@cyanheads/pubmed-mcp-server 2.10.20 → 2.10.22

package/dist/services/ncbi/formatting/citation-formatter.js

@@ -4,6 +4,27 @@
  * Pure TypeScript, zero dependencies, Workers-compatible.
  * @module src/services/ncbi/formatting/citation-formatter
  */
+/**
+ * NCBI `RefType`s that qualify the cited article itself — it was retracted,
+ * corrected, or questioned. A link from a record that is itself a notice
+ * (`RetractionOf`, `ErratumFor`) and comment or update links are not among them:
+ * citing a retraction notice is not citing a retracted work. (#215)
+ */
+export const CITATION_NOTICE_TYPES = [
+    'RetractionIn',
+    'RetractedandRepublishedIn',
+    'ExpressionOfConcernIn',
+    'ErratumIn',
+    'CorrectedandRepublishedIn',
+];
+const NOTICE_TYPES = new Set(CITATION_NOTICE_TYPES);
+function isCitationNotice(link) {
+    return NOTICE_TYPES.has(link.refType);
+}
+/** The record's linked notices that qualify the cited article, in NCBI's order. */
+export function citationNotices(article) {
+    return (article.commentsCorrections ?? []).filter(isCitationNotice);
+}
 /**
  * Whether a record cites a book rather than a journal article. Dispatch is on
  * `recordType`, never on `publicationTypes`: a Bookshelf record's publication
@@ -171,30 +192,13 @@ function terminate(text) {
     return text.endsWith('.') ? text : `${text}.`;
 }
 /**
- * Split a pages string like "45-67" into start and end components.
- * Handles en-dashes, em-dashes, and hyphens. Expands PubMed's truncated-end
- * convention (e.g., "737-8" → { start: "737", end: "738" }, "1639-41" →
- * "1639"/"1641") so downstream RIS/BibTeX consumers see absolute page numbers.
- */
-function splitPages(pages) {
-    if (!pages)
-        return {};
-    const parts = pages.split(/[-\u2013\u2014]/).map((p) => p.trim());
-    let [start, end] = parts;
-    if (start && end && end.length < start.length) {
-        end = start.slice(0, start.length - end.length) + end;
-    }
-    if (start && end)
-        return { start, end };
-    return start ? { start } : {};
-}
-/**
- * The electronic article locator that stands in for a page range, or undefined.
+ * The electronic article locator that Vancouver, BibTeX, and RIS print in place
+ * of a page range, or undefined.
  *
  * Returns a value only when the record carries no pagination. Publishers that
  * report the article number as both `Pagination` and `ELocationID` (PLoS ONE,
  * Scientific Reports) are already covered by the `pages` rendering, and printing
- * both would duplicate the number in every style.
+ * both would duplicate the number. APA and MLA use {@link articleNumber}.
  */
 function articleLocator(journal) {
     if (journal?.pages)
@@ -202,9 +206,256 @@ function articleLocator(journal) {
     return journal?.elocationId || undefined;
 }
 /**
- * Collapse internal whitespace (including embedded newlines from structured
- * abstracts) to single spaces. Strict RIS parsers treat blank lines as record
- * terminators, so abstract text must be flattened before emission.
+ * The article number APA and MLA cite in place of a page range: the record's
+ * non-DOI locator when it has no pagination, or when its pagination is that
+ * same value — NCBI copies an article number into `MedlinePgn` (Frontiers
+ * `662953`, Springer `42`). A page that matches no locator stays a page, though
+ * some are article numbers the record cannot identify (an Elsevier `105471`
+ * beside an `S…` pii looks exactly like a Lancet page `2413`). (#217)
+ *
+ * A dashless `MedlinePgn` also matches the locator with its hyphens removed:
+ * American Society for Microbiology journals deposit `e0364621` beside pii
+ * `e03646-21`. The locator's form is the one cited, as the publisher numbers
+ * the article. Pagination holding a dash is a range and never matches. (#231)
+ */
+function articleNumber(journal) {
+    const locator = journal?.elocationId;
+    if (!locator)
+        return;
+    const { pages } = journal;
+    if (!pages || pages === locator)
+        return locator;
+    return !PAGE_DASH.test(pages) && pages === locator.replaceAll('-', '') ? locator : undefined;
+}
+/** A dash NLM's pagination puts between the ends of a range. */
+const PAGE_DASH = /[-\u{2013}\u{2014}]/u;
+/**
+ * One page or range inside NLM's pagination: a start, then optionally a dash
+ * and an end. An item starts only after a separator, so a long run without a
+ * dash is read once rather than retried from each of its characters.
+ */
+const PAGE_ITEM = /(?<![^\s,;\u{2013}\u{2014}-])([^\s,;\u{2013}\u{2014}-]+)(?:\s*[-\u{2013}\u{2014}]\s*([^\s,;\u{2013}\u{2014}-]+))?/gu;
+/** The digits a page ends on, matched from the start of their run only. */
+const TRAILING_DIGITS = /(?<!\d)\d+$/;
+/**
+ * The end of a range as NLM's elision means it. An end made of digits is
+ * written against the start's digits and takes the start's prefix (`637-41` →
+ * `641`, `S5-15` → `S15`, `1255.e1-5` → `1255.e5`); any other end is already
+ * whole (`878.e8`, `e65`, `xv`).
+ */
+function fullPageEnd(start, end) {
+    const digits = /^\d+$/.test(end) ? TRAILING_DIGITS.exec(start) : null;
+    if (!digits)
+        return end;
+    const [run] = digits;
+    const elided = end.length < run.length;
+    return `${start.slice(0, digits.index)}${elided ? run.slice(0, run.length - end.length) : ''}${end}`;
+}
+/**
+ * NLM's pagination with each range written in full (`637-41` → `637-641`,
+ * `S5-15` → `S5-S15`), joined by `dash` or by NLM's own when none is given; the
+ * rest of NLM's text stays as given — a list's separators and labels too
+ * (`1407-8; author reply 1409-10` → `1407-1408; author reply 1409-1410`). APA
+ * joins with an en dash (`637–641`); BibTeX keeps NLM's. (#217, #228)
+ */
+function expandPageRanges(pages, dash) {
+    return pages.replace(PAGE_ITEM, (item, start, end) => {
+        if (end === undefined)
+            return item;
+        const join = dash ?? item.slice(start.length, item.length - end.length);
+        return `${start}${join}${fullPageEnd(start, end)}`;
+    });
+}
+/** A token that can be a page: it carries a digit or is a Roman numeral, unlike a label (`author reply`). */
+const PAGE_TOKEN = /\d|^[ivxlcdm]+$/i;
+/**
+ * The first and last page of NLM's pagination, for RIS `SP` and `EP`: the start
+ * of the first page or range, and the end of the last range written in full or
+ * the last listed page (`1308, 1310-4` → `1308`/`1314`, `xii-xv` → `xii`/`xv`,
+ * `S5-15` → `S5`/`S15`). A single page or article number has no end, and a
+ * label in a list (`author reply`) is not a page. NCBI's own `StartPage` and
+ * `EndPage` agree on every shape. (#228)
+ */
+function splitPages(pages) {
+    const items = [...pages.matchAll(PAGE_ITEM)];
+    const [, start] = items[0] ?? [];
+    const pageItems = items.filter(([, first = '', last]) => last !== undefined || PAGE_TOKEN.test(first));
+    const final = pageItems.at(-1);
+    if (!final || (pageItems.length === 1 && final[2] === undefined)) {
+        return { start, end: undefined };
+    }
+    const [, first = '', last] = final;
+    return { start, end: last === undefined ? first : fullPageEnd(first, last) };
+}
+/**
+ * Words APA 7 keeps lowercase in a title unless one opens it or its subtitle:
+ * articles, coordinating conjunctions, and prepositions of three letters or
+ * fewer — plus the articles, conjunctions, and prepositions of the non-English
+ * journal names NLM catalogs, in NLM's unaccented spelling: Romance (`Annales
+ * de dermatologie et de venereologie`), Germanic and Scandinavian (`Beitrage
+ * zur gerichtlichen Medizin`), Slavic (`Akusherstvo i ginekologiia`), and the
+ * Hungarian, Turkish, Finnish, Japanese, and Latin ones NLM's titles use. A
+ * word that is also a romanized Chinese syllable stays out (`za` of `za zhi`,
+ * `ai`, `nei`), as does one NLM uses for an abbreviation (`ed.`, `med.`).
+ */
+const APA_MINOR_WORDS = new Set([
+    ...['a', 'an', 'the', 'and', 'but', 'for', 'nor', 'or', 'so', 'yet'],
+    ...['as', 'at', 'by', 'in', 'of', 'off', 'on', 'per', 'to', 'up', 'via'],
+    ...['de', 'des', 'du', 'la', 'le', 'les', 'et', 'sur', 'au', 'aux', 'pour', 'par', 'dans'],
+    ...['avec', 'un', 'une', 'el', 'los', 'las', 'y', 'e', 'al', 'para', 'por', 'o', 'os', 'em'],
+    ...['no', 'na', 'ao', 'da', 'das', 'do', 'dos', 'di', 'del', 'della', 'dello', 'delle', 'dei'],
+    ...['degli', 'alla', 'alle', 'fra', 'si', 'din', 'pentru', 'der', 'die', 'dem', 'den', 'und'],
+    ...['für', 'fur', 'zu', 'zur', 'im', 'am', 'auf', 'aus', 'mit', 'uber', 'van', 'von', 'voor'],
+    ...['en', 'het', 'tot', 'op', 'uit', 'te', 'og', 'och', 'af', 'til', 'fran', 'mot', 'vir'],
+    ...['i', 'v', 'w', 'z', 'u', 'po', 'pro', 'nad', 've', 'es', 'az', 'ja', 'ad'],
+]);
+/** MLA 9 also keeps a preposition of any length lowercase. */
+const MLA_MINOR_WORDS = new Set([
+    ...APA_MINOR_WORDS,
+    ...['about', 'above', 'across', 'against', 'along', 'among', 'around', 'behind', 'below'],
+    ...['beneath', 'beside', 'between', 'beyond', 'during', 'from', 'into', 'onto', 'over'],
+    ...['through', 'throughout', 'toward', 'towards', 'under', 'upon', 'with', 'within'],
+    ...['without'],
+]);
+/** One hyphen-delimited part of a word: surrounding punctuation, then lowercase letters. */
+const LOWERCASE_WORD = /^([^\p{L}\p{N}]*)(\p{Ll}+(?:['\u{2019}]\p{Ll}+)*)([^\p{L}\p{N}]*)$/u;
+/** A French or Italian elided particle and the word it joins (`d'ophtalmologie`, `dell'apparato`). */
+const ELISION = /^(d|l|qu|dell|dall|nell|all|sull)(['\u{2019}])(.+)$/u;
+/** Brands NLM writes in lowercase, which keep it (`npj quantum materials`). */
+const LOWERCASE_BRANDS = new Set(['npj']);
+const capitalize = (word) => `${word.charAt(0).toUpperCase()}${word.slice(1)}`;
+/**
+ * Title-case a journal name, which NLM catalogs in sentence case (`Frontiers in
+ * pediatrics`), for APA and MLA, which cite it in title case. Only a word part
+ * written entirely in lowercase letters gains a capital — each part of a
+ * hyphenated word on its own — so a name NLM already cases (`eLife`, `mBio`,
+ * `JAMA`, `PloS`) and any part carrying a digit stay as written, as do a
+ * lowercase brand and a single-letter prefix (`npj`, `i-Perception`, `e-SPEN`).
+ * An elision capitalizes the word, not its particle (`d'Ophtalmologie`). A minor
+ * word stays lowercase unless it opens the name, or a subtitle or parallel title
+ * after a colon, semicolon, or ` = ` (`CA: A Cancer Journal`). (#217)
+ */
+function titleCaseJournal(title, minorWords) {
+    let opensPhrase = true;
+    const caseWord = (word, opens) => {
+        if (!opens && minorWords.has(word))
+            return word;
+        const elision = ELISION.exec(word);
+        if (!elision)
+            return capitalize(word);
+        const [, particle = '', apostrophe = '', rest = ''] = elision;
+        return `${opens ? capitalize(particle) : particle}${apostrophe}${caseWord(rest, false)}`;
+    };
+    return title
+        .split(/(\s+)/)
+        .map((token) => {
+        if (!/\S/.test(token))
+            return token;
+        if (token === ':' || token === ';' || token === '=') {
+            opensPhrase = true;
+            return token;
+        }
+        const parts = token.split('-');
+        const cased = parts
+            .map((part, index) => {
+            const match = LOWERCASE_WORD.exec(part);
+            if (!match)
+                return part;
+            const [, lead = '', word = '', trail = ''] = match;
+            const isPrefix = index === 0 && parts.length > 1 && word.length === 1;
+            if (isPrefix || LOWERCASE_BRANDS.has(word))
+                return part;
+            return `${lead}${caseWord(word, opensPhrase && index === 0)}${trail}`;
+        })
+            .join('-');
+        opensPhrase = /[:;]$/.test(token);
+        return cased;
+    })
+        .join('');
+}
+/** NLM's year qualifier: `Journal of women's health (2002)`. */
+const YEAR_QUALIFIER = /^\d{4}$/;
+/** An edition or format qualifier: `(Clinical research ed.)`, `(Landmark edition)`, `(Print)`. */
+const EDITION_QUALIFIER = /(?:\bed\.|\bedition)$|(?:^|: )(?:Online|Print)$/i;
+/** An abbreviation naming an edition: `Med Clin (Engl Ed)`, `Angew Chem Int Ed Engl`. */
+const EDITION_ABBREVIATION = /\b(?:Ed|Edn|Edition)\b/i;
+/**
+ * A place qualifier, `Place, Region` with an optional year: `London, England`,
+ * `New York, N.Y.`, `Chicago, Ill. : 1978`. Neither part holds a parenthesis,
+ * semicolon, or digit, and the region absorbs any space before the year's
+ * colon, so no two parts compete for a character and a near-miss fails in
+ * linear time.
+ */
+const PLACE_QUALIFIER = /^[^,():;\d]+,[^():;\d]+(?::\s*\d{4})?$/;
+/**
+ * The parenthetical a title ends on, a nested one read as one unit
+ * (`Bulletin (Hospital for Joint Diseases (New York, N.Y.))`), with the text
+ * before it; undefined when the title does not end on a balanced one.
+ */
+function trailingParenthetical(title) {
+    if (!title.endsWith(')'))
+        return;
+    let depth = 0;
+    for (let i = title.length - 1; i >= 0; i--) {
+        if (title[i] === ')')
+            depth++;
+        else if (title[i] === '(' && --depth === 0) {
+            return { base: title.slice(0, i).trimEnd(), inner: title.slice(i + 1, -1) };
+        }
+    }
+    return;
+}
+/**
+ * The journal name APA and MLA cite: NLM's title less the place, year, or
+ * edition NLM appends to tell same-named journals apart (`Lancet (London,
+ * England)` → `Lancet`, `Journal of women's health (2002)`, `BMJ (Clinical
+ * research ed.)`). The journal prints no such text. An edition stays when the
+ * ISO abbreviation names it too, as it is then a journal of its own (`Medicina
+ * clinica (English ed.)`, `Med Clin (Engl Ed)`). Any other parenthetical names
+ * the journal and stays: an issuing body (`Proceedings (Baylor University.
+ * Medical Center)`), a subject, a one-word place. Over NLM's journal list the
+ * rule is wrong on 18 of the 2,005 titles NLM's catalog settles — a place-shaped
+ * issuing body (`Collegian (Royal College of Nursing, Australia)`) is dropped,
+ * and a place with no region (`Medical record (New York : 1934)`) is kept. (#227)
+ */
+function citedJournalName(title, isoAbbreviation) {
+    const group = trailingParenthetical(title);
+    if (!group?.base)
+        return title;
+    const { base, inner } = group;
+    if (YEAR_QUALIFIER.test(inner))
+        return base;
+    if (EDITION_QUALIFIER.test(inner)) {
+        return EDITION_ABBREVIATION.test(isoAbbreviation ?? '') ? title : base;
+    }
+    return PLACE_QUALIFIER.test(inner) ? base : title;
+}
+/**
+ * Author keywords, then MeSH descriptor names, each term once: a term equal to
+ * an earlier one ignoring case is dropped, so the author's spelling and position
+ * win (`Deep learning` over MeSH `Deep Learning`). (#217)
+ */
+function mergedKeywords(article) {
+    const terms = new Map();
+    const candidates = [
+        ...(article.keywords ?? []),
+        ...(article.meshTerms ?? []).map((m) => m.descriptorName),
+    ];
+    for (const term of candidates) {
+        if (!term)
+            continue;
+        const key = term.toLowerCase();
+        if (!terms.has(key))
+            terms.set(key, term);
+    }
+    return [...terms.values()];
+}
+/**
+ * Collapse each whitespace run to one space and trim the ends. The parser trims
+ * only a field's ends, so a line break inside a title, name, or keyword — or
+ * between a structured abstract's paragraphs — would otherwise reach a
+ * citation: an untagged RIS line, or a blank one that strict parsers read as
+ * the end of the record. Every style writes record text through this. (#233)
  */
 function collapseWhitespace(text) {
     return text.replace(/\s+/g, ' ').trim();
@@ -254,8 +505,23 @@ function escapeBibtex(text) {
 // ---------------------------------------------------------------------------
 // Author formatters
 // ---------------------------------------------------------------------------
+/*
+ * A generational suffix (`Jr`, `2nd`) renders verbatim as NCBI writes it, as
+ * NCBI's own citation export prints it. It qualifies a surname, so a name with
+ * no surname renders without it. Inverted name forms set it off with a comma
+ * (`Frenck, R. W., Jr`); natural-order and NLM forms append it after a space
+ * (`R. W. Frenck Jr`, `Frenck RW Jr`). (#226)
+ */
+/** `, Jr` for an inverted name form; `''` without a suffix or a surname. */
+function invertedSuffix(author) {
+    return author.lastName && author.suffix ? `, ${author.suffix}` : '';
+}
+/** ` Jr` for a natural-order or NLM name form; `''` without a suffix or a surname. */
+function trailingSuffix(author) {
+    return author.lastName && author.suffix ? ` ${author.suffix}` : '';
+}
 /**
- * Format a single author in APA style: `Last, F. M.`
+ * Format a single author in APA style: `Last, F. M.`, or `Last, F. M., Jr`.
  * Collective/group authors return the group name directly.
  */
 function formatAuthorApa(author) {
@@ -270,7 +536,7 @@ function formatAuthorApa(author) {
             .map((part) => `${part[0]}.`)
             .join(' ');
     if (!initials)
-        return last;
+        return `${last}${invertedSuffix(author)}`;
     // Extract only letter characters (Unicode-aware: preserve Á, Ö, É, etc.),
     // format each as "X." separated by spaces.
     const formatted = Array.from(initials.replace(/[^\p{L}]/gu, ''))
@@ -278,7 +544,7 @@ function formatAuthorApa(author) {
         .join(' ');
     if (!last)
         return formatted;
-    return `${last}, ${formatted}`;
+    return `${last}, ${formatted}${invertedSuffix(author)}`;
 }
 /**
  * Format the full author list for APA 7th edition.
@@ -305,21 +571,19 @@ function formatAuthorsApa(authors) {
 }
 /**
  * Format a single author in MLA style.
- * First listed author: `Last, First Middle.`
- * Subsequent authors: `First Middle Last`
+ * First listed author: `Last, First Middle`, or `Last, First Middle, Jr`.
+ * Subsequent authors: `First Middle Last`, or `First Middle Last Jr`.
  */
 function formatAuthorMla(author, isFirst) {
     if (author.collectiveName)
         return author.collectiveName;
     const last = author.lastName ?? '';
     const first = author.firstName ?? '';
-    if (!last && !first)
-        return '';
-    if (!first)
-        return last;
     if (!last)
         return first;
-    return isFirst ? `${last}, ${first}` : `${first} ${last}`;
+    if (isFirst)
+        return `${first ? `${last}, ${first}` : last}${invertedSuffix(author)}`;
+    return `${first ? `${first} ${last}` : last}${trailingSuffix(author)}`;
 }
 /**
  * Format the full author list for MLA 9th edition.
@@ -342,27 +606,28 @@ function formatAuthorsMla(authors) {
     return `${formatAuthorMla(first, true)}, et al.`;
 }
 /**
- * Format a single author in BibTeX style: `{Last}, {First}`
+ * Format a single author in BibTeX style: `{Last}, First`, or the three-part
+ * `{Last}, Jr, First` when the name carries a suffix — the comma after `Jr`
+ * stays even with no first name, so the suffix is never read as one.
  */
 function formatAuthorBibtex(author) {
     if (author.collectiveName)
         return `{${escapeBibtex(author.collectiveName)}}`;
     const last = author.lastName ? escapeBibtex(author.lastName) : '';
     const first = author.firstName ? escapeBibtex(author.firstName) : '';
-    if (!last && !first)
-        return '';
-    if (!first)
-        return `{${last}}`;
     if (!last)
         return first;
-    return `{${last}}, ${first}`;
+    if (author.suffix)
+        return `{${last}}, ${escapeBibtex(author.suffix)},${first && ` ${first}`}`;
+    return first ? `{${last}}, ${first}` : `{${last}}`;
 }
 // ---------------------------------------------------------------------------
 // APA 7th Edition
 // ---------------------------------------------------------------------------
 /**
  * Format an editor in APA's "In E. E. Editor (Ed.)," position: initials first,
- * surname last — the inverse of the author position.
+ * surname last — the inverse of the author position — and a suffix after the
+ * surname (`M. P. Adam Jr`).
  */
 function formatEditorApa(editor) {
     if (editor.collectiveName)
@@ -378,7 +643,7 @@ function formatEditorApa(editor) {
             .map((c) => `${c}.`)
             .join(' ')
         : '';
-    return [initials, editor.lastName].filter(Boolean).join(' ');
+    return [initials, editor.lastName].filter(Boolean).join(' ') + trailingSuffix(editor);
 }
 /** APA editor list: comma-separated, `& ` before the last name. */
 function formatEditorsApa(editors) {
@@ -443,7 +708,7 @@ function formatApaBook(article) {
     const url = bookAccessUrl(article);
     if (url)
         parts.push(url);
-    return parts.join(' ');
+    return collapseWhitespace(parts.join(' '));
 }
 /**
  * Format a PubMed article as an APA 7th edition citation.
@@ -456,7 +721,8 @@ function formatApaBook(article) {
  * ```
  * Title. (Year). *Journal*, *Volume*(Issue), Pages. https://doi.org/DOI
  * ```
- * A Bookshelf record routes to {@link formatApaBook}.
+ * A retracted article ends on `(Retraction published <RefSource>)`. A Bookshelf
+ * record routes to {@link formatApaBook}.
  */
 export function formatApa(article) {
     if (isBookRecord(article))
@@ -487,21 +753,21 @@ export function formatApa(article) {
     // Journal, volume, issue, pages
     const journal = article.journalInfo;
     if (journal?.title) {
-        let journalPart = `*${journal.title}*`;
+        const name = citedJournalName(journal.title, journal.isoAbbreviation);
+        let journalPart = `*${titleCaseJournal(name, APA_MINOR_WORDS)}*`;
         if (journal.volume) {
             journalPart += `, *${journal.volume}*`;
             if (journal.issue) {
                 journalPart += `(${journal.issue})`;
             }
         }
-        if (journal.pages) {
-            journalPart += `, ${journal.pages}`;
+        // APA 7 p. 294-295: an article number takes the page range's place
+        const number = articleNumber(journal);
+        if (number) {
+            journalPart += `, Article ${number}`;
         }
-        else {
-            // APA 7 p. 294-295: an article number takes the page range's place
-            const locator = articleLocator(journal);
-            if (locator)
-                journalPart += `, Article ${locator}`;
+        else if (journal.pages) {
+            journalPart += `, ${expandPageRanges(journal.pages, '\u{2013}')}`;
         }
         journalPart += '.';
         parts.push(journalPart);
@@ -510,18 +776,23 @@ export function formatApa(article) {
     if (article.doi) {
         parts.push(`https://doi.org/${article.doi}`);
     }
-    return parts.join(' ');
+    // APA 7 notes a retraction after the DOI. Its own form names the notice's
+    // year, full journal title, and location, which `RefSource` does not carry
+    // apart; NCBI's exporter fills the parenthetical with `RefSource` instead.
+    const retractions = citationNotices(article)
+        .filter((notice) => notice.refType === 'RetractionIn')
+        .map((notice) => stripTrailingPeriod(notice.refSource));
+    if (retractions.length > 0) {
+        parts.push(`(Retraction published ${retractions.join('; ')})`);
+    }
+    return collapseWhitespace(parts.join(' '));
 }
 // ---------------------------------------------------------------------------
 // MLA 9th Edition
 // ---------------------------------------------------------------------------
-/** MLA renders editors first-name-first after `edited by`. */
+/** MLA renders editors first-name-first after `edited by`, as it does a non-first author. */
 function formatEditorsMla(editors) {
-    const names = editors
-        .map((editor) => editor.collectiveName
-        ? editor.collectiveName
-        : [editor.firstName, editor.lastName].filter(Boolean).join(' '))
-        .filter(Boolean);
+    const names = editors.map((editor) => formatAuthorMla(editor, false)).filter(Boolean);
     if (names.length === 0)
         return '';
     if (names.length === 1)
@@ -565,7 +836,7 @@ function formatMlaBook(article) {
         detailParts.push(date);
     if (detailParts.length)
         parts.push(terminate(detailParts.join(', ')));
-    return parts.join(' ');
+    return collapseWhitespace(parts.join(' '));
 }
 /**
  * Format a PubMed article as an MLA 9th edition citation.
@@ -595,7 +866,8 @@ export function formatMla(article) {
     const journal = article.journalInfo;
     if (journal?.title) {
         const detailParts = [];
-        detailParts.push(`*${journal.title}*`);
+        const name = citedJournalName(journal.title, journal.isoAbbreviation);
+        detailParts.push(`*${titleCaseJournal(name, MLA_MINOR_WORDS)}*`);
         if (journal.volume) {
             detailParts.push(`vol. ${journal.volume}`);
         }
@@ -606,25 +878,24 @@ export function formatMla(article) {
         if (year !== 'n.d.') {
             detailParts.push(year);
         }
-        if (journal.pages) {
+        // MLA 9 codifies no article-number form; citation guides converge on
+        // "art. <value>" in the page position.
+        const number = articleNumber(journal);
+        if (number) {
+            detailParts.push(`art. ${number}`);
+        }
+        else if (journal.pages) {
             // MLA 9 §6.56: "p." for a single page, "pp." for a range
             const isRange = /[-\u2013\u2014]/.test(journal.pages);
             detailParts.push(`${isRange ? 'pp.' : 'p.'} ${journal.pages}`);
         }
-        else {
-            // MLA 9 codifies no article-number form; citation guides converge on
-            // "art. <value>" in the page position.
-            const locator = articleLocator(journal);
-            if (locator)
-                detailParts.push(`art. ${locator}`);
-        }
         parts.push(`${detailParts.join(', ')}.`);
     }
     // DOI
     if (article.doi) {
         parts.push(`https://doi.org/${article.doi}.`);
     }
-    return parts.join(' ');
+    return collapseWhitespace(parts.join(' '));
 }
 // ---------------------------------------------------------------------------
 // BibTeX
@@ -701,11 +972,12 @@ export function formatBibtex(article) {
     if (journal?.issue) {
         fields.push(['number', escapeBibtex(journal.issue)]);
     }
-    // Pages — or, with no pagination, biblatex's `eid` for the article number.
-    // Classic BibTeX has no article-number field and overloading `pages` is
-    // imprecise; `eid` is broadly supported (acmart included).
+    // Pages, each range written in full (#228) — or, with no pagination,
+    // biblatex's `eid` for the article number. Classic BibTeX has no
+    // article-number field and overloading `pages` is imprecise; `eid` is broadly
+    // supported (acmart included).
     if (journal?.pages) {
-        fields.push(['pages', escapeBibtex(journal.pages)]);
+        fields.push(['pages', escapeBibtex(expandPageRanges(journal.pages))]);
     }
     else if (!book) {
         const locator = articleLocator(journal);
@@ -738,29 +1010,38 @@ export function formatBibtex(article) {
         if (url)
             fields.push(['url', url]);
     }
-    // Keywords — merge article keywords with MeSH descriptor names
-    const keywordSet = new Set();
-    for (const k of article.keywords ?? [])
-        keywordSet.add(k);
-    for (const m of article.meshTerms ?? []) {
-        if (m.descriptorName)
-            keywordSet.add(m.descriptorName);
+    // Keywords — article keywords merged with MeSH descriptor names. Brace-wrap
+    // each term so MeSH descriptors that carry internal commas in their inverted
+    // form (e.g. "Databases, Protein") stay a single keyword — biblatex's
+    // comma-separated list parser reads a braced item as one element.
+    const keywords = mergedKeywords(article);
+    if (keywords.length > 0) {
+        fields.push(['keywords', keywords.map((k) => `{${escapeBibtex(k)}}`).join(', ')]);
     }
-    // Brace-wrap each term so MeSH descriptors that carry internal commas in
-    // their inverted form (e.g. "Databases, Protein") stay a single keyword —
-    // biblatex's comma-separated list parser reads a braced item as one element.
-    if (keywordSet.size > 0) {
-        const keywords = [...keywordSet].map((k) => `{${escapeBibtex(k)}}`).join(', ');
-        fields.push(['keywords', keywords]);
-    }
-    // Build entry
+    // Build entry — each value on its own line, whitespace runs collapsed (#233)
     const maxKeyLen = Math.max(...fields.map(([k]) => k.length));
-    const fieldLines = fields.map(([k, v]) => `  ${k.padEnd(maxKeyLen)} = {${v}}`).join(',\n');
+    const fieldLines = fields
+        .map(([k, v]) => `  ${k.padEnd(maxKeyLen)} = {${collapseWhitespace(v)}}`)
+        .join(',\n');
     return `@${entryType}{${key},\n${fieldLines}\n}`;
 }
 // ---------------------------------------------------------------------------
 // RIS
 // ---------------------------------------------------------------------------
+/**
+ * An `AU` or `A2` value: `Lastname, Firstname`, or `Lastname, Firstname, Suffix`.
+ * Positional, so a suffixed name with no first name keeps the empty slot
+ * (`Frenck, , Jr`) rather than reading the suffix as a first name.
+ */
+function formatNameRis(person) {
+    if (person.collectiveName)
+        return person.collectiveName;
+    const last = person.lastName ?? '';
+    const first = person.firstName ?? '';
+    if (last && person.suffix)
+        return `${last}, ${first}, ${person.suffix}`;
+    return first ? `${last}, ${first}` : last;
+}
 /**
  * Format a PubMed article as a RIS record.
  *
@@ -769,9 +1050,12 @@ export function formatBibtex(article) {
  */
 export function formatRis(article) {
     const lines = [];
+    // Every value on its own tag line: a line break inside one would leave an
+    // untagged line, or a blank one that ends the record for strict parsers (#233)
     const tag = (code, value) => {
-        if (value)
-            lines.push(`${code}  - ${value}`);
+        const text = value && collapseWhitespace(value);
+        if (text)
+            lines.push(`${code}  - ${text}`);
     };
     const book = isBookRecord(article) ? article.book : undefined;
     // Type of reference — the record type for a book, else the publication types
@@ -782,20 +1066,8 @@ export function formatRis(article) {
         : firstMappedType(article.publicationTypes, RIS_REFERENCE_TYPE, 'JOUR');
     lines.push(`TY  - ${refType}`);
     // Authors — one AU tag per author
-    if (article.authors?.length) {
-        for (const author of article.authors) {
-            if (author.collectiveName) {
-                tag('AU', author.collectiveName);
-            }
-            else {
-                const last = author.lastName ?? '';
-                const first = author.firstName ?? '';
-                if (last || first) {
-                    tag('AU', first ? `${last}, ${first}` : last);
-                }
-            }
-        }
-    }
+    for (const author of article.authors ?? [])
+        tag('AU', formatNameRis(author));
     // Title
     tag('TI', article.title);
     // Container — the journal, or the book a chapter sits in. `BT` is the book
@@ -810,14 +1082,8 @@ export function formatRis(article) {
     if (book) {
         if (article.recordType === 'book-chapter')
             tag('BT', book.title);
-        for (const editor of book.editors ?? []) {
-            const last = editor.lastName ?? '';
-            const first = editor.firstName ?? '';
-            if (editor.collectiveName)
-                tag('A2', editor.collectiveName);
-            else if (last || first)
-                tag('A2', first ? `${last}, ${first}` : last);
-        }
+        for (const editor of book.editors ?? [])
+            tag('A2', formatNameRis(editor));
         tag('PB', book.publisher);
         tag('CY', book.publisherLocation);
         tag('ET', book.edition);
@@ -834,7 +1100,7 @@ export function formatRis(article) {
     // Volume & Issue
     tag('VL', journal?.volume);
     tag('IS', journal?.issue);
-    // Pages — split into start/end, expanding PubMed's truncated-end convention.
+    // Pages — the first page and the last, written in full ({@link splitPages}).
     // With no pagination, the article number goes on `C7` (the attested RIS
     // convention for Article Number), never on SP/EP, which hold absolute pages.
     if (journal?.pages) {
@@ -866,22 +1132,12 @@ export function formatRis(article) {
         if (url)
             lines.push(`UR  - ${url}`);
     }
-    // Keywords — merge article keywords with MeSH descriptor names
-    const keywordSet = new Set();
-    for (const k of article.keywords ?? [])
-        keywordSet.add(k);
-    for (const m of article.meshTerms ?? []) {
-        if (m.descriptorName)
-            keywordSet.add(m.descriptorName);
-    }
-    for (const kw of keywordSet) {
+    // Keywords — article keywords merged with MeSH descriptor names
+    for (const kw of mergedKeywords(article)) {
         tag('KW', kw);
     }
-    // Abstract — collapse internal whitespace so blank lines don't break strict
-    // RIS parsers that terminate records at blank lines
-    if (article.abstractText) {
-        tag('AB', collapseWhitespace(article.abstractText));
-    }
+    // Abstract — a structured abstract's paragraphs flatten onto the one line
+    tag('AB', article.abstractText);
     // End of record (trailing space per RIS spec)
     lines.push('ER  - ');
     return lines.join('\n');
@@ -891,8 +1147,8 @@ export function formatRis(article) {
 // ---------------------------------------------------------------------------
 /**
  * Format a single author in Vancouver style: `Surname AB` — surname, a space,
- * then initials with no periods or internal spaces. Collective/group authors
- * return their name directly.
+ * then initials with no periods or internal spaces, then any suffix after a
+ * space (`Frenck RW Jr`). Collective/group authors return their name directly.
  */
 function formatAuthorVancouver(author) {
     if (author.collectiveName)
@@ -913,9 +1169,7 @@ function formatAuthorVancouver(author) {
         : '';
     if (!last)
         return initials;
-    if (!initials)
-        return last;
-    return `${last} ${initials}`;
+    return `${initials ? `${last} ${initials}` : last}${trailingSuffix(author)}`;
 }
 /**
  * Format the full author list for Vancouver: list every author for six or
@@ -976,8 +1230,16 @@ function formatVancouverBook(article) {
     const url = bookshelfUrl(book);
     if (url)
         segments.push(`Available from: ${url}`);
-    return segments.join(' ');
+    return collapseWhitespace(segments.join(' '));
 }
+/**
+ * The notices Vancouver notes after the reference, each with Citing Medicine's
+ * label (Boxes 58–59). NCBI's own NLM export prints these two and no others.
+ */
+const VANCOUVER_NOTE_LABELS = {
+    RetractionIn: 'Retraction in',
+    ErratumIn: 'Erratum in',
+};
 /**
  * Format a PubMed article as a Vancouver (ICMJE/NLM) reference — the numbered
  * style used by NEJM, Lancet, JAMA, BMJ, and most biomedical journals.
@@ -989,8 +1251,9 @@ function formatVancouverBook(article) {
  * ```
  * Journal name uses the NLM/ISO abbreviation when available; pages are used as
  * PubMed supplies them (often elided, e.g. "583-9"); the DOI carries no trailing
- * period so it stays copy-pasteable. An article number replaces nothing — with
+ * period so it stays copy-pasteable, unless a note follows it. An article number replaces nothing — with
  * no pagination it trails the source as an NLM note (`. pii: 2400512.`). A
+ * retraction or erratum follows as a note (`Retraction in: <RefSource>`). A
  * Bookshelf record routes to {@link formatVancouverBook}.
  */
 export function formatVancouver(article) {
@@ -1035,11 +1298,20 @@ export function formatVancouver(article) {
         const label = journal?.elocationIdType;
         segments.push(label ? `${label}: ${locator}.` : `${locator}.`);
     }
-    // DOI — NLM "doi: <doi>" form; no trailing period (would corrupt the DOI)
+    // Citing Medicine's retraction and erratum notes (Boxes 58–59), one per
+    // notice in NCBI's order with `RefSource` verbatim, as NCBI's NLM export
+    // prints them. Each ends on exactly one period: some `RefSource`s carry none.
+    const notes = citationNotices(article).flatMap((notice) => {
+        const label = VANCOUVER_NOTE_LABELS[notice.refType];
+        return label ? [terminate(`${label}: ${notice.refSource}`)] : [];
+    });
+    // DOI — NLM "doi: <doi>" form with no trailing period, which a reader copying
+    // it would take as part of the DOI; a note after it closes it, as in NLM's export
     if (article.doi) {
-        segments.push(`doi: ${article.doi}`);
+        segments.push(notes.length > 0 ? `doi: ${article.doi}.` : `doi: ${article.doi}`);
     }
-    return segments.join(' ');
+    segments.push(...notes);
+    return collapseWhitespace(segments.join(' '));
 }
 // ---------------------------------------------------------------------------
 // Dispatchers