chaffjs 0.19.0 → 0.20.0

package/src/detectors/index.ts

@@ -1,152 +1,20 @@
-import type { Detector } from "../plugin.ts";
-import { countPerSection } from "./count-per-section.ts";
-import { sentenceLength } from "./sentence-length.ts";
-import { headingEcho } from "./heading-echo.ts";
-import { repeatedHead } from "./repeated-head.ts";
-import { sentenceRhythm } from "./sentence-rhythm.ts";
-import { phraseCount, phraseMatch } from "./phrase-match.ts";
-import { agentlessPassive } from "./agentless-passive.ts";
-import { doubledWord } from "./doubled-word.ts";
-import { agreementSlip } from "./agreement-slip.ts";
-import { sentenceEnding } from "./sentence-ending.ts";
-import { doubledParticle, nounEnding } from "./token-shape.ts";
-import { kanjiRun, middleDot } from "./char-shape.ts";
-import { adverbDensity, conjunctionRun, expletive, titleCaseMix } from "./en-shape.ts";
-import { oxfordComma } from "./oxford-comma.ts";
-import { paragraphLength, paragraphVariance, preambleLength, ruleOfThree, sectionUniformity } from "./structure.ts";
-import { concreteEvidence, dashDensity, emojiDensity, ngramRepetition, undefinedAcronym } from "./signals.ts";
-import { aiTell, contractionMix, cushionDensity, hedging, repeatedConjunction, unqualifiedSuperlative } from "./lexicon.ts";
-import { internalJargon, properNounDensity, requiredSections } from "./team.ts";
-import { nameVariant } from "./name-variant.ts";
-import { latinSpacing, preferredTerm } from "./orthography.ts";
-import { straySpace } from "./stray-space.ts";
-import { announcedCount } from "./announced-count.ts";
-import { outsideAnnouncedList } from "./outside-announced-list.ts";
-import { danglingFigure } from "./dangling-figure.ts";
-import { nominalization } from "./nominalization.ts";
-import { requirementSmell } from "./requirement-smell.ts";
-import { requirementModal } from "./requirement-modal.ts";
-import { modalConflict } from "./modal-conflict.ts";
-import { vagueFigurePointer } from "./vague-figure-pointer.ts";
-import {
-  danglingReference,
-  dateOrder,
-  dateRangeReversed,
-  dateWeekdayMismatch,
-  duplicateDefinition,
-  numberingGap,
-  percentSumMismatch,
-  totalMismatch,
-} from "./structure-tree.ts";
-import { headingLevelSkip } from "./heading-level-skip.ts";
-import { imageAltText } from "./image-alt-text.ts";
-import { brokenLink } from "./broken-link.ts";
-import { urlRunOn } from "./url-run-on.ts";
-import { unbalancedBracket } from "./unbalanced-bracket.ts";
-import { doubledPunctuation } from "./doubled-punctuation.ts";
-import { kutotenConsistency } from "./kutoten-consistency.ts";
-import { hankakuKana } from "./hankaku-kana.ts";
-import { invisibleCharacter } from "./invisible-character.ts";
-import { spaceBeforePunctuation } from "./space-before-punctuation.ts";
-import { duplicateHeading } from "./duplicate-heading.ts";
-import { emptySection } from "./empty-section.ts";
-import { fullwidthAlnum } from "./fullwidth-alnum.ts";
-import { spellingVariety } from "./spelling-variety.ts";
-import { raNuki } from "./ra-nuki.ts";
-import { katakanaLongVowel } from "./long-vowel.ts";
-import { customPattern, customTokens, customWords } from "./custom.ts";
-import { boldLabelList } from "./bold-label.ts";
-import { acronymExpansionConflict, unusedDefinition, useBeforeDefinition } from "./definition-use.ts";
-import { assistantResidue, colonLeadIn, contrastFraming, openerDensity, openerPile, unfilledPlaceholder } from "./ai-phrasing.ts";
+import { extname, join } from "node:path";
+import type { CrossDetector, Detector } from "../plugin.ts";
+import { loadRegistry } from "./registry-load.ts";
+
+// Each detector is registered by its own file, registry/<how_to_find>.ts, so a new rule adds a file and edits no shared
+// list. A rule that compares the documents of one run registers in cross-registry/<how_to_find>.ts instead.
+
+const isDetector = (value: unknown): value is Detector => typeof value === "function";
+
+const isCrossDetector = (value: unknown): value is CrossDetector => typeof value === "function";
 
 /** rule 定義の how_to_find がここを引く。rule 側は実装を知らない。 */
-export const DETECTORS: Readonly<Record<string, Detector>> = {
-  "count-per-section": countPerSection,
-  "sentence-length": sentenceLength,
-  "heading-echo": headingEcho,
-  "repeated-head": repeatedHead,
-  "sentence-rhythm": sentenceRhythm,
-  "phrase-match": phraseMatch,
-  "phrase-count": phraseCount,
-  "agentless-passive": agentlessPassive,
-  "sentence-ending": sentenceEnding,
-  "noun-ending": nounEnding,
-  "doubled-particle": doubledParticle,
-  "doubled-word": doubledWord,
-  "agreement-slip": agreementSlip,
-  "kanji-run": kanjiRun,
-  "middle-dot": middleDot,
-  "adverb-density": adverbDensity,
-  expletive: expletive,
-  "conjunction-run": conjunctionRun,
-  "title-case-mix": titleCaseMix,
-  "oxford-comma": oxfordComma,
-  "paragraph-length": paragraphLength,
-  "paragraph-variance": paragraphVariance,
-  "section-uniformity": sectionUniformity,
-  "rule-of-three": ruleOfThree,
-  "preamble-length": preambleLength,
-  "emoji-density": emojiDensity,
-  "dash-density": dashDensity,
-  "ngram-repetition": ngramRepetition,
-  "undefined-acronym": undefinedAcronym,
-  "concrete-evidence": concreteEvidence,
-  hedging: hedging,
-  "cushion-density": cushionDensity,
-  "unqualified-superlative": unqualifiedSuperlative,
-  "repeated-conjunction": repeatedConjunction,
-  "ai-tell": aiTell,
-  "contrast-framing": contrastFraming,
-  "stock-transition": openerDensity,
-  "opener-pile": openerPile,
-  "colon-lead-in": colonLeadIn,
-  "assistant-residue": assistantResidue,
-  "unfilled-placeholder": unfilledPlaceholder,
-  "contraction-mix": contractionMix,
-  "internal-jargon": internalJargon,
-  "required-sections": requiredSections,
-  "proper-noun-density": properNounDensity,
-  "name-variant": nameVariant,
-  "preferred-term": preferredTerm,
-  "latin-spacing": latinSpacing,
-  "stray-space": straySpace,
-  "katakana-long-vowel": katakanaLongVowel,
-  "custom-words": customWords,
-  "custom-pattern": customPattern,
-  "custom-tokens": customTokens,
-  "dangling-reference": danglingReference,
-  "date-weekday-mismatch": dateWeekdayMismatch,
-  "date-order": dateOrder,
-  "total-mismatch": totalMismatch,
-  "numbering-gap": numberingGap,
-  "duplicate-definition": duplicateDefinition,
-  "announced-count": announcedCount,
-  "outside-announced-list": outsideAnnouncedList,
-  "dangling-figure": danglingFigure,
-  nominalization: nominalization,
-  "requirement-smell": requirementSmell,
-  "requirement-modal": requirementModal,
-  "modal-conflict": modalConflict,
-  "vague-figure-pointer": vagueFigurePointer,
-  "date-range-reversed": dateRangeReversed,
-  "percent-sum-mismatch": percentSumMismatch,
-  "heading-level-skip": headingLevelSkip,
-  "image-alt-text": imageAltText,
-  "broken-link": brokenLink,
-  "url-run-on": urlRunOn,
-  "unbalanced-bracket": unbalancedBracket,
-  "doubled-punctuation": doubledPunctuation,
-  "kutoten-consistency": kutotenConsistency,
-  "hankaku-kana": hankakuKana,
-  "invisible-character": invisibleCharacter,
-  "space-before-punctuation": spaceBeforePunctuation,
-  "duplicate-heading": duplicateHeading,
-  "empty-section": emptySection,
-  "fullwidth-alnum": fullwidthAlnum,
-  "spelling-variety": spellingVariety,
-  "ra-nuki": raNuki,
-  "bold-label-list": boldLabelList,
-  "unused-definition": unusedDefinition,
-  "use-before-definition": useBeforeDefinition,
-  "acronym-expansion-conflict": acronymExpansionConflict,
-};
+export const DETECTORS: Readonly<Record<string, Detector>> = loadRegistry(join(import.meta.dirname, "registry"), extname(import.meta.filename), isDetector);
+
+/** The detectors of the rules that compare the documents of one run (requires: [documents]), by how_to_find. */
+export const CROSS_DETECTORS: Readonly<Record<string, CrossDetector>> = loadRegistry(
+  join(import.meta.dirname, "cross-registry"),
+  extname(import.meta.filename),
+  isCrossDetector,
+);

package/src/detectors/invisible-character.ts

@@ -80,8 +80,8 @@ const CANCEL_TAG = String.fromCodePoint(0xe007f);
 
 /** タグ文字の並びをさかのぼった先が黒い旗で、並びが取り消しのタグ（U+E007F）で閉じていれば、地域の旗の中。 */
 const isInFlag = (source: string, at: number): boolean => {
-  const before = [...source.slice(Math.max(0, at - FLAG_REACH), at)];
-  const after = [...source.slice(at, at + FLAG_REACH)];
+  const before = Array.from(source.slice(Math.max(0, at - FLAG_REACH), at));
+  const after = Array.from(source.slice(at, at + FLAG_REACH));
   const closing = after.findIndex((char) => !TAG.test(char) || char === CANCEL_TAG);
   return before[before.findLastIndex((char) => !TAG.test(char))] === BLACK_FLAG && after[closing] === CANCEL_TAG;
 };
@@ -185,6 +185,6 @@ export const invisibleCharacter: Detector = (doc): Finding[] =>
     line: 0,
     column: 0,
     quote: visibleQuote(doc.source, run.span),
-    values: { code: run.codes.join(" "), count: [...doc.source.slice(run.span.start, run.span.end)].length, offset: run.span.start },
+    values: { code: run.codes.join(" "), count: Array.from(doc.source.slice(run.span.start, run.span.end)).length, offset: run.span.start },
     variant: run.kind,
   }));

package/src/detectors/katakana-density.ts

@@ -0,0 +1,62 @@
+import type { Detector, Finding, ProseDocument, Sentence } from "../plugin.ts";
+import { proseText } from "../measure.ts";
+import { rateOver } from "./lexicon.ts";
+
+/** A katakana word, full or half width: two or more katakana, not starting with the long-vowel mark. A middle dot (ジョン・スミス) splits it. */
+const KATAKANA_WORD = /[ァ-ヴ][ァ-ヴー]+|[ｦ-ﾝ][ｦ-ﾟ]+/gu;
+
+/** A line break inside a katakana word (アジェ↵ンダ): a wrapped line, not two words. */
+const WRAP_IN_WORD = /(?<=[ァ-ヴーｦ-ﾟ])[^\S\n]*\n[^\S\n]*(?=[ァ-ヴーｦ-ﾟ])/gu;
+
+/** How many of the most frequent words the message names. */
+const NAMED_WORDS = 3;
+
+type KatakanaWord = { readonly sentence: Sentence; readonly word: string };
+
+/** The text with each of the team's names (chaff.yaml names:) blanked out, longest first: a name is not a loanword to trim. */
+const withoutNames = (text: string, names: readonly string[]): string =>
+  names
+    .filter((name) => name !== "")
+    .toSorted((left, right) => right.length - left.length)
+    .reduce((rest, name) => rest.replaceAll(name, " "), text);
+
+/** The katakana words of one sentence's text, a line wrapped inside a word joined back, the team's names left out. */
+export const katakanaWordsIn = (text: string, names: readonly string[]): string[] =>
+  [...withoutNames(text.replace(WRAP_IN_WORD, ""), names).matchAll(KATAKANA_WORD)].map((match) => match[0]);
+
+/** The words used most, most frequent first; ties keep the order they first appear in. */
+export const mostFrequent = (words: readonly string[], count: number): string[] => {
+  const tally = words.reduce((counts, word) => counts.set(word, (counts.get(word) ?? 0) + 1), new Map<string, number>());
+  return [...tally.entries()]
+    .toSorted((left, right) => right[1] - left[1])
+    .slice(0, count)
+    .map(([word]) => word);
+};
+
+const wordsOfDocument = (doc: ProseDocument): KatakanaWord[] =>
+  doc.sentences.flatMap((sentence) => katakanaWordsIn(sentence.text, doc.names ?? []).map((word) => ({ sentence, word })));
+
+/**
+ * Katakana loanwords per 1000 characters, over the whole document. Reported once, at the first sentence holding one, with the
+ * words used most: those are the ones a Japanese word could replace first. Short documents are not measured.
+ */
+export const katakanaDensity: Detector = (doc, options): Finding[] => {
+  const found = wordsOfDocument(doc);
+  const density = rateOver(doc, found.length, options.limit);
+  const first = found[0];
+  if (density === undefined || first === undefined) return [];
+  const words = mostFrequent(
+    found.map((hit) => hit.word),
+    NAMED_WORDS,
+  );
+  return [
+    {
+      rule: "katakana-density",
+      severity: "info",
+      line: 0,
+      column: 0,
+      quote: proseText(first.sentence),
+      values: { count: found.length, density, limit: options.limit, words: words.join("、"), offset: first.sentence.span.start },
+    },
+  ];
+};

package/src/detectors/known-correction.ts

@@ -0,0 +1,89 @@
+import type { Detector, Finding, LexiconEntry, ProseDocument, Sentence } from "../plugin.ts";
+import { entryRanges } from "./lexicon-match.ts";
+import { escapeRegExp } from "../orthography.ts";
+import { correctionIn, editBetween, type Correction, type Edit } from "../correction-edit.ts";
+
+/** 語彙表の一組。pattern が誤った形、instead_of が正しい形。 */
+type Pair = { readonly entry: LexiconEntry; readonly wrong: string; readonly edit: Edit };
+
+/** 見つけた誤り 1 つ。offset は文書の上の位置。 */
+type Hit = { readonly sentence: Sentence; readonly offset: number; readonly correction: Correction };
+
+const pairsOf = (entries: readonly LexiconEntry[]): Pair[] =>
+  entries.flatMap((entry) =>
+    entry.instead_of === undefined || entry.pattern === "" ? [] : [{ entry, wrong: entry.pattern, edit: editBetween(entry.pattern, entry.instead_of) }],
+  );
+
+const LATIN_EDGE = /[A-Za-z0-9]/u;
+
+/** 英字の語は語の途中に当てない（"teh" は "Tehran" の中に無い）。かなと漢字は字で語の切れ目を決められないので、品詞が無ければ字の並びで照らす。 */
+const wordPattern = (wrong: string): RegExp => {
+  const before = LATIN_EDGE.test(wrong.charAt(0)) ? "(?<![\\p{L}\\p{N}])" : "";
+  const after = LATIN_EDGE.test(wrong.charAt(wrong.length - 1)) ? "(?![\\p{L}\\p{N}])" : "";
+  return new RegExp(`${before}${escapeRegExp(wrong)}${after}`, "giu");
+};
+
+const hitAt = (sentence: Sentence, start: number, correction: Correction | undefined): Hit[] =>
+  correction === undefined ? [] : [{ sentence, offset: start, correction }];
+
+/** 品詞が無い文。字の並びで、英字の語の切れ目を守って照らす。指すのは当たった字全体。 */
+const textHits = (sentence: Sentence, pair: Pair): Hit[] =>
+  [...sentence.text.matchAll(wordPattern(pair.wrong))].flatMap((match) =>
+    hitAt(sentence, sentence.span.start + match.index, correctionIn(match[0], pair.wrong, pair.edit, [match[0].length])),
+  );
+
+/** 品詞のある文。語の並びで照らすので、語の途中には当たらない（「ですすべて」の「ですす」）。指すのは違う所を含む語まで。 */
+const tokenHits = (sentence: Sentence, pair: Pair, source: string): Hit[] => {
+  const tokens = sentence.tokens ?? [];
+  return entryRanges(sentence, pair.entry).flatMap((range) => {
+    const words = tokens.slice(range.start, range.end);
+    const start = words[0]?.span.start ?? 0;
+    const written = source.slice(start, words.at(-1)?.span.end ?? start);
+    const boundaries = words.map((word) => word.span.end - start);
+    return written === "" ? [] : hitAt(sentence, start, correctionIn(written, pair.wrong, pair.edit, boundaries));
+  });
+};
+
+/** 字の並びで当たった所のうち、両端が文の語の切れ目にあるもの。語彙表の語を解析器が文の中と違う切り方にした（「ますす」）ときの拾い直し。 */
+const alignedTextHits = (sentence: Sentence, pair: Pair): Hit[] => {
+  const tokens = sentence.tokens ?? [];
+  const starts = new Set(tokens.map((token) => token.span.start));
+  const ends = new Set(tokens.map((token) => token.span.end));
+  return textHits(sentence, pair).filter((hit) => starts.has(hit.offset) && ends.has(hit.offset + pair.wrong.length));
+};
+
+const hitsIn = (sentence: Sentence, pair: Pair, source: string): Hit[] => {
+  if (sentence.tokens === undefined || (pair.entry.tokens?.length ?? 0) === 0) return textHits(sentence, pair);
+  return [...tokenHits(sentence, pair, source), ...alignedTextHits(sentence, pair)];
+};
+
+/** チームが names に並べた名前の中の字（「黒ひげ危機一発」の「危機一発」）は、その字のまま正しい。 */
+const isInTeamName = (doc: ProseDocument, hit: Hit): boolean =>
+  (doc.names ?? []).some((name) =>
+    [...hit.sentence.text.matchAll(new RegExp(escapeRegExp(name), "giu"))].some((match) => {
+      const start = hit.sentence.span.start + match.index;
+      return start <= hit.offset && hit.offset + hit.correction.matched.length <= start + match[0].length;
+    }),
+  );
+
+/** 一つの所に二つの組が当たったら、語彙表で先に書いた組だけ。 */
+const firstAtEachOffset = (hits: readonly Hit[]): Hit[] => hits.filter((hit, at) => hits.findIndex((other) => other.offset === hit.offset) === at);
+
+/**
+ * 誤りと正しい形の組の語彙表（pattern が誤り、instead_of が正しい形）。文書の中で誤った形を見つけ、正しい形を添えて言う。
+ * どの組を誤りとするかは語彙表が決め、detector は語の切れ目と大文字と、組のどこが違うかだけを扱う。
+ */
+export const knownCorrection: Detector = (doc, options): Finding[] => {
+  const pairs = pairsOf(options.lexicon ?? []);
+  const hits = doc.sentences.flatMap((sentence) => pairs.flatMap((pair) => hitsIn(sentence, pair, doc.source)));
+  return firstAtEachOffset(hits.toSorted((left, right) => left.offset - right.offset))
+    .filter((hit) => !isInTeamName(doc, hit))
+    .map((hit) => ({
+      rule: "",
+      severity: "warning",
+      line: 0,
+      column: 0,
+      quote: hit.sentence.text.trim(),
+      values: { matched: hit.correction.matched, suggestion: hit.correction.suggestion, offset: hit.offset },
+    }));
+};

package/src/detectors/lexicon-match.ts

@@ -81,7 +81,7 @@ export const entryOpens = (sentence: Sentence, entry: LexiconEntry): boolean =>
 };
 
 /** Marks and closing brackets that may follow the last word of a sentence (こと。」, must.). */
-const TRAILING_MARKS = new Set([..." \t\n。．.!！?？」』)）\"'”’"]);
+const TRAILING_MARKS = new Set(" \t\n。．.!！?？」』)）\"'”’");
 
 const withoutTrailingMarks = (text: string): string => {
   const chars = [...text];

package/src/detectors/lexicon.ts

@@ -66,6 +66,9 @@ const densityRule =
 
 const hedgingDensity = densityRule("excessive-hedging");
 export const cushionDensity = densityRule("cushion-phrase-density", "at-floor");
+/** 語彙表の語の密度だけを見る rule が共通で使う。rule の id は実行側が入れる。 */
+export const phraseDensity = densityRule("");
+export const vagueDensity = densityRule("vague-word-density");
 
 /** 書いたとおりの字。行の折り返しをまたいでいたら空白 1 つに。 */
 const writtenAt = (source: string, span: Span): string => source.slice(span.start, span.end).replace(/\s+/gu, " ").trim();

package/src/detectors/list-parallel.ts

@@ -0,0 +1,155 @@
+// Whether the items of one bulleted list share a form: all phrases or all sentences, all opening with a verb or all with
+// a noun. Pure; reads the item sentences' tokens. Which form is right is the writer's choice, so only the minority within
+// one list is reported. です/ます against plain endings inside a list is no-mixed-desumasu's, not this.
+import { minorityWithin } from "../orthography.ts";
+import { isQuotedAt } from "./bold-label.ts";
+import type { BulletList, Detector, Finding, ProseDocument, Sentence, Span, Token } from "../plugin.ts";
+
+/** What an item is made of: a phrase or a sentence (ending), and the word it opens with (head, English only). */
+export type Axis = "ending" | "head";
+
+/** For each axis, the two forms. The first is the side minorityWithin calls true. */
+export const FORMS: Readonly<Record<Axis, readonly [string, string]>> = { ending: ["phrase", "sentence"], head: ["verb", "noun"] };
+
+const SKIPPED_AT_END = new Set(["PUNCT", "SYM", "X"]);
+const NOMINAL = new Set(["NOUN", "PROPN", "NUM", "PRON"]);
+const PREDICATE = new Set(["VERB", "AUX", "ADJ"]);
+const NOUN_HEAD = new Set(["NOUN", "ADJ", "DET", "NUM", "PRON", "PROPN"]);
+const SENTENCE_END = /[.!?。．！？]["'”’)）」』]*$/u;
+const OPENING_BRACKETS = new Set(["(", "（", "[", "［", "【"]);
+const CLOSING_BRACKETS = new Set([")", "）", "]", "］", "】"]);
+
+/** A word the tagger only guessed at, or one that is also a verb (Design, Document): its form is not known. */
+const isUnsure = (token: Token | undefined): boolean => token?.features?.["Guess"] === "Yes" || token?.features?.["AlsoVerb"] === "Yes";
+
+/** A finite verb or auxiliary: an English item with one says something. A gerund or participle (Running, nested) does not. */
+const isFiniteVerb = (token: Token): boolean => (token.pos === "VERB" || token.pos === "AUX") && !["Ger", "Part"].includes(token.features?.["VerbForm"] ?? "");
+
+/** The tokens before a closing aside: 「健康診断書（3か月以内のもの）」 ends, for its form, at 健康診断書. */
+const withoutClosingAside = (tokens: readonly Token[]): readonly Token[] => {
+  const last = tokens.findLastIndex((token) => !SKIPPED_AT_END.has(token.pos) || CLOSING_BRACKETS.has(token.surface));
+  if (last === -1 || !CLOSING_BRACKETS.has(tokens[last]?.surface ?? "")) return tokens;
+  const open = tokens.findLastIndex((token, index) => index < last && OPENING_BRACKETS.has(token.surface));
+  return open <= 0 ? tokens : tokens.slice(0, open);
+};
+
+/** Japanese: a predicate at the end (出します, 済ませる, 早い) is a sentence; a noun (月末, 確認すること) is a phrase. */
+const japaneseEnding = (sentences: readonly Sentence[]): string | undefined => {
+  if (sentences.length > 1) return "sentence";
+  const last = withoutClosingAside(sentences[0]?.tokens ?? []).findLast((token) => !SKIPPED_AT_END.has(token.pos));
+  if (last === undefined) return undefined;
+  if (NOMINAL.has(last.pos)) return "phrase";
+  return PREDICATE.has(last.pos) ? "sentence" : undefined;
+};
+
+/** English: a sentence closes with a full stop and has a finite verb; a phrase has neither, nor any verb at all. Anything between is not judged. */
+const englishEnding = (sentences: readonly Sentence[]): string | undefined => {
+  const tokens = sentences.flatMap((sentence) => sentence.tokens ?? []);
+  if (tokens.length === 0) return undefined;
+  const closed = SENTENCE_END.test(sentences.at(-1)?.text.trim() ?? "");
+  if (sentences.length > 1 || (closed && tokens.some(isFiniteVerb))) return "sentence";
+  const verbless = !tokens.some((token) => token.pos === "VERB" || token.pos === "AUX");
+  return !closed && verbless && !isUnsure(tokens[0]) ? "phrase" : undefined;
+};
+
+/** English: the word an item opens with. Install and Running open with a verb; The, Fast and Docker with a noun phrase. */
+/** Small words a title leaves in lower case (Terms of Service). */
+const TITLE_SMALL_WORDS = new Set(["of", "and", "or", "the", "a", "an", "to", "for", "in", "on"]);
+
+/** Every word capitalised (Cover Letter, Terms of Service): a name or a title, whose first word is not a verb whatever its tag. */
+const isTitleCase = (text: string): boolean =>
+  text
+    .trim()
+    .split(/\s+/u)
+    .every((word) => TITLE_SMALL_WORDS.has(word) || /^\p{Lu}/u.test(word));
+
+const englishHead = (sentences: readonly Sentence[]): string | undefined => {
+  const first = (sentences[0]?.tokens ?? []).find((token) => !SKIPPED_AT_END.has(token.pos));
+  if (first === undefined || isUnsure(first) || isTitleCase(sentences[0]?.text ?? "")) return undefined;
+  // A participle (Generated files, Nested lists) modifies the noun after it; the item is a noun phrase, but the tag does not say so.
+  if (first.pos === "VERB") return first.features?.["VerbForm"] === "Part" ? undefined : "verb";
+  return NOUN_HEAD.has(first.pos) ? "noun" : undefined;
+};
+
+/** An item's form on an axis, or undefined when its words do not tell. Japanese items have no head axis (the verb comes last). */
+export const formOf = (axis: Axis, sentences: readonly Sentence[], language: string): string | undefined => {
+  if (language === "ja") return axis === "ending" ? japaneseEnding(sentences) : undefined;
+  return axis === "ending" ? englishEnding(sentences) : englishHead(sentences);
+};
+
+const within = (inner: Span, outer: Span): boolean => inner.start >= outer.start && inner.end <= outer.end;
+
+/** The item's own sentences: those in its span, less those of a list nested inside it. */
+export const itemSentences = (doc: ProseDocument, item: Span): Sentence[] => {
+  const nested = doc.lists.map((list) => list.span).filter((span) => span.start > item.start && within(span, item));
+  return doc.sentences.filter((sentence) => within(sentence.span, item) && !nested.some((span) => within(sentence.span, span)));
+};
+
+type Item = { readonly sentences: readonly Sentence[]; readonly form: string };
+
+const findingOf = (axis: Axis, item: Item, count: number, of: number, limit: number): Finding => {
+  const first = item.sentences[0];
+  return {
+    rule: "list-item-form-mix",
+    severity: "info",
+    line: 0,
+    column: 0,
+    quote: first?.text.trim() ?? "",
+    values: { written: first?.text.trim() ?? "", count, of, limit, offset: first?.span.start ?? 0 },
+    variant: `${axis}-${item.form}`,
+  };
+};
+
+const AXES: readonly Axis[] = ["ending", "head"];
+
+/** An item that opens with a label and a colon (大倉さんの肌感：…, Last modified: …): a field and its value, not judged. */
+const LABELLED = /^[^:：。.]{1,40}\s*[:：]\s*\S/u;
+
+/** Code or a link in the item as written: chaff masks it out of the sentence, so the item's words are not all there. */
+const CODE_OR_LINK = /`|\]\(|\]\[|https?:\/\//u;
+
+/** An item ending like a clause of one running sentence (…conduct;, …、, …and): the list is one sentence split up. */
+const RUNNING_ON = /(?:[;,、，]|\b(?:and|or))$/iu;
+
+/** An item of one word (Resume, 例) is a name: either form would read it the same. */
+const MIN_WORDS = 2;
+
+const wordCount = (sentences: readonly Sentence[]): number =>
+  sentences.flatMap((sentence) => sentence.tokens ?? []).filter((token) => !SKIPPED_AT_END.has(token.pos)).length;
+
+const isJudged = (written: string, sentences: readonly Sentence[]): boolean => {
+  const text = sentences.map((sentence) => sentence.text).join(" ");
+  return wordCount(sentences) >= MIN_WORDS && !LABELLED.test(text.trim()) && !CODE_OR_LINK.test(written);
+};
+
+const runsOn = (sentences: readonly Sentence[]): boolean => RUNNING_ON.test(sentences.at(-1)?.text.trim() ?? "");
+
+const listFindings = (doc: ProseDocument, list: BulletList, limit: number): Finding[] => {
+  const sentencesOf = list.itemSpans.map((span) => itemSentences(doc, span));
+  if (sentencesOf.some(runsOn)) return [];
+  const judged = sentencesOf.filter((sentences, index) => {
+    const span = list.itemSpans[index];
+    return span !== undefined && isJudged(doc.source.slice(span.start, sentences.at(-1)?.span.end ?? span.end), sentences);
+  });
+  return AXES.flatMap((axis) => {
+    const items = judged.flatMap((sentences) => {
+      const form = formOf(axis, sentences, doc.language);
+      return form === undefined ? [] : [{ sentences, form }];
+    });
+    const odd = minorityWithin(items, (item) => item.form === FORMS[axis][0], limit);
+    return odd.map((item) => findingOf(axis, item, odd.length, items.length, limit));
+  });
+};
+
+/** Each list item whose form is the minority in its list, on each axis. A quoted list is someone else's words. */
+export const listItemFormMix: Detector = (doc, options): Finding[] => {
+  const findings = doc.lists.filter((list) => !isQuotedAt(doc.source, list.span.start)).flatMap((list) => listFindings(doc, list, options.limit));
+  // The same item can be read in two lists (its own and one the parser nests it in); report it once.
+  const seen = new Set<string>();
+  return findings.filter((finding) => {
+    const key = `${String(finding.values["offset"])} ${finding.variant ?? ""}`;
+    if (seen.has(key)) return false;
+    seen.add(key);
+    return true;
+  });
+};

package/src/detectors/long-vowel.ts

@@ -10,17 +10,17 @@ const wordsOf = (value: OptionValue | undefined): readonly string[] =>
   Array.isArray(value) ? value.filter((entry): entry is string => typeof entry === "string") : [];
 
 /** The minimum when no setting gives one: two morae before the final ー, which leaves カー alone. */
-const DEFAULT_MIN_MORAE = 2;
+export const DEFAULT_MIN_MORAE = 2;
 
 /** A listed word is exempt written either way: listing コンピューター exempts コンピュータ too. */
-const exemptStems = (except: readonly string[]): ReadonlySet<string> => new Set(except.map((word) => stemOf(word.trim())));
+export const exemptStems = (except: readonly string[]): ReadonlySet<string> => new Set(except.map((word) => stemOf(word.trim())));
 
 /**
  * Words that become another word with a final ー (フリ, as in 振り, and フリー, free): the language's long-vowel-distinct word list.
  * Both are nouns to the tagger, so only the list tells them apart; they are left out as if listed under except.
  * The list is optional: without it the rule still runs and compares every word.
  */
-const distinctWordsOf = (doc: ProseDocument): string[] => (doc.lexicons["long-vowel-distinct"] ?? []).map((entry) => entry.pattern);
+export const distinctWordsOf = (doc: ProseDocument): string[] => (doc.lexicons["long-vowel-distinct"] ?? []).map((entry) => entry.pattern);
 
 /**
  * The kana a final ー follows where drop drops it: the language's long-vowel-drop-after list (ア段, the -er, -or, -ar of
@@ -35,7 +35,7 @@ const dropAfterOf = (doc: ProseDocument): ReadonlySet<string> | undefined => {
  * The katakana nouns of a sentence, as the language adapter split them: ユーザー inside ユーザーインターフェース is its own word.
  * Proper nouns (PROPN) and the team's names keep their own spelling (ディズニー).
  */
-const kanaWordsOf = (sentence: Sentence, names: readonly string[], exempt: ReadonlySet<string>): KanaWord[] => {
+export const kanaWordsOf = (sentence: Sentence, names: readonly string[], exempt: ReadonlySet<string>): KanaWord[] => {
   const named = nameSpans(sentence.text, names);
   return (sentence.tokens ?? [])
     .filter((token) => token.pos === "NOUN" && isKatakanaWord(token.surface) && !exempt.has(stemOf(token.surface)))

package/src/detectors/misuse-match.ts

@@ -0,0 +1,28 @@
+import type { Detector, Finding, Lexicon, LexiconEntry, Sentence } from "../plugin.ts";
+import { entryIn } from "./lexicon-match.ts";
+
+type Misuse = { readonly sentence: Sentence; readonly entry: LexiconEntry };
+
+const misuseIn = (sentence: Sentence, lexicon: Lexicon): Misuse | undefined => {
+  const entry = lexicon.find((candidate) => entryIn(sentence, candidate));
+  return entry === undefined ? undefined : { sentence, entry };
+};
+
+const findingOf = (misuse: Misuse, limit: number): Finding => ({
+  rule: "",
+  severity: "warning",
+  line: 0,
+  column: 0,
+  quote: misuse.sentence.text.trim(),
+  values: { matched: misuse.entry.pattern, correct: misuse.entry.rewrite ?? "", limit, offset: misuse.sentence.span.start },
+});
+
+/**
+ * A phrase from a lexicon of established misuses, each entry carrying its correct form in `rewrite`. The message names
+ * both, so the writer does not have to look the right form up. One finding per sentence, for its first listed misuse.
+ */
+export const misuseMatch: Detector = (doc, options): Finding[] => {
+  const lexicon = options.lexicon ?? [];
+  const misuses = doc.sentences.flatMap((sentence) => misuseIn(sentence, lexicon) ?? []);
+  return misuses.length < options.limit ? [] : misuses.map((misuse) => findingOf(misuse, options.limit));
+};

package/src/detectors/modal-conflict.ts

@@ -4,7 +4,7 @@ import { modalConflicts, modalStatements, type Obligation } from "../structure/m
 import { linesOf, lineNumberAt } from "../structure/lines.ts";
 import { quoteAt } from "./structure-tree.ts";
 
-const obligationsOf = (tree: StructureNode): Obligation[] =>
+export const obligationsOf = (tree: StructureNode): Obligation[] =>
   inDocumentOrder(tree).flatMap((node) =>
     node.kind === "obligation"
       ? [{ start: node.span.start, end: node.span.end, type: String(node.attrs["type"] ?? ""), marker: String(node.attrs["marker"] ?? "") }]