euparliamentmonitor 0.9.21 → 0.9.23

package/scripts/aggregator/metadata/heading-rules.d.ts

@@ -1,83 +1,4 @@
-/**
- * Headings inside an editorial artefact that carry the journalist's lede
- * paragraph (a one-paragraph summary of "what happened, why it matters").
- * When the resolver sees one of these as a `## …` heading inside the
- * editorial artefact, it prefers the first prose paragraph that follows
- * it as the description (and as a title fallback) over a generic line
- * walk. Names are matched case-insensitively against the heading text
- * (after stripping inline Markdown).
- */
-export declare const EDITORIAL_LEDE_HEADINGS: readonly string[];
-/**
- * Artifact-category prefixes that appear inside editorial-artefact H1s as
- * a structural label rather than an editorial headline (e.g. `# Synthesis
- * Summary — Week in Review (3 Apr – 1 May 2026)`). When a candidate H1
- * starts with one of these prefixes followed by a separator (em/en dash,
- * hyphen, or colon), the resolver treats it as **generic** so it does
- * not leak into the article `<title>`. Compared lower-case, with leading
- * punctuation stripped.
- */
-export declare const ARTIFACT_CATEGORY_PREFIXES: readonly string[];
-/**
- * Normalise a Markdown heading's text for comparison against the
- * editorial-lede heading whitelist. Strips inline Markdown decorations
- * (`*`, `_`, `` ` ``, `#`), then strips any leading non-alphanumeric
- * characters (emoji, punctuation, spaces) so a heading like
- * `🎯 Headline Judgement` compares equal to `headline judgement`.
- *
- * @param raw - Raw heading text (no leading hashes)
- * @returns Lower-cased, decoration-stripped heading text
- */
-export declare function normaliseHeadingText(raw: string): string;
-/**
- * Word-boundary match against an editorial-lede whitelist entry. Matches
- * when the normalised heading equals the whitelist entry exactly, or when
- * the entry is followed by any non-alphanumeric character — covering
- * localized parenthetical glosses written with ASCII or full-width
- * punctuation (e.g. `bluf (bottom line up front)`, `bluf（結論先出し）`,
- * `bluf — 핵심 결론`, `60-second read — what happened`).
- *
- * @param headingText - Normalised heading text (lower-case, decoration-stripped)
- * @param whitelistEntry - Lower-case whitelist entry from
- *                        {@link EDITORIAL_LEDE_HEADINGS}
- * @returns `true` when `headingText` begins with `whitelistEntry` at a
- *          word boundary
- */
-export declare function isLedeHeadingMatch(headingText: string, whitelistEntry: string): boolean;
-/**
- * Return `true` when an artefact-H1 begins with one of the
- * `ARTIFACT_CATEGORY_PREFIXES` followed by a separator. Such H1s
- * carry the artefact's structural label rather than a journalist's
- * headline (e.g. `# Synthesis Summary — Week in Review (3 Apr – 1 May
- * 2026)`) and must not leak into the article `<title>`.
- *
- * @param heading - Plain-text H1 (after `stripInlineMarkdown`)
- * @returns `true` when the heading is an artefact-category label
- */
-export declare function isArtifactCategoryHeading(heading: string): boolean;
-/**
- * Strip a leading or trailing artifact-category label from a heading and
- * return the editorial-topic core. When neither end carries a category
- * label, the heading is returned unchanged. When the category label is
- * the **entire** heading (e.g. `# Executive Brief`) the result is the
- * empty string.
- *
- * Examples:
- * - `Executive Brief — EU Parliament Motions` → `EU Parliament Motions`
- * - `EU Parliament Propositions — Executive Brief` → `EU Parliament Propositions`
- * - `EP10 Term Outlook — Executive Brief` → `EP10 Term Outlook`
- * - `Key Legislative Developments — Deep Analysis (2026-05-08)` → `Key Legislative Developments`
- * - `Synthesis Summary — EP Motions & Adopted Texts` → `EP Motions & Adopted Texts`
- *
- * Trailing parenthesised metadata (`(2026-05-08)`, `(May 2026)`) is also
- * stripped because it functions as a date stamp rather than editorial
- * copy. The returned core is trimmed of whitespace and trailing
- * punctuation.
- *
- * @param heading - Raw heading text (post-{@link stripInlineMarkdown})
- * @returns Editorial-topic core, or empty string when only the category survived
- */
-export declare function stripArtifactCategoryAffix(heading: string): string;
+export { EDITORIAL_LEDE_HEADINGS, ARTIFACT_CATEGORY_PREFIXES, normaliseHeadingText, isLedeHeadingMatch, isArtifactCategoryHeading, stripArtifactCategoryAffix, } from './artifact-category-heading.js';
 /**
  * Return `true` when the supplied heading matches the generic
  * `${humanize(articleType)} — ${date}` form that the aggregator writes as
@@ -85,7 +6,7 @@ export declare function stripArtifactCategoryAffix(heading: string): string;
  * separators, and matches the `breaking-breaking` variant that some
  * same-day collision runs produce.
  *
- * @param heading - Plain-text heading (post-{@link stripInlineMarkdown})
+ * @param heading - Plain-text heading (post-`stripInlineMarkdown`)
  * @param articleType - Article type slug
  * @param date - ISO date string
  * @returns `true` when the heading carries no editorial information

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

@@ -2,280 +2,104 @@
 // SPDX-License-Identifier: Apache-2.0
 /**
  * @module Aggregator/Metadata/HeadingRules
- * @description Heading-classification helpers extracted from
- * `article-metadata.ts`. Owns the editorial-lede whitelist, the artefact
- * category prefix list, the institutional-noun whitelist, and the
- * `isGenericHeading` / `stripArtifactCategoryAffix` predicates that
- * drive title-tier selection in the resolver.
+ * @description Heading-classification helpers used by the article
+ * metadata resolver. Owns:
  *
- * Pure leaf module — the only runtime dependencies are
- * {@link stripInlineMarkdown} (text-utils) and {@link humanizeSlug}
- * (slug). Re-exported through `article-metadata.ts` for back-compat
- * with existing call sites.
+ *   - {@link isGenericHeading} — the resolver's master generic-heading
+ *     predicate (drives title-tier selection).
+ *   - Internal helpers for institutional-noun, category-noun, and
+ *     `<label><sep><date>` boilerplate detection.
+ *
+ * The editorial-lede whitelist, artifact-category prefix list, and the
+ * {@link isArtifactCategoryHeading} / {@link stripArtifactCategoryAffix}
+ * helpers were extracted to `./artifact-category-heading.ts` in May 2026
+ * to keep this file under the 600-raw-line drift-guard. They are
+ * **re-exported here** so existing call sites (`lede-extractor.ts`,
+ * `artifact-walker.ts`, `article-metadata.ts`) keep working unchanged.
+ *
+ * Pure leaf module. The only runtime dependencies are
+ * {@link humanizeSlug} (slug) and the helpers re-exported from
+ * `./artifact-category-heading.js`.
  */
-import { stripInlineMarkdown } from './text-utils.js';
 import { humanizeSlug } from './slug.js';
+import { isArtifactCategoryHeading } from './artifact-category-heading.js';
+// Re-export the artifact-category surface so existing imports continue
+// to work without touching consumers.
+export { EDITORIAL_LEDE_HEADINGS, ARTIFACT_CATEGORY_PREFIXES, normaliseHeadingText, isLedeHeadingMatch, isArtifactCategoryHeading, stripArtifactCategoryAffix, } from './artifact-category-heading.js';
 /**
- * Headings inside an editorial artefact that carry the journalist's lede
- * paragraph (a one-paragraph summary of "what happened, why it matters").
- * When the resolver sees one of these as a `## …` heading inside the
- * editorial artefact, it prefers the first prose paragraph that follows
- * it as the description (and as a title fallback) over a generic line
- * walk. Names are matched case-insensitively against the heading text
- * (after stripping inline Markdown).
+ * Article-type aliases that author-templates use interchangeably with
+ * the humanized slug. `breaking` runs in particular alternate between
+ * `Breaking` and `Breaking News` in brief H1s. The aliases are matched
+ * alongside the canonical `humanizeSlug(articleType)` value so the
+ * downstream pattern + trailing-date regex pick them all up.
  */
-export const EDITORIAL_LEDE_HEADINGS = [
-    '60-second read',
-    '60 second read',
-    'sixty-second read',
-    'lede',
-    'lead',
-    'tl;dr',
-    'tldr',
-    'synopsis',
-    'in brief',
-    'at a glance',
-    'bottom line',
-    'bluf',
-    'bluf — bottom line up front',
-    'bottom line up front',
-    'executive summary',
-    'executive briefing',
-    'master narrative',
-    'overview',
-    'headline judgement',
-    'headline judgment',
-    'key findings',
-    'key judgements',
-    'key judgments',
-    'situation summary',
-    'situation report',
-    'situation update',
-];
+const ARTICLE_TYPE_ALIASES = {
+    breaking: ['Breaking News'],
+};
 /**
- * Artifact-category prefixes that appear inside editorial-artefact H1s as
- * a structural label rather than an editorial headline (e.g. `# Synthesis
- * Summary — Week in Review (3 Apr – 1 May 2026)`). When a candidate H1
- * starts with one of these prefixes followed by a separator (em/en dash,
- * hyphen, or colon), the resolver treats it as **generic** so it does
- * not leak into the article `<title>`. Compared lower-case, with leading
- * punctuation stripped.
+ * Separators observed in the wild for brief H1s mixing the
+ * article-type label with a single ISO or human-friendly date.
  */
-export const ARTIFACT_CATEGORY_PREFIXES = [
-    'actor mapping',
-    'analytical quality',
-    'breaking news analysis',
-    'coalition dynamics',
-    'commission wp alignment',
-    'committee activity report',
-    'cross run continuity',
-    'deep analysis',
-    'economic context',
-    'executive brief',
-    'executive briefing',
-    'executive intelligence brief',
-    'executive intelligence briefing',
-    'executive summary',
-    'forward indicators',
-    'historical baseline',
-    'impact matrix',
-    'intelligence assessment',
-    'intelligence briefing',
-    'intelligence synthesis summary',
-    'legislative output analysis',
-    'legislative pipeline analysis',
-    'legislative pipeline forecast',
-    'mandate fulfilment scorecard',
-    'master intelligence synthesis',
-    'mcp reliability audit',
-    'methodology reflection',
-    'monthly outlook',
-    'motions analysis',
-    'parliamentary calendar projection',
-    'pestle analysis',
-    'political intelligence brief',
-    'political risk',
-    'political threat landscape',
-    'presidency trio context',
-    'propositions analysis',
-    'quantitative swot',
-    'risk assessment',
-    'risk matrix',
-    'risk scoring',
-    'scenario forecast',
-    'seat projection',
-    'significance classification',
-    'situation report',
-    'situation summary',
-    'stakeholder analysis',
-    'stakeholder impact',
-    'stakeholder map',
-    'swot analysis',
-    'synthesis summary',
-    'threat assessment',
-    'threat model',
-    'voting patterns',
-    'weekly outlook',
-    'wildcards blackswans',
-];
+const GENERIC_HEADING_SEPARATORS = [' — ', ' - ', ' – ', ': ', ' ', ' | ', ', '];
 /**
- * Normalise a Markdown heading's text for comparison against the
- * editorial-lede heading whitelist. Strips inline Markdown decorations
- * (`*`, `_`, `` ` ``, `#`), then strips any leading non-alphanumeric
- * characters (emoji, punctuation, spaces) so a heading like
- * `🎯 Headline Judgement` compares equal to `headline judgement`.
- *
- * @param raw - Raw heading text (no leading hashes)
- * @returns Lower-cased, decoration-stripped heading text
+ * Date-shape character class: digits, dashes (ISO) plus letters and
+ * single spaces (human-friendly forms like `8 April 2026`). Single-day
+ * only — date *ranges* are preserved as editorial scope-window content.
  */
-export function normaliseHeadingText(raw) {
-    return stripInlineMarkdown(raw)
-        .replace(/[*_`#]+/g, '')
-        .replace(/^[^A-Za-z0-9]+/, '')
-        .trim()
-        .toLowerCase();
-}
+const GENERIC_HEADING_DATE_SHAPE = '[\\d][\\d\\-]*|\\d{1,2}\\s+[A-Za-z]+\\s+\\d{4}';
 /**
- * Word-boundary match against an editorial-lede whitelist entry. Matches
- * when the normalised heading equals the whitelist entry exactly, or when
- * the entry is followed by any non-alphanumeric character — covering
- * localized parenthetical glosses written with ASCII or full-width
- * punctuation (e.g. `bluf (bottom line up front)`, `bluf（結論先出し）`,
- * `bluf — 핵심 결론`, `60-second read — what happened`).
+ * Aliases used for one article-type slug, including the canonical
+ * humanised slug plus any registered aliases.
  *
- * @param headingText - Normalised heading text (lower-case, decoration-stripped)
- * @param whitelistEntry - Lower-case whitelist entry from
- *                        {@link EDITORIAL_LEDE_HEADINGS}
- * @returns `true` when `headingText` begins with `whitelistEntry` at a
- *          word boundary
+ * @param articleType - Article-type slug
+ * @returns Ordered list of label aliases
  */
-export function isLedeHeadingMatch(headingText, whitelistEntry) {
-    if (headingText === whitelistEntry)
-        return true;
-    if (!headingText.startsWith(whitelistEntry))
-        return false;
-    const next = headingText.charAt(whitelistEntry.length);
-    // Word boundary — anything that is not an ASCII letter/digit is a
-    // separator we accept. This works uniformly across ASCII parentheses,
-    // CJK full-width brackets `（`, dashes `— – -`, colons `:`, and the
-    // ideographic full-width colon `：`.
-    return next === '' || !/[a-z0-9]/.test(next);
+function resolveLabelAliases(articleType) {
+    const human = humanizeSlug(articleType);
+    return [human, ...(ARTICLE_TYPE_ALIASES[articleType] ?? [])];
 }
 /**
- * Return `true` when an artefact-H1 begins with one of the
- * `ARTIFACT_CATEGORY_PREFIXES` followed by a separator. Such H1s
- * carry the artefact's structural label rather than a journalist's
- * headline (e.g. `# Synthesis Summary — Week in Review (3 Apr – 1 May
- * 2026)`) and must not leak into the article `<title>`.
+ * Match an exact `<prefix?><label><sep><date>` shape, including the
+ * `EU Parliament ` / `EP ` prefix variants and the redundant
+ * `<label> <label> — <date>` form occasionally emitted by same-day
+ * collision runs.
  *
- * @param heading - Plain-text H1 (after `stripInlineMarkdown`)
- * @returns `true` when the heading is an artefact-category label
+ * @param normalized - Heading text after whitespace collapse
+ * @param label - Article-type label to test against
+ * @param date - ISO date string
+ * @returns `true` when the heading matches a known literal shape
  */
-export function isArtifactCategoryHeading(heading) {
-    const normalized = normaliseCategoryHeading(heading);
-    if (normalized === '')
-        return false;
-    for (const prefix of ARTIFACT_CATEGORY_PREFIXES) {
-        if (normalized === prefix)
+function matchesLiteralLabelDateShape(normalized, label, date) {
+    for (const sep of GENERIC_HEADING_SEPARATORS) {
+        const p = `${label}${sep}${date}`;
+        if (normalized === p)
             return true;
-        if (normalized.startsWith(`${prefix} —`) ||
-            normalized.startsWith(`${prefix} –`) ||
-            normalized.startsWith(`${prefix} -`) ||
-            normalized.startsWith(`${prefix}:`)) {
+        if (normalized === `EU Parliament ${p}`)
             return true;
-        }
-        if (normalized.endsWith(` — ${prefix}`) ||
-            normalized.endsWith(` – ${prefix}`) ||
-            normalized.endsWith(` - ${prefix}`) ||
-            normalized.endsWith(`: ${prefix}`)) {
+        if (normalized === `EP ${p}`)
             return true;
-        }
-    }
-    return false;
-}
-/**
- * Strip a leading or trailing artifact-category label from a heading and
- * return the editorial-topic core. When neither end carries a category
- * label, the heading is returned unchanged. When the category label is
- * the **entire** heading (e.g. `# Executive Brief`) the result is the
- * empty string.
- *
- * Examples:
- * - `Executive Brief — EU Parliament Motions` → `EU Parliament Motions`
- * - `EU Parliament Propositions — Executive Brief` → `EU Parliament Propositions`
- * - `EP10 Term Outlook — Executive Brief` → `EP10 Term Outlook`
- * - `Key Legislative Developments — Deep Analysis (2026-05-08)` → `Key Legislative Developments`
- * - `Synthesis Summary — EP Motions & Adopted Texts` → `EP Motions & Adopted Texts`
- *
- * Trailing parenthesised metadata (`(2026-05-08)`, `(May 2026)`) is also
- * stripped because it functions as a date stamp rather than editorial
- * copy. The returned core is trimmed of whitespace and trailing
- * punctuation.
- *
- * @param heading - Raw heading text (post-{@link stripInlineMarkdown})
- * @returns Editorial-topic core, or empty string when only the category survived
- */
-export function stripArtifactCategoryAffix(heading) {
-    const trimmed = heading.trim();
-    if (trimmed === '')
-        return '';
-    const sortedPrefixes = [...ARTIFACT_CATEGORY_PREFIXES].sort((a, b) => b.length - a.length);
-    const normalized = normaliseCategoryHeading(trimmed);
-    const skip = trimmed.length - normalized.length;
-    const visible = trimmed.slice(skip < 0 ? 0 : skip);
-    const visibleClean = visible.replace(/\s*\([^)]{1,80}\)\s*$/u, '').trim();
-    const normalizedClean = normaliseCategoryHeading(visibleClean);
-    for (const prefix of sortedPrefixes) {
-        for (const sep of [' — ', ' – ', ' - ', ': ']) {
-            const candidate = `${prefix}${sep}`;
-            if (normalizedClean.startsWith(candidate)) {
-                const core = visibleClean.slice(candidate.length).trim();
-                return cleanupAffixCore(core);
-            }
-        }
-        for (const sep of [' — ', ' – ', ' - ', ': ']) {
-            const candidate = `${sep}${prefix}`;
-            if (normalizedClean.endsWith(candidate)) {
-                const core = visibleClean.slice(0, visibleClean.length - candidate.length).trim();
-                return cleanupAffixCore(core);
-            }
-        }
-        if (normalizedClean === prefix)
-            return '';
     }
-    return trimmed;
-}
-/**
- * Tidy the editorial-topic core returned by
- * {@link stripArtifactCategoryAffix}: drop trailing parenthesised
- * metadata (`(2026-05-08)`, `(May 2026)`) and trailing punctuation. When
- * stripping leaves the string too short to be meaningful (<5 chars),
- * return the empty string so callers fall through to lower tiers.
- *
- * @param core - Heading with the category label already stripped
- * @returns Cleaned editorial-topic core, or empty string when too short
- */
-function cleanupAffixCore(core) {
-    const withoutTrailingParens = core.replace(/\s*\([^)]{1,80}\)\s*$/u, '').trim();
-    const withoutTrailingPunct = withoutTrailingParens.replace(/[—–:;,.\s-]+$/u, '').trim();
-    if (withoutTrailingPunct.length < 5)
-        return '';
-    return withoutTrailingPunct;
+    const labelRedundant = `${label} ${label}`;
+    return normalized === `${labelRedundant} — ${date}`;
 }
 /**
- * Lower-case, decoration-stripped form used by the artifact-category
- * matchers. Strips inline Markdown, leading non-alphanumeric runs (emoji,
- * decoration), and collapses whitespace to a single space.
+ * Match `<prefix?><label><sep-or-space><any-date>` patterns where the
+ * date token can be any ISO / human / single-day-range shape. Anchored
+ * to end-of-string so it cannot fire on editorial sentences that
+ * happen to contain a date token mid-clause.
  *
- * @param raw - Raw heading text
- * @returns Lower-case normalised form
+ * @param normalized - Heading text after whitespace collapse
+ * @param label - Article-type label to test against
+ * @returns `true` when the heading matches the trailing-date shape
  */
-function normaliseCategoryHeading(raw) {
-    return stripInlineMarkdown(raw)
-        .trim()
-        .toLowerCase()
-        .replace(/^[^a-z0-9]+/, '')
-        .replace(/\s+/g, ' ');
+function matchesTrailingDateShape(normalized, label) {
+    const trailingDateOnly = new RegExp(`^(?:EU Parliament |EP )?${escapeRegex(label)}\\s*[—–\\-|,:]\\s*(?:${GENERIC_HEADING_DATE_SHAPE})$`, 'u');
+    if (trailingDateOnly.test(normalized))
+        return true;
+    // Same shape but label followed directly by a date with whitespace only
+    // (e.g. `Breaking News 2026-04-01`).
+    const labelSpaceDate = new RegExp(`^(?:EU Parliament |EP )?${escapeRegex(label)}\\s+(?:${GENERIC_HEADING_DATE_SHAPE})$`, 'u');
+    return labelSpaceDate.test(normalized);
 }
 /**
  * Return `true` when the supplied heading matches the generic
@@ -284,7 +108,7 @@ function normaliseCategoryHeading(raw) {
  * separators, and matches the `breaking-breaking` variant that some
  * same-day collision runs produce.
  *
- * @param heading - Plain-text heading (post-{@link stripInlineMarkdown})
+ * @param heading - Plain-text heading (post-`stripInlineMarkdown`)
  * @param articleType - Article type slug
  * @param date - ISO date string
  * @returns `true` when the heading carries no editorial information
@@ -295,27 +119,12 @@ export function isGenericHeading(heading, articleType, date) {
         return true;
     if (isArtifactCategoryHeading(normalized))
         return true;
-    const human = humanizeSlug(articleType);
-    const patterns = [
-        `${human} — ${date}`,
-        `${human} - ${date}`,
-        `${human} – ${date}`,
-        `${human}: ${date}`,
-        `${human} ${date}`,
-    ];
-    const humanRedundant = `${human} ${human}`;
-    for (const p of patterns) {
-        if (normalized === p)
-            return true;
-        if (normalized === `EU Parliament ${p}`)
+    for (const label of resolveLabelAliases(articleType)) {
+        if (matchesLiteralLabelDateShape(normalized, label, date))
             return true;
-        if (normalized === `${humanRedundant} — ${date}`)
+        if (matchesTrailingDateShape(normalized, label))
             return true;
     }
-    const trailingDateOnly = new RegExp(`^${escapeRegex(human)}\\s*[—–-]\\s*[\\d-]+$`, 'u');
-    if (trailingDateOnly.test(normalized)) {
-        return true;
-    }
     if (isCategoryNounHeading(normalized, articleType))
         return true;
     if (isBareInstitutionalHeading(normalized))

package/scripts/aggregator/metadata/keyword-filters.d.ts

@@ -0,0 +1,60 @@
+/**
+ * @module Aggregator/Metadata/KeywordFilters
+ * @description Cross-site keyword catalogue and noise-token filter used
+ * by {@link buildSeoKeywords} in `resolve-helpers.ts`.
+ *
+ * Two responsibilities:
+ *
+ *   1. **Always-on cross-site keywords** ({@link CROSS_SITE_KEYWORDS})
+ *      are prepended to every article's `<meta name="keywords">` list
+ *      regardless of language, so search-engine discovery of the
+ *      Hack23 civic-tech portfolio (EU Parliament Monitor +
+ *      Riksdagsmonitor + CIA) is consistent across all 14 localized
+ *      surfaces. The user explicitly requested
+ *      `riksdagsmonitor, political intelligence, riksdag, regeringen`
+ *      (the sister Swedish-Parliament project) plus EP analogues.
+ *
+ *   2. **Noise-token rejection** ({@link isNoiseKeywordToken}) drops
+ *      the UUID-fragment tokens (`77fc920c`, `3a76`, `9db5`, …) and
+ *      synthetic run-id slugs (`propositions-run261-1779431162`) that
+ *      the previous keyword extractor leaked into `<head>` when a
+ *      brief mentioned its own run id editorially (e.g.
+ *      `Analysis run 77fc920c-3a76-4813-9db5-43a7e9acc25e returned
+ *      0 classified actors`).
+ *
+ * Pure leaf module — no imports.
+ */
+/**
+ * Cross-site SEO keywords prepended to every article in every
+ * language. Order is meaningful: stronger civic-tech-portfolio terms
+ * first so they appear ahead of the per-article-type keywords when
+ * the 16-entry budget is exceeded.
+ */
+export declare const CROSS_SITE_KEYWORDS: readonly string[];
+/**
+ * Decide whether a single keyword token should be discarded as noise.
+ *
+ * The current rules reject tokens that:
+ *
+ *   - Look like a UUID hex chunk: ≥4 chars and consist solely of the
+ *     `[0-9a-f]` alphabet **and** contain at least one digit (so
+ *     real English words like `dead` / `face` survive). Tokens of
+ *     length ≥8 are always rejected (a real English word of that
+ *     length composed exclusively of hex letters is vanishingly rare;
+ *     the allowlist guards the short cases).
+ *   - Are mostly digits (≥80 % digit characters) — runtime epoch
+ *     suffixes such as `1779431162` and committee-codeoid mashes like
+ *     `2024k1234`.
+ *   - Start with `run` and end with all-digits (`run261`, `run17`),
+ *     the per-run slug suffix the aggregator stamps onto run ids.
+ *   - Match the full opaque-runId shape `<type>-run<digits>-<digits>`
+ *     after a strip / normalization round-trip.
+ *
+ * Returns `false` for normal vocabulary so the keyword list stays
+ * useful — every reject path is intentionally narrow.
+ *
+ * @param token - Single token candidate
+ * @returns `true` when the token should be dropped from keywords
+ */
+export declare function isNoiseKeywordToken(token: string): boolean;
+//# sourceMappingURL=keyword-filters.d.ts.map