brief.md`
@@ -174,6 +175,28 @@ export function resolveLocalizedBriefHighlight(runDir, lang, articleType, date)
         const body = readArtefactBody(abs);
         if (!body)
             continue;
+        // Tier 1 (NEW, May-2026): structural extraction of `## Strategic
+        // Intelligence Summary` / `## Reader Briefing` sections. The
+        // briefing extractor is language-agnostic — it matches on the
+        // English section headings, which the translation pipeline
+        // preserves verbatim under the localized brief contract — so a
+        // Swedish brief whose synthesis section is still written as
+        // `## Strategic Intelligence Summary` (with translated body
+        // prose) will resolve correctly here. When the translator has
+        // additionally localized the section heading the matcher falls
+        // back to the legacy lede/H1 path below, producing the
+        // localized H1 as headline.
+        const briefing = extractBriefingHighlight(body);
+        if (briefing && (briefing.headline || briefing.summary)) {
+            const fallbackHeadline = deriveHeadline(body, articleType, date);
+            return {
+                headline: briefing.headline || fallbackHeadline,
+                summary: briefing.summary,
+                extendedSummary: briefing.extendedSummary || extractExtendedLedeAfterHeading(body),
+                sourceFile: rel,
+                sourceLang: lang,
+            };
+        }
         const headline = deriveHeadline(body, articleType, date);
         const lede = extractLedeAfterHeading(body);
         const summary = lede || extractStrongProseLine(body);

package/scripts/aggregator/generator/slug.js

@@ -37,27 +37,6 @@ export function buildArticleSlug(date, articleType, runSuffix) {
 export function sanitizeRunSuffix(runId) {
     return _sanitizeRunSuffix(runId);
 }
-/**
- * Return `true` when a line should be skipped when hunting for the default
- * description. Thin wrapper preserved for back-compat — real logic lives
- * in `src/aggregator/article-metadata.ts`'s `shouldSkipDescriptionLine`.
- *
- * @param line - Trimmed line from the aggregated Markdown source
- * @returns `true` when the line is not prose and should be skipped
- */
-function shouldSkipDescriptionLine(line) {
-    if (line.length === 0)
-        return true;
-    if (line.startsWith('#'))
-        return true;
-    if (line.startsWith('>'))
-        return true;
-    if (line.startsWith('<'))
-        return true;
-    if (line.startsWith('|'))
-        return true;
-    return false;
-}
 /** Description used when no prose paragraph qualifies. */
 const FALLBACK_DESCRIPTION = 'EU Parliament intelligence summary derived from committed analysis artifacts.';
 /**
@@ -73,7 +52,6 @@ const FALLBACK_DESCRIPTION = 'EU Parliament intelligence summary derived from co
  * @returns Plain-text description, truncated to ≤300 characters
  */
 export function extractDefaultDescription(markdown) {
-    void shouldSkipDescriptionLine;
     const strong = extractStrongProseLine(markdown);
     return strong.length > 0 ? strong : FALLBACK_DESCRIPTION;
 }

package/scripts/aggregator/html/headline.d.ts

@@ -1,4 +1,5 @@
-import type { LanguageCode } from '../../types/index.js';
+import type { LanguageCode, LanguageMap } from '../../types/index.js';
+import { type SeoSurface } from '../metadata/seo-budgets.js';
 /**
  * Resolve a localized article type label *without* the leading icon
  * emoji. Used for the OpenGraph `article:section` meta and the JSON-LD
@@ -28,19 +29,50 @@ export declare const HEADLINE_LIMIT = 110;
 export declare function truncateHeadline(title: string): string;
 /**
  * Build the localized `<title>` separator for the
- * `{articleTitle} {sep} {siteTitle}` pattern. LTR locales use the
- * right-pointing guillemet (»); RTL locales (Arabic, Hebrew) use the
- * left-pointing guillemet («) so the visual hierarchy reads from the
- * primary title towards the site name without breaking bidi flow.
+ * `{articleTitle} {sep} {siteTitle}` pattern.
  *
- * The previous em-dash separator collided with em-dashes inside
- * article titles (the editorial style uses `Title — Subtitle`) and
- * rendered ambiguously in screen readers.
+ * Latin scripts use the policy-mandated ASCII pipe (`" | "`), which
+ * scans cleanly in SERP cards and never collides with em-dashes that
+ * the editorial style routinely uses inside titles. CJK locales use
+ * the Katakana middle-dot (`" ・ "`, U+30FB) which is the documented
+ * Google CJK separator and renders correctly in JP / KO / ZH fonts.
+ * RTL locales use the Hebrew paseq (`" ׀ "`, U+05C0) — a vertical
+ * stroke that preserves bidi flow without injecting a Latin guillemet
+ * that would force a direction change mid-title.
  *
  * @param lang - Target language code
- * @returns `" » "` for LTR locales, `" « "` for RTL
+ * @returns Per-script separator
  */
 export declare function getTitleSeparator(lang: LanguageCode): string;
+/**
+ * Short brand fallback per script family. Used by
+ * {@link buildPageTitle} when the full `SITE_NAME` would push the
+ * `<title>` past the SERP budget but a shorter variant would fit.
+ *
+ * - Latin → "EPM" (3 chars, ASCII-safe in news cards)
+ * - CJK → "EPM" (Latin abbreviation reads correctly in JP / KO / ZH SERPs)
+ * - RTL → Arabic abbreviation "EPM" works in both Arabic and Hebrew
+ *   SERP cards (Bing/Google render the Latin token RTL-isolated)
+ *
+ * Per-locale overrides live in {@link SHORT_SITE_NAMES} below so a
+ * future editorial change (e.g. a registered Arabic brand) only
+ * touches the table.
+ */
+export declare const SHORT_SITE_NAMES: LanguageMap;
+/**
+ * Compose a title for one SEO surface using the per-script byte
+ * budget from `metadata/seo-budgets.ts`. Drops the brand suffix when
+ * the article title alone fills the budget (better SERP than a
+ * truncated headline followed by a clipped brand) and falls through
+ * to a short-brand variant when that fits but the full one does not.
+ *
+ * @param title - Article title (plain text, already markdown-stripped)
+ * @param lang - Target language code
+ * @param siteTitle - Full brand suffix (e.g. "EU Parliament Monitor")
+ * @param surface - Optional SEO surface; defaults to `<title>` budget
+ * @returns Composed, budget-clamped title
+ */
+export declare function buildPageTitle(title: string, lang: LanguageCode, siteTitle: string, surface?: SeoSurface): string;
 /**
  * Resolve a localized article type label with icon. Falls back to the
  * humanised slug when a translation isn't available.

package/scripts/aggregator/html/headline.js

@@ -7,8 +7,9 @@
  * icon), the page-title separator that respects bidi direction, and the
  * Schema.org-compatible truncated headline used in JSON-LD.
  */
-import { ARTICLE_TYPE_LABELS, ARTICLE_TYPE_ICONS, getLocalizedString, getTextDirection, } from '../../constants/languages.js';
+import { ARTICLE_TYPE_LABELS, ARTICLE_TYPE_ICONS, getLocalizedString, } from '../../constants/languages.js';
 import { ArticleCategory } from '../../types/index.js';
+import { classifyScript, clampTitleForSurface } from '../metadata/seo-budgets.js';
 /**
  * Resolve a localized article type label *without* the leading icon
  * emoji. Used for the OpenGraph `article:section` meta and the JSON-LD
@@ -51,20 +52,78 @@ export function truncateHeadline(title) {
 }
 /**
  * Build the localized `<title>` separator for the
- * `{articleTitle} {sep} {siteTitle}` pattern. LTR locales use the
- * right-pointing guillemet (»); RTL locales (Arabic, Hebrew) use the
- * left-pointing guillemet («) so the visual hierarchy reads from the
- * primary title towards the site name without breaking bidi flow.
+ * `{articleTitle} {sep} {siteTitle}` pattern.
  *
- * The previous em-dash separator collided with em-dashes inside
- * article titles (the editorial style uses `Title — Subtitle`) and
- * rendered ambiguously in screen readers.
+ * Latin scripts use the policy-mandated ASCII pipe (`" | "`), which
+ * scans cleanly in SERP cards and never collides with em-dashes that
+ * the editorial style routinely uses inside titles. CJK locales use
+ * the Katakana middle-dot (`" ・ "`, U+30FB) which is the documented
+ * Google CJK separator and renders correctly in JP / KO / ZH fonts.
+ * RTL locales use the Hebrew paseq (`" ׀ "`, U+05C0) — a vertical
+ * stroke that preserves bidi flow without injecting a Latin guillemet
+ * that would force a direction change mid-title.
  *
  * @param lang - Target language code
- * @returns `" » "` for LTR locales, `" « "` for RTL
+ * @returns Per-script separator
  */
 export function getTitleSeparator(lang) {
-    return getTextDirection(lang) === 'rtl' ? ' « ' : ' » ';
+    const family = classifyScript(lang);
+    if (family === 'cjk')
+        return ' ・ ';
+    if (family === 'rtl')
+        return ' ׀ ';
+    return ' | ';
+}
+/**
+ * Short brand fallback per script family. Used by
+ * {@link buildPageTitle} when the full `SITE_NAME` would push the
+ * `<title>` past the SERP budget but a shorter variant would fit.
+ *
+ * - Latin → "EPM" (3 chars, ASCII-safe in news cards)
+ * - CJK → "EPM" (Latin abbreviation reads correctly in JP / KO / ZH SERPs)
+ * - RTL → Arabic abbreviation "EPM" works in both Arabic and Hebrew
+ *   SERP cards (Bing/Google render the Latin token RTL-isolated)
+ *
+ * Per-locale overrides live in {@link SHORT_SITE_NAMES} below so a
+ * future editorial change (e.g. a registered Arabic brand) only
+ * touches the table.
+ */
+export const SHORT_SITE_NAMES = {
+    en: 'EPM',
+    sv: 'EPM',
+    da: 'EPM',
+    no: 'EPM',
+    fi: 'EPM',
+    de: 'EPM',
+    fr: 'EPM',
+    es: 'EPM',
+    nl: 'EPM',
+    ar: 'EPM',
+    he: 'EPM',
+    ja: 'EPM',
+    ko: 'EPM',
+    zh: 'EPM',
+};
+/**
+ * Compose a title for one SEO surface using the per-script byte
+ * budget from `metadata/seo-budgets.ts`. Drops the brand suffix when
+ * the article title alone fills the budget (better SERP than a
+ * truncated headline followed by a clipped brand) and falls through
+ * to a short-brand variant when that fits but the full one does not.
+ *
+ * @param title - Article title (plain text, already markdown-stripped)
+ * @param lang - Target language code
+ * @param siteTitle - Full brand suffix (e.g. "EU Parliament Monitor")
+ * @param surface - Optional SEO surface; defaults to `<title>` budget
+ * @returns Composed, budget-clamped title
+ */
+export function buildPageTitle(title, lang, siteTitle, surface = 'title') {
+    const shortSiteTitle = getLocalizedString(SHORT_SITE_NAMES, lang);
+    return clampTitleForSurface(title, lang, surface, {
+        siteTitle,
+        shortSiteTitle,
+        separator: getTitleSeparator(lang),
+    });
 }
 /**
  * Resolve a localized article type label with icon. Falls back to the

package/scripts/aggregator/html/shell.js

@@ -20,7 +20,8 @@ import { stripHtmlTags } from '../../utils/html-sanitize.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 { truncateHeadline, getTitleSeparator, getLocalizedArticleType, getLocalizedArticleTypePlain, } from './headline.js';
+import { truncateHeadline, getTitleSeparator, buildPageTitle, getLocalizedArticleType, getLocalizedArticleTypePlain, } from './headline.js';
+import { clampForBudget } from '../metadata/seo-budgets.js';
 import { getArticleFilename, buildArticleHreflangLinks, buildLanguageSwitcher, } from './hreflang.js';
 import { buildArticleToc } from './toc.js';
 import { blobUrl } from '../infra/github-urls.js';
@@ -28,6 +29,57 @@ import { blobUrl } from '../infra/github-urls.js';
 export const PUBLISHER_NAME = 'Hack23 AB';
 /** Site name used across meta tags and structured data. */
 export const SITE_NAME = 'EU Parliament Monitor';
+/**
+ * Compute the per-surface SEO-budget-clamped variants of the article
+ * title and description for a single render. See
+ * `analysis/methodologies/seo-headers-policy.md` § 1.1 for the
+ * documented sources of every cap.
+ *
+ * @param options - The {@link WrapArticleOptions} carrying title /
+ *                  description / extendedDescription
+ * @param lang - Validated publishing locale (already coerced to a
+ *               supported `LanguageCode`)
+ * @param siteTitle - Resolved localized site title used as the brand
+ *                    suffix
+ * @returns One {@link SeoClampedSurfaces} record per article render
+ */
+function computeSeoClamps(options, lang, siteTitle) {
+    const pageTitle = buildPageTitle(options.title, lang, siteTitle);
+    const ogTitleClamped = clampForBudget(options.title, lang, 'ogTitle');
+    const twitterTitleClamped = clampForBudget(options.title, lang, 'twitterTitle');
+    const metaDescriptionClamped = clampForBudget(options.description, lang, 'metaDescription');
+    // og:description and twitter:description prefer the longer BLUF
+    // paragraph (extendedDescription) so social-card previews show the
+    // full lede; fall back to the short meta description when the
+    // extended one is empty.
+    const socialSource = options.extendedDescription && options.extendedDescription.length > 0
+        ? options.extendedDescription
+        : options.description;
+    const ogDescriptionClamped = clampForBudget(socialSource, lang, 'ogDescription');
+    const twitterDescriptionClamped = clampForBudget(socialSource, lang, 'twitterDescription');
+    const imageAltClamped = clampForBudget(`${options.title}${getTitleSeparator(lang)}${siteTitle}`, lang, 'imageAlt');
+    const jsonLdHeadline = truncateHeadline(options.title);
+    // Emit an `alternativeHeadline` whenever the headline truncator
+    // dropped more than a handful of characters from the full title.
+    // Schema.org's `NewsArticle.alternativeHeadline` field is exactly
+    // for the long-form variant of `headline` and lets Google's
+    // Knowledge Graph keep both versions for retrieval. The 5-char
+    // threshold avoids emitting trivially redundant pairs when the
+    // truncator only trimmed trailing whitespace or punctuation.
+    const fullTitleTrimmed = options.title.trim();
+    const altCandidate = fullTitleTrimmed.length - jsonLdHeadline.length > 5 ? fullTitleTrimmed : undefined;
+    return {
+        pageTitle,
+        ogTitleClamped,
+        twitterTitleClamped,
+        metaDescriptionClamped,
+        ogDescriptionClamped,
+        twitterDescriptionClamped,
+        imageAltClamped,
+        jsonLdHeadline,
+        ...(altCandidate ? { alternativeHeadline: altCandidate } : {}),
+    };
+}
 /**
  * Render the full article HTML document with the shared chrome.
  *
@@ -65,6 +117,17 @@ export function wrapArticleHtml(options) {
     // CodeQL-safe.
     const bodyText = stripHtmlTags(options.body);
     const wordCount = bodyText.split(/\s+/u).filter((w) => w.length > 0).length;
+    // Pre-compute the per-surface SEO-budget-clamped variants of title
+    // and description. Each surface gets its own clamp tuned to the
+    // documented platform envelope (Google/Bing SERP, Facebook/LinkedIn
+    // OG, Twitter card) and the script family (Latin / CJK / RTL —
+    // CJK glyphs render at ~2× Latin pixel width, so the same byte
+    // count occupies twice the SERP width). See
+    // `src/aggregator/metadata/seo-budgets.ts` for the budget table and
+    // `analysis/methodologies/seo-headers-policy.md` § 1.1 for the
+    // documented sources of every cap.
+    const seoClamps = computeSeoClamps(options, safeLang, siteTitle);
+    const { pageTitle, ogTitleClamped, twitterTitleClamped, metaDescriptionClamped, ogDescriptionClamped, twitterDescriptionClamped, imageAltClamped, jsonLdHeadline, alternativeHeadline, } = seoClamps;
     // Build the JSON-LD image graph. Google requires NewsArticle.image
     // to be an array (or single ImageObject) with explicit width/height
     // covering at least one of the 1:1, 4:3, 16:9 aspect ratios for
@@ -92,8 +155,9 @@ export function wrapArticleHtml(options) {
     const jsonLd = {
         '@context': 'https://schema.org',
         '@type': 'NewsArticle',
-        headline: truncateHeadline(options.title),
-        description: options.description,
+        headline: jsonLdHeadline,
+        ...(alternativeHeadline ? { alternativeHeadline } : {}),
+        description: metaDescriptionClamped,
         datePublished: options.date,
         dateModified: options.date,
         inLanguage: safeLang,
@@ -165,18 +229,10 @@ export function wrapArticleHtml(options) {
     };
     const structuredData = [jsonLd, breadcrumbLd];
     const jsonLdString = JSON.stringify(structuredData).replace(/</g, '\\u003c');
-    const pageTitle = `${options.title}${getTitleSeparator(safeLang)}${siteTitle}`;
     const keywords = (options.keywords ?? []).map((keyword) => keyword.trim()).filter(Boolean);
     const keywordsMeta = keywords.length > 0
         ? `  <meta name="keywords" content="${escapeHTML(keywords.join(', '))}">\n`
         : '';
-    // Use the longer extended description for og:description/twitter:description
-    // when available so social-card previews show the full BLUF
-    // paragraph; the short meta description stays within Google's
-    // ~160-char snippet budget.
-    const socialDescription = options.extendedDescription && options.extendedDescription.length > 0
-        ? options.extendedDescription
-        : options.description;
     const ogLocaleTags = buildOgLocaleTags(safeLang);
     const twitterAttribution = buildTwitterAttributionTags();
     const twitterAttributionBlock = twitterAttribution ? `\n${twitterAttribution}` : '';
@@ -196,7 +252,7 @@ export function wrapArticleHtml(options) {
   <meta http-equiv="Content-Language" content="${safeLang}">
   <meta name="referrer" content="no-referrer">
   <title>${escapeHTML(pageTitle)}</title>
-  <meta name="description" content="${escapeHTML(options.description)}">
+  <meta name="description" content="${escapeHTML(metaDescriptionClamped)}">
 ${keywordsMeta}  <meta name="robots" content="index, follow, max-snippet:-1, max-image-preview:large">
   <meta name="author" content="${PUBLISHER_NAME}">
   <meta name="publisher" content="${PUBLISHER_NAME}">
@@ -211,15 +267,15 @@ ${hreflangLinks}
   <link rel="alternate" type="application/rss+xml" title="EU Parliament Monitor RSS" href="${BASE_URL}/rss.xml">
   <link rel="preconnect" href="https://hack23.com" crossorigin>
   <meta property="og:type" content="article">
-  <meta property="og:title" content="${escapeHTML(options.title)}">
-  <meta property="og:description" content="${escapeHTML(socialDescription)}">
+  <meta property="og:title" content="${escapeHTML(ogTitleClamped)}">
+  <meta property="og:description" content="${escapeHTML(ogDescriptionClamped)}">
   <meta property="og:url" content="${canonicalUrl}">
   <meta property="og:site_name" content="EU Parliament Monitor">
 ${ogLocaleTags}
-${buildResponsiveSocialImageMeta(`${options.title}${getTitleSeparator(safeLang)}EU Parliament Monitor`)}
+${buildResponsiveSocialImageMeta(imageAltClamped)}
   <meta name="twitter:card" content="summary_large_image">
-  <meta name="twitter:title" content="${escapeHTML(options.title)}">
-  <meta name="twitter:description" content="${escapeHTML(socialDescription)}">${twitterAttributionBlock}
+  <meta name="twitter:title" content="${escapeHTML(twitterTitleClamped)}">
+  <meta name="twitter:description" content="${escapeHTML(twitterDescriptionClamped)}">${twitterAttributionBlock}
 ${buildResponsiveIconLinks('../')}
   <link rel="manifest" href="../site.webmanifest">
   <meta name="color-scheme" content="light dark">

package/scripts/aggregator/manifest/index.d.ts

@@ -3,7 +3,7 @@
  * @description Public re-exports for the manifest bounded context.
  */
 export type { HorizonProfile, Manifest, ManifestFiles, ManifestHistoryEntry, ManifestMetadataOverride, MetadataManifest, } from './types.js';
-export { resolveArticleType, resolveDate, resolveRunId, latestGateResult, flattenManifestFiles, UNKNOWN_ARTICLE_TYPE, } from './resolver.js';
+export { resolveArticleType, resolveDate, resolveRunId, latestGateResult, flattenManifestFiles, stripRunSuffix, RUN_SUFFIX_PATTERN, UNKNOWN_ARTICLE_TYPE, } from './resolver.js';
 export { readManifest, parseManifest, type ReadManifestResult } from './reader.js';
 export { applyHorizonProfile, buildHorizonProfile } from './manifest-writer.js';
 //# sourceMappingURL=index.d.ts.map

package/scripts/aggregator/manifest/index.js

@@ -1,6 +1,6 @@
 // SPDX-FileCopyrightText: 2024-2026 Hack23 AB
 // SPDX-License-Identifier: Apache-2.0
-export { resolveArticleType, resolveDate, resolveRunId, latestGateResult, flattenManifestFiles, UNKNOWN_ARTICLE_TYPE, } from './resolver.js';
+export { resolveArticleType, resolveDate, resolveRunId, latestGateResult, flattenManifestFiles, stripRunSuffix, RUN_SUFFIX_PATTERN, UNKNOWN_ARTICLE_TYPE, } from './resolver.js';
 export { readManifest, parseManifest } from './reader.js';
 export { applyHorizonProfile, buildHorizonProfile } from './manifest-writer.js';
 //# sourceMappingURL=index.js.map

package/scripts/aggregator/manifest/resolver.d.ts

@@ -10,6 +10,30 @@
 import type { Manifest, ManifestFiles } from './types.js';
 /** Sentinel used when no schema variant supplies a usable article type. */
 export declare const UNKNOWN_ARTICLE_TYPE = "unknown";
+/**
+ * Pattern matching trailing `-run<N>` taxonomy noise that historic
+ * Stage-B writers occasionally encode into `articleType` (e.g.
+ * `committee-reports-run47`, `motions-run41`, `breaking-run193`). Also
+ * tolerates the legacy double-prefixed `motions-runmotions-run-1777010709`
+ * pattern observed in 2025 manifests where the writer concatenated the
+ * articleType and runId. The leading `-run` makes the match greedy enough
+ * to catch both single-suffix and double-prefixed forms.
+ *
+ * Exported for unit tests.
+ */
+export declare const RUN_SUFFIX_PATTERN: RegExp;
+/**
+ * Strip a trailing `-run<N>` taxonomy-noise suffix from an article-type
+ * slug, but only when doing so yields a {@link CANONICAL_ARTICLE_TYPES}
+ * token. This is conservative: a non-canonical leading token (e.g.
+ * `custom-type-run5`) is returned untouched so we never silently
+ * collapse a genuinely new article type into something it isn't.
+ *
+ * @param slug - Raw article-type slug from a manifest field
+ * @returns Canonical slug when the suffix was successfully stripped,
+ *   otherwise the original input
+ */
+export declare function stripRunSuffix(slug: string): string;
 /**
  * Resolve the article-type slug from a manifest, tolerating historic schemas.
  *
@@ -19,7 +43,10 @@ export declare const UNKNOWN_ARTICLE_TYPE = "unknown";
  *   3. `articleTypes[0]` — pre-aggregator-pipeline plural array
  *   4. `runType` — historic field on older breaking-run manifests
  *
- * Falls back to `'unknown'` when none of the above is a non-empty string.
+ * Each candidate is passed through {@link stripRunSuffix} so trailing
+ * `-run<N>` taxonomy noise never leaks into JSON-LD `articleSection`,
+ * the filesystem slug, or the SEO dump's article-type histogram. Falls
+ * back to `'unknown'` when none of the above is a non-empty string.
  *
  * @param manifest - Parsed manifest (any of the supported schemas)
  * @returns Article-type slug usable as a filename component