euparliamentmonitor 0.9.13 → 0.9.14

package/scripts/aggregator/article-metadata.js

@@ -51,40 +51,12 @@ import path from 'path';
 import { ALL_LANGUAGES, getLocalizedString } from '../constants/language-core.js';
 import { BREAKING_NEWS_TITLES, COMMITTEE_REPORTS_TITLES, ELECTION_CYCLE_TITLES, LOCALIZED_KEYWORDS, MONTH_AHEAD_TITLES, MONTHLY_REVIEW_TITLES, MOTIONS_TITLES, PROPOSITIONS_TITLES, QUARTER_AHEAD_TITLES, QUARTER_IN_REVIEW_TITLES, TERM_OUTLOOK_TITLES, WEEK_AHEAD_TITLES, WEEKLY_REVIEW_TITLES, YEAR_AHEAD_TITLES, YEAR_IN_REVIEW_TITLES, } from '../constants/language-articles.js';
 import { resolveLocalizedBriefHighlight } from './editorial-brief-resolver.js';
-/** Maximum `<meta description>` length we will emit. */
-const DESCRIPTION_MAX_LENGTH = 180;
-/** Target minimum `<meta description>` length before we append context. */
-const DESCRIPTION_MIN_LENGTH = 140;
-/**
- * Length below which a raw description is considered too short to stand
- * on its own and gets enriched with date/context. Independent from
- * {@link DESCRIPTION_MIN_LENGTH} (which controls sentence-boundary
- * truncation behaviour). Set lower than DESCRIPTION_MIN_LENGTH so a
- * clean 100-140 char prose lede is preserved verbatim instead of being
- * padded with date/context boilerplate.
- */
-const ENRICHMENT_TRIGGER_LENGTH = 100;
-/** Maximum `<title>` length — anything longer is truncated with an ellipsis. */
-const TITLE_MAX_LENGTH = 140;
-/**
- * Soft target for headline-style titles produced as a fallback from
- * BLUF/lede prose. When the candidate exceeds `TITLE_MAX_LENGTH`, the
- * truncator first looks for a natural clause boundary
- * (`.`, `:`, `—`, `;`) inside the `[HEADLINE_SOFT_MIN, TITLE_MAX_LENGTH]`
- * window and breaks there instead of mid-clause-with-ellipsis. This
- * turns a 137-character truncated prose paragraph into a complete
- * journalistic clause, which scans much better in news cards and SERP
- * snippets without sacrificing the keyword-rich opening.
- */
-const HEADLINE_SOFT_MIN = 60;
-/**
- * Punctuation marks that signal a natural clause boundary inside a
- * BLUF / lede paragraph. Listed in preferred-break order: a colon or
- * em-dash that introduces a list of consequences is the best break,
- * full stops are next, and semicolons last. Single ASCII space is
- * always a fallback boundary handled separately.
- */
-const HEADLINE_CLAUSE_BOUNDARIES = [': ', ' — ', ' – ', '. ', '; '];
+// Text-utility constants + helpers — extracted into the `metadata/`
+// bounded context as pure leaf-module functions. Re-exported here for
+// back-compat with existing call sites; new code should import directly
+// from `./metadata/text-utils.js`.
+import { DESCRIPTION_MAX_LENGTH, EXTENDED_DESCRIPTION_MAX_LENGTH, ENRICHMENT_TRIGGER_LENGTH, shouldSkipDescriptionLine, stripLeadingProseLabel, stripInlineMarkdown, truncateDescription, truncateExtendedDescription, truncateTitle, extractFirstSentence, } from './metadata/text-utils.js';
+export { shouldSkipDescriptionLine, stripLeadingProseLabel, stripInlineMarkdown, truncateDescription, truncateExtendedDescription, truncateTitle, extractFirstSentence, } from './metadata/text-utils.js';
 /** Localized labels used to enrich short or duplicate-prone meta descriptions. */
 const SEO_CONTEXT_LABELS = {
     en: {
@@ -315,435 +287,6 @@ const ARTIFACT_CATEGORY_PREFIXES = [
     'weekly outlook',
     'wildcards blackswans',
 ];
-/**
- * Emoji-banner prefixes that Stage-B agents use to decorate metadata rows
- * (e.g. `📋 Analysis Owner:`). Any line starting with one of these is
- * metadata, never prose.
- */
-const EMOJI_BANNER_CHARS = [
-    '📋',
-    '📅',
-    '🔍',
-    '🏛',
-    '📰',
-    '📊',
-    '🏷',
-    '📈',
-    '📉',
-    '⚠',
-    '🔔',
-    '🎯',
-    '🗳',
-    '🏢',
-    '📄',
-];
-/**
- * Label prefixes that a prose description must never start with. Every
- * entry matches case-insensitively at the start of a trimmed line, followed
- * by optional space and a colon.
- */
-const METADATA_LINE_PREFIXES = [
-    'Admiralty Grade',
-    'Analysis Date',
-    'Analysis Owner',
-    'Article Type',
-    'Article Window',
-    'Assessment Date',
-    'Briefing',
-    'Briefing Date',
-    'Classification',
-    'Classification Date',
-    'Confidence',
-    'Confidence in Evidence',
-    'Data Sources',
-    'Date',
-    'Document Type',
-    'Filing Date',
-    'Generated',
-    'Horizon',
-    'IMF Status',
-    'Last Updated',
-    'Parliamentary Status',
-    'Parliamentary Term',
-    'Period',
-    'Prepared',
-    'Purpose',
-    'Region',
-    'Reporting',
-    'Reporting Period',
-    'Reporting Window',
-    'Run',
-    'Run ID',
-    'Series',
-    'Series Run',
-    'Source',
-    'Sources',
-    'SPDX-FileCopyrightText',
-    'SPDX-License-Identifier',
-    'Topic',
-    'Type',
-    'WEP Band',
-    'WEP Grade',
-    'Window',
-];
-/**
- * Return `true` when a line cannot serve as a prose description. Rejects
- * Markdown structural lines (headings, blockquotes, tables, HTML),
- * mermaid/chart directives, emoji-banner metadata rows, and the known
- * `Key: value` banners that Stage-B agents emit as artefact preamble.
- *
- * @param line - Trimmed line from the aggregated Markdown source
- * @returns `true` when the line is not prose and should be skipped
- */
-export 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;
-    if (line.startsWith('---') || line.startsWith('==='))
-        return true;
-    if (line.startsWith('```') || line.startsWith('~~~'))
-        return true;
-    if (line.startsWith('%%'))
-        return true;
-    if (/^title\s/i.test(line))
-        return true;
-    if (EMOJI_BANNER_CHARS.some((char) => line.startsWith(char)))
-        return true;
-    const labelSource = line.replace(/^\*+/, '').replace(/^\*\*/, '').replace(/^_+/, '').trim();
-    for (const prefix of METADATA_LINE_PREFIXES) {
-        const lower = labelSource.toLowerCase();
-        const prefixLower = prefix.toLowerCase();
-        if (lower.startsWith(`${prefixLower}:`) ||
-            lower.startsWith(`${prefixLower} :`) ||
-            lower.startsWith(`${prefixLower}**:`) ||
-            lower.startsWith(`${prefixLower}*:`)) {
-            return true;
-        }
-    }
-    if (/^[-*_=~.]{3,}$/.test(line))
-        return true;
-    if (isLocalizedBannerRow(line))
-        return true;
-    return false;
-}
-/**
- * Language-agnostic banner-row detector. Stage-B artefacts open with a
- * metadata banner of the shape
- *   `**Date:** 2026-05-15 | **Type:** Breaking | **Run:** breaking-run-001`
- * and its localized siblings — notably Japanese / Chinese / Korean briefs
- * which place the full-width colon `：` **inside** the bold span
- * (`**日付：**`) rather than after it. The `METADATA_LINE_PREFIXES` table
- * only covers the English vocabulary; this helper catches the structural
- * shape directly: a line that starts with `**`, contains at least one
- * `|` separator, and carries two-or-more bold key markers that end with
- * — or are followed by — an ASCII colon `:` or full-width colon `：`.
- * Banner rows look identical in every language we publish, so detecting
- * them here keeps localized briefs from leaking their first banner line
- * into the `<meta description>`.
- *
- * @param line - Trimmed source line
- * @returns `true` when the line is a banner row in any locale
- */
-function isLocalizedBannerRow(line) {
-    if (!line.startsWith('**'))
-        return false;
-    if (!line.includes('|'))
-        return false;
-    const inside = (line.match(/\*\*[^*]+[:：]\s*\*\*/g) ?? []).length;
-    const after = (line.match(/\*\*[^*]+\*\*\s*[:：]/g) ?? []).length;
-    return inside + after >= 2;
-}
-/**
- * Strip inline Markdown decorations so we can use the remaining text as
- * plain-text meta-tag content. Removes link syntax, emphasis, inline code
- * backticks, and HTML-entity fragments that the Markdown source sometimes
- * smuggles in. Keeps the visible text readable.
- *
- * @param raw - Trimmed Markdown line
- * @returns Plain-text variant
- */
-/**
- * Strip a leading all-caps prose label (e.g. `SITUATION:`, `KEY MOTION:`,
- * `BLUF:`, `BOTTOM LINE:`, `TIER-1:`) from a prose line. These labels
- * are common in BLUF-style editorial writing — they survive
- * {@link stripInlineMarkdown} (which strips the `**bold**` wrapper but
- * keeps the literal text) and would otherwise leak into the SEO
- * description as a confusing all-caps shout.
- *
- * Matches up to 4 hyphenated all-caps tokens, optionally followed by a
- * digit suffix (`TIER-1`), terminating at a colon. Returns the original
- * line when no opener is present.
- *
- * @param line - Plain prose line (post-{@link stripInlineMarkdown})
- * @returns Line with the all-caps opener removed
- */
-export function stripLeadingProseLabel(line) {
-    const colonIdx = line.indexOf(': ');
-    if (colonIdx < 2 || colonIdx > 80)
-        return line;
-    const label = line.slice(0, colonIdx);
-    const rest = line.slice(colonIdx + 2).trim();
-    if (rest.length < 20)
-        return line;
-    if (!/^[A-Z][A-Z0-9 -]{1,79}$/.test(label))
-        return line;
-    if (label.length < 3)
-        return line;
-    return rest;
-}
-/**
- * Strip inline Markdown decorations so we can use the remaining text as
- * plain-text meta-tag content. Removes link syntax, emphasis, inline code
- * backticks, and HTML-entity fragments that the Markdown source sometimes
- * smuggles in. Keeps the visible text readable.
- *
- * @param raw - Trimmed Markdown line
- * @returns Plain-text variant
- */
-export function stripInlineMarkdown(raw) {
-    return raw
-        .replace(/!\[([^\]\n]{0,500})\]\(([^)\n]{0,500})\)/g, '$1')
-        .replace(/\[([^\]\n]{1,500})\]\(([^)\n]{0,500})\)/g, '$1')
-        .replace(/`([^`\n]{1,500})`/g, '$1')
-        .replace(/\*\*([^*\n]{1,500})\*\*/g, '$1')
-        .replace(/__([^_\n]{1,500})__/g, '$1')
-        .replace(/\*([^*\n]{1,500})\*/g, '$1')
-        .replace(/_([^_\n]{1,500})_/g, '$1')
-        .replace(/~~([^~\n]{1,500})~~/g, '$1')
-        .replace(/\s+/g, ' ')
-        .trim();
-}
-/** Connector / determiner words that read as broken copy when they are
- *  the final token before a truncation ellipsis. */
-const TRAILING_STOP_WORDS = new Set([
-    'the',
-    'a',
-    'an',
-    'of',
-    'to',
-    'for',
-    'in',
-    'on',
-    'at',
-    'by',
-    'and',
-    'or',
-    'with',
-    'from',
-]);
-/** Trailing characters we always strip before appending our own ellipsis,
- *  so we never emit double-ellipsis or stray punctuation. */
-const TRAILING_PUNCT = /[.,;:—\-…\s]/u;
-/**
- * Repeatedly strip trailing stop-words (separated by a single space) and
- * trailing punctuation (including any pre-existing ellipsis). Implemented
- * imperatively to avoid super-linear regex backtracking on the
- * `(?:\s+stop-word)+$` pattern flagged by `security/detect-unsafe-regex`.
- *
- * @param input - Pre-clipped string to clean up
- * @returns Cleaned string with no trailing stop-words or punctuation
- */
-function stripTrailingStopWordsAndPunctuation(input) {
-    let result = input;
-    let changed = true;
-    while (changed) {
-        changed = false;
-        while (result.length > 0 && TRAILING_PUNCT.test(result.charAt(result.length - 1))) {
-            result = result.slice(0, -1);
-            changed = true;
-        }
-        const lastSpace = result.lastIndexOf(' ');
-        if (lastSpace >= 0) {
-            const tail = result.slice(lastSpace + 1).toLowerCase();
-            if (TRAILING_STOP_WORDS.has(tail)) {
-                result = result.slice(0, lastSpace);
-                changed = true;
-            }
-        }
-    }
-    return result;
-}
-/**
- * Clamp a string to `DESCRIPTION_MAX_LENGTH` characters, appending
- * an ellipsis when truncation actually happens. Does not break words if
- * avoidable — a trailing partial word is trimmed back to the previous
- * space first.
- *
- * @param text - Raw description text
- * @returns Truncated description with trailing ellipsis when clipped
- */
-export function truncateDescription(text) {
-    if (text.length <= DESCRIPTION_MAX_LENGTH)
-        return text;
-    const cut = text.slice(0, DESCRIPTION_MAX_LENGTH - 1);
-    // Prefer the last full sentence terminator within the cut so we don't
-    // end on a dangling determiner ("…year. The"). Period/!/? followed by
-    // a space marks a clean boundary. Only honour the boundary when it
-    // sits past the soft minimum so we keep enough body text to be useful.
-    const sentenceEnd = Math.max(cut.lastIndexOf('. '), cut.lastIndexOf('! '), cut.lastIndexOf('? '));
-    if (sentenceEnd >= DESCRIPTION_MIN_LENGTH) {
-        return cut.slice(0, sentenceEnd + 1).replace(/\s+$/, '');
-    }
-    const lastSpace = cut.lastIndexOf(' ');
-    let safe = lastSpace > DESCRIPTION_MAX_LENGTH - 60 ? cut.slice(0, lastSpace) : cut;
-    // Drop dangling stop-words and trailing punctuation/ellipsis so we
-    // never emit broken copy ("…year. The" → "…year.") or double-ellipsis
-    // ("The……") when the upstream input already carried an ellipsis.
-    safe = stripTrailingStopWordsAndPunctuation(safe);
-    return `${safe}…`;
-}
-/**
- * Clamp a title to `TITLE_MAX_LENGTH` characters in the same
- * word-boundary-preserving fashion as {@link truncateDescription}.
- *
- * @param text - Raw title text
- * @returns Truncated title with trailing ellipsis when clipped
- */
-export function truncateTitle(text) {
-    if (text.length <= TITLE_MAX_LENGTH)
-        return text;
-    // Prefer ending at a natural clause boundary inside the
-    // `[HEADLINE_SOFT_MIN, TITLE_MAX_LENGTH]` window so the truncated
-    // title reads as a complete journalistic clause rather than a
-    // mid-sentence prose snippet. Iterate boundaries in priority order;
-    // when a candidate falls in the window, break there and drop the
-    // ellipsis since the result is grammatically complete.
-    const search = text.slice(0, TITLE_MAX_LENGTH);
-    for (const boundary of HEADLINE_CLAUSE_BOUNDARIES) {
-        const idx = search.lastIndexOf(boundary);
-        if (idx >= HEADLINE_SOFT_MIN) {
-            const clean = stripTrailingStopWordsAndPunctuation(text.slice(0, idx));
-            if (clean.length >= HEADLINE_SOFT_MIN)
-                return clean;
-        }
-    }
-    const cut = text.slice(0, TITLE_MAX_LENGTH - 1);
-    const lastSpace = cut.lastIndexOf(' ');
-    let safe = lastSpace > TITLE_MAX_LENGTH - 40 ? cut.slice(0, lastSpace) : cut;
-    safe = stripTrailingStopWordsAndPunctuation(safe);
-    return `${safe}…`;
-}
-/**
- * Return the first complete sentence from a prose paragraph, suitable
- * for use as a fallback editorial title when the artefact H1 is
- * categorical (e.g. `# EU Parliament Committee Reports`) and the
- * resolver must derive `<title>` from the BLUF / lede summary instead.
- *
- * A "sentence" is the prefix up to the first sentence-terminator
- * (`. `, `! `, `? `, `; `) inside the `[HEADLINE_SOFT_MIN,
- * TITLE_MAX_LENGTH]` window. Common abbreviations (`Q1.`, `Q2.`,
- * `H1.`, `H2.`, `Mr.`, `Mrs.`, `e.g.`, `i.e.`, `vs.`) are skipped
- * so they don't terminate the sentence prematurely. When no
- * acceptable terminator exists in the window, returns the entire
- * input unchanged so {@link truncateTitle} can handle clause-boundary
- * truncation downstream.
- *
- * This produces journalistically clean titles even for the
- * propositions / committee-reports cases where the BLUF paragraph
- * opens with a single long sentence that exceeds 140 chars —
- * `truncateTitle` then breaks on a clause boundary, and the result is
- * still grammatical because the input was a sentence prefix rather
- * than an arbitrary paragraph slice.
- *
- * @param paragraph - Prose paragraph (post-{@link stripInlineMarkdown})
- * @returns First sentence, or the original paragraph when none can be
- *   identified within the soft-min window
- */
-export function extractFirstSentence(paragraph) {
-    const trimmed = paragraph.trim();
-    if (trimmed.length <= HEADLINE_SOFT_MIN)
-        return trimmed;
-    // Limit terminator search to TITLE_MAX_LENGTH * 1.5 — beyond that
-    // we'd rather let truncateTitle clause-truncate the original
-    // paragraph than return a too-long first sentence.
-    const window = trimmed.slice(0, Math.floor(TITLE_MAX_LENGTH * 1.5));
-    // Skip common abbreviations that contain a period inside a token
-    // (Q1., e.g., i.e., vs., Mr., Mrs., No., U.S., E.U.). We walk
-    // candidate terminator positions; a position counts only when the
-    // char before it is *not* part of a known abbreviation token.
-    const terminators = ['. ', '! ', '? ', '; '];
-    let bestIdx = -1;
-    for (const t of terminators) {
-        let from = HEADLINE_SOFT_MIN;
-        let idx;
-        while ((idx = window.indexOf(t, from)) !== -1) {
-            if (!isAbbreviationBoundary(window, idx) && idx < window.length - 1) {
-                if (bestIdx === -1 || idx < bestIdx)
-                    bestIdx = idx;
-                break;
-            }
-            from = idx + t.length;
-        }
-    }
-    if (bestIdx >= HEADLINE_SOFT_MIN) {
-        return trimmed.slice(0, bestIdx + 1).trim();
-    }
-    return trimmed;
-}
-/**
- * Abbreviation tokens (lowercase, including the trailing period) that
- * should NOT count as sentence terminators when {@link extractFirstSentence}
- * scans for a `.` boundary. Single-letter all-caps initials
- * (`U.S.`, `E.U.`) are handled by the all-caps-initial check below.
- */
-const ABBREVIATION_PREFIXES = [
-    'mr.',
-    'mrs.',
-    'ms.',
-    'dr.',
-    'st.',
-    'no.',
-    'vs.',
-    'e.g.',
-    'i.e.',
-    'etc.',
-    'cf.',
-    'al.',
-    // EP fiscal-year and quarter shorthand: Q1., Q2., Q3., Q4., H1., H2., FY.
-    'q1.',
-    'q2.',
-    'q3.',
-    'q4.',
-    'h1.',
-    'h2.',
-    'fy.',
-];
-/**
- * Check whether the character preceding the `.` at `idx` in `text`
- * indicates an abbreviation (so the `.` is not a sentence terminator).
- * Matches the {@link ABBREVIATION_PREFIXES} table and the all-caps
- * single-letter initials pattern (`U.S.`, `E.U.`).
- *
- * @param text - Source text (lowercased segment + original mixed-case)
- * @param idx - Index of the `.` character in `text`
- * @returns `true` when the period at `idx` is part of an abbreviation
- */
-function isAbbreviationBoundary(text, idx) {
-    // All-caps single-letter initial like `U.S.` or `E.U.` — char at
-    // idx-1 is a capital letter, and idx-2 is either start of string,
-    // whitespace, or another single-letter+period pair.
-    if (idx >= 1) {
-        const prev = text.charCodeAt(idx - 1);
-        const isUpperLetter = prev >= 65 && prev <= 90;
-        if (isUpperLetter && (idx === 1 || text[idx - 2] === ' ' || text[idx - 2] === '.')) {
-            return true;
-        }
-    }
-    // ABBREVIATION_PREFIXES lookup — scan backwards from `.` to find the
-    // start of the word, then compare lowercased.
-    let start = idx;
-    while (start > 0 && /[a-zA-Z]/u.test(text[start - 1] ?? ''))
-        start--;
-    const token = text.slice(start, idx + 1).toLowerCase();
-    return ABBREVIATION_PREFIXES.includes(token);
-}
 /**
  * Return the first Markdown H1 (`# …`) in the supplied text, stripped of
  * the leading `#` and trailing anchor syntax. Returns an empty string when
@@ -905,6 +448,41 @@ export function extractLedeAfterHeading(markdown) {
         return '';
     return truncateDescription(buf.lines.join(' '));
 }
+/**
+ * Same parsing rules as {@link extractLedeAfterHeading} but with a
+ * larger byte budget so the full BLUF paragraph (typically 200-300
+ * characters in the editorial style guide) is captured for use as
+ * `og:description` / `twitter:description`. Returns the joined
+ * paragraph clamped via {@link truncateExtendedDescription} (which
+ * returns `''` when the result wouldn't be longer than the regular
+ * meta description).
+ *
+ * @param markdown - Brief body (SPDX preamble already stripped)
+ * @returns Extended lede paragraph, or `''` when not worth emitting
+ */
+export function extractExtendedLedeAfterHeading(markdown) {
+    const state = { inFence: false, inLede: false };
+    const buf = { lines: [], byteCount: 0 };
+    for (const raw of markdown.split('\n')) {
+        const line = raw.trim();
+        const directive = classifyLedeLine(line, state.inFence, state.inLede, buf.lines.length > 0);
+        const action = applyLedeDirective(directive, state, buf.lines.length > 0);
+        if (action === 'break')
+            break;
+        if (action === 'continue')
+            continue;
+        const collect = collectProseLine(line, buf);
+        if (collect === 'continue')
+            continue;
+        if (collect === 'break')
+            break;
+        if (buf.byteCount >= EXTENDED_DESCRIPTION_MAX_LENGTH)
+            break;
+    }
+    if (buf.lines.length === 0)
+        return '';
+    return truncateExtendedDescription(buf.lines.join(' '));
+}
 /**
  * Normalise a Markdown heading's text for comparison against the
  * editorial-lede heading whitelist. Strips inline Markdown decorations
@@ -2112,189 +1690,12 @@ function templateForType(lang, articleType, inputs) {
             };
     }
 }
-/** Milliseconds in one UTC day — used by date-window derivation helpers. */
-const MS_PER_DAY = 86_400_000;
-/**
- * Parse an ISO date and return the `[start, end]` week range as ISO
- * strings. Week starts on Monday and ends on the following Sunday.
- *
- * @param date - ISO date string (`YYYY-MM-DD`)
- * @returns `{ start, end }` both in `YYYY-MM-DD` form
- */
-export function deriveWeekRange(date) {
-    const parsed = parseIsoDate(date);
-    if (!parsed)
-        return { start: date, end: date };
-    const day = parsed.getUTCDay();
-    const shift = (day + 6) % 7;
-    const startMs = parsed.getTime() - shift * MS_PER_DAY;
-    const endMs = startMs + 6 * MS_PER_DAY;
-    return { start: formatIsoDate(new Date(startMs)), end: formatIsoDate(new Date(endMs)) };
-}
-/**
- * Return the D-36 → D-8 reporting window for the `week-in-review`
- * article type. EP roll-call voting data is published with a 2–6 week
- * lag, so using the most-recent 7 days structurally produces a
- * vote-empty dataset. Shifting 8 days back and widening to 28 days
- * (start = D-36, end = D-8) ensures the window always contains at
- * least one full EP plenary week with published roll-call data
- * (ADR-006). Direction is consistent with the workflow's
- * `DATE_FROM` (start = D-36) → `DATE_TO` (end = D-8) variables.
- *
- * @param date - ISO article date string (`YYYY-MM-DD`) — typically TODAY
- * @returns `{ start: D-36, end: D-8 }` both as `YYYY-MM-DD` ISO strings
- */
-export function deriveReportingWindowForWeekInReview(date) {
-    const parsed = parseIsoDate(date);
-    if (!parsed)
-        return { start: date, end: date };
-    return {
-        start: formatIsoDate(new Date(parsed.getTime() - 36 * MS_PER_DAY)),
-        end: formatIsoDate(new Date(parsed.getTime() - 8 * MS_PER_DAY)),
-    };
-}
-/**
- * Return a human-friendly month label for an ISO date — English month
- * name + four-digit year (e.g. `April 2026`). The non-English template
- * generators accept this same label verbatim because they interpolate it
- * into a localized sentence rather than translating the month itself.
- *
- * @param date - ISO date string
- * @returns Month label, or the input when parsing fails
- */
-export function deriveMonthLabel(date) {
-    const parsed = parseIsoDate(date);
-    if (!parsed)
-        return date;
-    const monthNames = [
-        'January',
-        'February',
-        'March',
-        'April',
-        'May',
-        'June',
-        'July',
-        'August',
-        'September',
-        'October',
-        'November',
-        'December',
-    ];
-    const name = monthNames[parsed.getUTCMonth()] ?? '';
-    return `${name} ${parsed.getUTCFullYear()}`.trim();
-}
-/**
- * Return a quarter label for an ISO date — `Q<n> <YYYY>` (e.g. `Q2 2026`).
- * Used by `quarter-ahead` and `quarter-in-review` title generators.
- *
- * @param date - ISO date string
- * @returns Quarter label, or the input when parsing fails
- */
-export function deriveQuarterLabel(date) {
-    const parsed = parseIsoDate(date);
-    if (!parsed)
-        return date;
-    const quarter = Math.floor(parsed.getUTCMonth() / 3) + 1;
-    return `Q${quarter} ${parsed.getUTCFullYear()}`;
-}
-/**
- * Return a four-digit year label for an ISO date. Used by `year-ahead`
- * and `year-in-review` title generators.
- *
- * @param date - ISO date string
- * @returns Year label, or the input when parsing fails
- */
-export function deriveYearLabel(date) {
-    const parsed = parseIsoDate(date);
-    if (!parsed)
-        return date;
-    return String(parsed.getUTCFullYear());
-}
-/**
- * EP-term constants — keep these in sync with
- * {@link analysis/methodologies/electoral-cycle-methodology.md}.
- *  - EP10: 16 Jul 2024 → ~end of June 2029
- *  - EP11: ~Jul 2029 → ~Jun 2034
- */
-const EP10_START_YEAR = 2024;
-const EP10_END_YEAR = 2029;
-const EP11_END_YEAR = 2034;
-const EP_ELECTION_MONTH = 6; // June
-/**
- * Return the EP-term label for an ISO date — `EP10 → 2029` or `EP11 → 2034`.
- * Used by `term-outlook` title generator.
- *
- * @param date - ISO date string
- * @returns Term label, or the input when parsing fails
- */
-export function deriveTermLabel(date) {
-    const parsed = parseIsoDate(date);
-    if (!parsed)
-        return date;
-    const year = parsed.getUTCFullYear();
-    const month = parsed.getUTCMonth() + 1;
-    if (year < EP10_START_YEAR)
-        return `EP9 → ${EP10_START_YEAR}`;
-    if (year < EP10_END_YEAR || (year === EP10_END_YEAR && month <= EP_ELECTION_MONTH)) {
-        return `EP10 → ${EP10_END_YEAR}`;
-    }
-    if (year < EP11_END_YEAR || (year === EP11_END_YEAR && month <= EP_ELECTION_MONTH)) {
-        return `EP11 → ${EP11_END_YEAR}`;
-    }
-    const yearsBeyond = year - EP11_END_YEAR;
-    const offset = month <= EP_ELECTION_MONTH ? 0 : 1;
-    const termsBeyond = Math.floor((yearsBeyond - 1 + offset) / 5) + 1;
-    const termIndex = 11 + termsBeyond;
-    const termEnd = EP11_END_YEAR + 5 * termsBeyond;
-    return `EP${termIndex} → ${termEnd}`;
-}
-/**
- * Return the election-cycle label for an ISO date — pairs the outgoing
- * and incoming EP terms with the election year (e.g. `EP10 → EP11 (2029)`).
- * Used by the `election-cycle` title generator.
- *
- * @param date - ISO date string
- * @returns Cycle label, or the input when parsing fails
- */
-export function deriveElectionCycleLabel(date) {
-    const parsed = parseIsoDate(date);
-    if (!parsed)
-        return date;
-    const year = parsed.getUTCFullYear();
-    if (year <= EP10_END_YEAR)
-        return `EP10 → EP11 (${EP10_END_YEAR})`;
-    if (year <= EP11_END_YEAR)
-        return `EP11 → EP12 (${EP11_END_YEAR})`;
-    const cyclesBeyond = Math.ceil((year - EP11_END_YEAR) / 5);
-    const electionYear = EP11_END_YEAR + 5 * cyclesBeyond;
-    const out = 11 + cyclesBeyond;
-    return `EP${out} → EP${out + 1} (${electionYear})`;
-}
-/**
- * Parse an ISO date string as UTC midnight. Returns `null` for malformed
- * input so callers can skip month/week derivation gracefully.
- *
- * @param iso - ISO date string
- * @returns Parsed `Date` or `null`
- */
-function parseIsoDate(iso) {
-    if (!/^\d{4}-\d{2}-\d{2}$/.test(iso))
-        return null;
-    const parsed = new Date(`${iso}T00:00:00Z`);
-    return Number.isNaN(parsed.getTime()) ? null : parsed;
-}
-/**
- * Format a `Date` as `YYYY-MM-DD` in UTC.
- *
- * @param d - Date object
- * @returns ISO date string
- */
-function formatIsoDate(d) {
-    const y = d.getUTCFullYear();
-    const m = String(d.getUTCMonth() + 1).padStart(2, '0');
-    const day = String(d.getUTCDate()).padStart(2, '0');
-    return `${y}-${m}-${day}`;
-}
+// Date-label helpers — extracted into the `metadata/` bounded context as
+// pure leaf-module functions. Re-exported here for back-compat with
+// existing call sites; new code should import directly from
+// `./metadata/date-labels.js`.
+import { deriveWeekRange, deriveReportingWindowForWeekInReview, deriveMonthLabel, deriveQuarterLabel, deriveYearLabel, deriveTermLabel, deriveElectionCycleLabel, } from './metadata/date-labels.js';
+export { deriveWeekRange, deriveReportingWindowForWeekInReview, deriveMonthLabel, deriveQuarterLabel, deriveYearLabel, deriveTermLabel, deriveElectionCycleLabel, } from './metadata/date-labels.js';
 /**
  * Extract a manifest override value for a single language. Accepts either
  * a plain string (applied to every language) or a `LanguageMap` object.
@@ -2334,6 +1735,7 @@ function resolveEditorialContent(opts) {
             return {
                 headline: highlight.headline,
                 summary: highlight.summary,
+                extendedSummary: extractExtendedLedeAfterHeading(markdown),
             };
         }
         if (highlight?.summary) {
@@ -2342,10 +1744,12 @@ function resolveEditorialContent(opts) {
     }
     const aggregatedH1 = extractFirstH1(markdown);
     const aggregatedSummary = extractStrongProseLine(markdown);
+    const aggregatedExtended = extractExtendedLedeAfterHeading(markdown);
     if (aggregatedH1 && !isGenericHeading(aggregatedH1, articleType, date)) {
         return {
             headline: truncateTitle(aggregatedH1),
             summary: artefactSummary || aggregatedSummary,
+            extendedSummary: aggregatedExtended,
         };
     }
     const summary = artefactSummary || aggregatedSummary;
@@ -2357,9 +1761,13 @@ function resolveEditorialContent(opts) {
         // to clause-boundary truncation downstream when the sentence
         // itself overruns TITLE_MAX_LENGTH.
         const firstSentence = extractFirstSentence(summary);
-        return { headline: truncateTitle(firstSentence), summary };
+        return {
+            headline: truncateTitle(firstSentence),
+            summary,
+            extendedSummary: aggregatedExtended,
+        };
     }
-    return { headline: '', summary: '' };
+    return { headline: '', summary: '', extendedSummary: '' };
 }
 /**
  * Pick the per-language SEO title from the resolved editorial pair and
@@ -2603,10 +2011,23 @@ function resolveOneLanguage(input) {
         : composeContextualDescription(input.lang, rawDescription, editorial, input.date, input.runId);
     const truncatedTitle = truncateTitle(title);
     const truncatedDescription = truncateDescription(description);
+    // The extended description tracks the same source as the short
+    // description: when a manifest description overrides, use it
+    // verbatim (no point synthesising an extended form from the brief
+    // when the editor explicitly chose the manifest copy); otherwise
+    // use the editorial extended summary lifted from the brief BLUF.
+    // `truncateExtendedDescription` returns `''` when the candidate
+    // wouldn't be longer than the regular meta description, so callers
+    // can fall back to {@link description} via a simple `||`.
+    const extendedSource = manifestDescription
+        ? manifestDescription
+        : editorial.extendedSummary || rawDescription;
+    const truncatedExtendedDescription = truncateExtendedDescription(extendedSource);
     const source = manifestTitle || manifestDescription ? 'manifest' : perLanguage.source;
     return {
         title: truncatedTitle,
         description: truncatedDescription,
+        extendedDescription: truncatedExtendedDescription,
         keywords: buildSeoKeywords(input.lang, input.articleType, input.date, input.runId, truncatedTitle, truncatedDescription),
         source,
     };
@@ -2641,6 +2062,7 @@ function resolvePerLanguageEditorial(input) {
                 editorial: {
                     headline: localized.headline,
                     summary: localized.summary,
+                    extendedSummary: localized.extendedSummary,
                 },
                 source: 'localized-brief',
             };
@@ -2656,7 +2078,7 @@ function resolvePerLanguageEditorial(input) {
     // Nothing editorial at all → caller will fall back to the localized
     // template.
     return {
-        editorial: { headline: '', summary: '' },
+        editorial: { headline: '', summary: '', extendedSummary: '' },
         source: 'template',
     };
 }