flappa-doormal 2.17.1 → 2.19.0

package/dist/index.mjs

@@ -1,141 +1,25 @@
-//#region src/utils/textUtils.ts
-/**
-* Normalizes line endings to Unix-style (`\n`).
-*
-* Converts Windows (`\r\n`) and old Mac (`\r`) line endings to Unix style
-* for consistent pattern matching across platforms.
-*
-* @param content - Raw content with potentially mixed line endings
-* @returns Content with all line endings normalized to `\n`
-*/
-const normalizeLineEndings = (content) => {
-	return content.includes("\r") ? content.replace(/\r\n?/g, "\n") : content;
-};
-/**
-* Escapes regex metacharacters (parentheses and brackets) in template patterns,
-* but preserves content inside `{{...}}` token delimiters.
-*
-* This allows users to write intuitive patterns like `({{harf}}):` instead of
-* the verbose `\\({{harf}}\\):`. The escaping is applied BEFORE token expansion,
-* so tokens like `{{harf}}` which expand to `[أ-ي]` work correctly.
-*
-* @param pattern - Template pattern that may contain `()[]` and `{{tokens}}`
-* @returns Pattern with `()[]` escaped outside of `{{...}}` delimiters
-*
-* @example
-* escapeTemplateBrackets('({{harf}}): ')
-* // → '\\({{harf}}\\): '
-*
-* @example
-* escapeTemplateBrackets('[{{raqm}}] ')
-* // → '\\[{{raqm}}\\] '
-*
-* @example
-* escapeTemplateBrackets('{{harf}}')
-* // → '{{harf}}' (unchanged - no brackets outside tokens)
-*/
-const escapeTemplateBrackets = (pattern) => {
-	return pattern.replace(/(\{\{[^}]*\}\})|([()[\]])/g, (_match, token, bracket) => token || `\\${bracket}`);
-};
+//#region src/segmentation/tokens.ts
 /**
-* Character class matching all Arabic diacritics (Tashkeel/Harakat).
+* Arabic base letters used by low-level dictionary-style regex helpers.
 *
-* Includes the following diacritical marks:
-* - U+064B: ً (fathatan - double fatha)
-* - U+064C: ٌ (dammatan - double damma)
-* - U+064D: ٍ (kasratan - double kasra)
-* - U+064E: َ (fatha - short a)
-* - U+064F: ُ (damma - short u)
-* - U+0650: ِ (kasra - short i)
-* - U+0651: ّ (shadda - gemination)
-* - U+0652: ْ (sukun - no vowel)
-*
-* @internal
+* This is intentionally broader than `{{harf}}`:
+* - includes standalone hamza `ء`
+* - stays as a raw regex fragment rather than a template token
 */
-const DIACRITICS_CLASS = "[ًٌٍَُِّْ]";
+const ARABIC_BASE_LETTER_CLASS = "[ء-غف-ي]";
 /**
-* Groups of equivalent Arabic characters.
-*
-* Characters within the same group are considered equivalent for matching purposes.
-* This handles common variations in Arabic text where different characters are
-* used interchangeably or have the same underlying meaning.
-*
-* Equivalence groups:
-* - Alef variants: ا (bare), آ (with madda), أ (with hamza above), إ (with hamza below)
-* - Ta marbuta and Ha: ة ↔ ه (often interchangeable at word endings)
-* - Alef maqsura and Ya: ى ↔ ي (often interchangeable at word endings)
-*
-* @internal
+* Arabic combining marks / annotation signs used by low-level regex helpers.
 */
-const EQUIV_GROUPS = [
-	[
-		"ا",
-		"آ",
-		"أ",
-		"إ"
-	],
-	["ة", "ه"],
-	["ى", "ي"]
-];
+const ARABIC_MARKS_CLASS = "[\\u0610-\\u061A\\u0640\\u064B-\\u065F\\u0670\\u06D6-\\u06ED]";
 /**
-* Escapes a string for safe inclusion in a regular expression.
-*
-* Escapes all regex metacharacters: `.*+?^${}()|[\]\\`
-*
-* @param s - Any string to escape
-* @returns String with regex metacharacters escaped
-*
-* @example
-* escapeRegex('hello.world')   // → 'hello\\.world'
-* escapeRegex('[test]')        // → '\\[test\\]'
-* escapeRegex('a+b*c?')        // → 'a\\+b\\*c\\?'
+* A single Arabic base letter followed by zero or more combining marks.
 */
-const escapeRegex = (s) => s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
-const getEquivClass = (ch) => {
-	const group = EQUIV_GROUPS.find((g) => g.includes(ch));
-	return group ? `[${group.map(escapeRegex).join("")}]` : escapeRegex(ch);
-};
-const normalizeArabicLight = (str) => {
-	return str.normalize("NFC").replace(/[\u200C\u200D]/g, "").replace(/\s+/g, " ").trim();
-};
-const makeDiacriticInsensitive = (text) => {
-	const diacriticsMatcher = `${DIACRITICS_CLASS}*`;
-	return Array.from(normalizeArabicLight(text)).map((ch) => getEquivClass(ch) + diacriticsMatcher).join("");
-};
-const isCombiningMarkOrSelector = (char) => {
-	if (!char) return false;
-	return /\p{M}/u.test(char) || char === "︎" || char === "️";
-};
-const isJoiner = (char) => char === "‌" || char === "‍";
+const ARABIC_LETTER_WITH_OPTIONAL_MARKS_PATTERN = `${ARABIC_BASE_LETTER_CLASS}${ARABIC_MARKS_CLASS}*`;
 /**
-* Ensures the position does not split a grapheme cluster (surrogate pairs,
-* combining marks, or zero-width joiners / variation selectors).
-*
-* This is only used as a last-resort fallback when we are forced to split
-* near a hard limit (e.g. maxContentLength with no safe whitespace/punctuation).
+* One or more Arabic letters, where each letter may carry combining marks.
 */
-const adjustForUnicodeBoundary = (content, position) => {
-	let adjusted = position;
-	while (adjusted > 0) {
-		const high = content.charCodeAt(adjusted - 1);
-		const low = content.charCodeAt(adjusted);
-		if (high >= 55296 && high <= 56319 && low >= 56320 && low <= 57343) {
-			adjusted -= 1;
-			continue;
-		}
-		const nextChar = content[adjusted];
-		const prevChar = content[adjusted - 1];
-		if (isCombiningMarkOrSelector(nextChar) || isJoiner(nextChar) || isJoiner(prevChar)) {
-			adjusted -= 1;
-			continue;
-		}
-		break;
-	}
-	return adjusted;
-};
-
-//#endregion
-//#region src/segmentation/tokens.ts
+const ARABIC_WORD_WITH_OPTIONAL_MARKS_PATTERN = `(?:${ARABIC_LETTER_WITH_OPTIONAL_MARKS_PATTERN})+`;
+const ARABIC_SPACED_CODE_ATOM = `[أ-غف-ي]${ARABIC_MARKS_CLASS}*`;
 const RUMUZ_ATOM = `(?:${[
 	"تمييز(?![\\u064B-\\u0652\\u0670أ-ي])",
 	"خت",
@@ -166,15 +50,25 @@ const RUMUZ_ATOM = `(?:${[
 ].join("|")})`;
 const RUMUZ_BLOCK = `${RUMUZ_ATOM}(?:\\s+${RUMUZ_ATOM})*`;
 const BASE_TOKENS = {
+	/** Chapter marker (باب). */
 	bab: "باب",
+	/** Basmala (بسم الله). Also matches ﷽. */
 	basmalah: ["بسم الله", "﷽"].join("|"),
+	/** Bullet point variants: `•`, `*`, `°`. */
 	bullet: "[•*°]",
+	/** Dash variants: `-` (U+002D), `–` (U+2013), `—` (U+2014), `ـ` (tatweel U+0640). */
 	dash: "[-–—ـ]",
+	/** Section marker (فصل / مسألة). */
 	fasl: ["مسألة", "فصل"].join("|"),
+	/** Single Arabic letter (أ-ي). Does NOT include diacritics. */
 	harf: "[أ-ي]",
-	harfs: "[أ-ي](?:\\s+[أ-ي])*",
+	/** One or more single Arabic letters separated by spaces, allowing marks/tatweel on each isolated letter (e.g. `د ت س`, `هـ ث`). For multi-letter codes use `{{rumuz}}`. */
+	harfs: `${ARABIC_SPACED_CODE_ATOM}(?:\\s+${ARABIC_SPACED_CODE_ATOM})*`,
+	/** Horizontal rule / separator: 5+ repeated dashes, underscores, equals, or tatweels. Mixed allowed. */
 	hr: "[-–—ـ_=]{5,}",
+	/** Book marker (كتاب). */
 	kitab: "كتاب",
+	/** Hadith transmission phrases (حدثنا, أخبرنا, حدثني, etc.). */
 	naql: [
 		"حدثني",
 		"وأخبرنا",
@@ -186,33 +80,58 @@ const BASE_TOKENS = {
 		"وحدثني",
 		"وحدثنيه"
 	].join("|"),
+	/** Newline character. Useful for breakpoints that split on line boundaries. */
 	newline: "\\n",
+	/** Single ASCII digit (0-9). */
 	num: "\\d",
+	/** One or more ASCII digits (0-9)+. */
 	nums: "\\d+",
+	/** Single Arabic-Indic digit (٠-٩, U+0660-U+0669). */
 	raqm: "[\\u0660-\\u0669]",
+	/** One or more Arabic-Indic digits (٠-٩)+. */
 	raqms: "[\\u0660-\\u0669]+",
+	/** Rijāl/takhrīj source abbreviations. Matches one or more codes separated by whitespace. */
 	rumuz: RUMUZ_BLOCK,
+	/** Arabic/common punctuation: `.`, `!`, `?`, `؟`, `؛`. */
 	tarqim: "[.!?؟؛]"
 };
 /** Pre-defined token constants for use in patterns. */
 const Token = {
+	/** Chapter marker - باب */
 	BAB: "{{bab}}",
+	/** Basmala - بسم الله */
 	BASMALAH: "{{basmalah}}",
+	/** Bullet point variants */
 	BULLET: "{{bullet}}",
+	/** Dash variants (hyphen, en-dash, em-dash, tatweel) */
 	DASH: "{{dash}}",
+	/** Section marker - فصل / مسألة */
 	FASL: "{{fasl}}",
+	/** Single Arabic letter */
 	HARF: "{{harf}}",
+	/** Multiple Arabic letters separated by spaces, allowing marks/tatweel on each isolated letter */
 	HARFS: "{{harfs}}",
+	/** Horizontal rule / separator (repeated dashes) */
 	HR: "{{hr}}",
+	/** Book marker - كتاب */
 	KITAB: "{{kitab}}",
+	/** Hadith transmission phrases */
 	NAQL: "{{naql}}",
+	/** Newline character (for breakpoints) */
 	NEWLINE: "{{newline}}",
+	/** Single ASCII digit */
 	NUM: "{{num}}",
+	/** Composite: {{raqms}} {{dash}} (space) */
 	NUMBERED: "{{numbered}}",
+	/** One or more ASCII digits */
 	NUMS: "{{nums}}",
+	/** Single Arabic-Indic digit */
 	RAQM: "{{raqm}}",
+	/** One or more Arabic-Indic digits */
 	RAQMS: "{{raqms}}",
+	/** Source abbreviations (rijāl/takhrīj) */
 	RUMUZ: "{{rumuz}}",
+	/** Punctuation marks */
 	TARQIM: "{{tarqim}}"
 };
 /** Wraps a token constant with a named capture: `{{token}}` → `{{token:name}}`. */
@@ -222,7 +141,9 @@ const withCapture = (token, name) => {
 	return `{{${match[1]}:${name}}}`;
 };
 /** Composite tokens that reference base tokens. Pre-expanded at load time. @internal */
-const COMPOSITE_TOKENS = { numbered: "{{raqms}} {{dash}} " };
+const COMPOSITE_TOKENS = { 
+/** Common hadith numbering format: Arabic-Indic digits + dash + space. */
+numbered: "{{raqms}} {{dash}} " };
 /** Expands composite tokens (e.g. `{{numbered}}`) to their underlying template form. */
 const expandCompositeTokensInTemplate = (template) => {
 	let out = template;
@@ -473,11 +394,11 @@ const templateToRegex = (template) => {
 * Useful for documentation, validation, or building user interfaces
 * that show available tokens.
 *
-* @returns Array of token names (e.g., `['bab', 'basmala', 'bullet', ...]`)
+* @returns Array of token names (e.g., `['bab', 'basmalah', 'bullet', ...]`)
 *
 * @example
 * getAvailableTokens()
-* // → ['bab', 'basmala', 'bullet', 'dash', 'harf', 'kitab', 'naql', 'raqm', 'raqms']
+* // → ['bab', 'basmalah', 'bullet', 'dash', 'harf', 'kitab', 'naql', 'raqm', 'raqms']
 */
 const getAvailableTokens = () => Object.keys(TOKEN_PATTERNS);
 /**
@@ -486,13 +407,13 @@ const getAvailableTokens = () => Object.keys(TOKEN_PATTERNS);
 * Returns the raw pattern string as defined in `TOKEN_PATTERNS`,
 * without any expansion or capture group wrapping.
 *
-* @param tokenName - The token name to look up (e.g., 'raqms', 'dash')
-* @returns The regex pattern string, or `undefined` if token doesn't exist
+* @param tokenName - The token name to look up (e.g., `'raqms'`, `'dash'`, `'harfs'`)
+* @returns The regex pattern string for that known token
 *
 * @example
 * getTokenPattern('raqms')   // → '[\\u0660-\\u0669]+'
 * getTokenPattern('dash')    // → '[-–—ـ]'
-* getTokenPattern('unknown') // → undefined
+* getTokenPattern('harfs')     // → pattern for spaced isolated Arabic letter codes
 */
 const getTokenPattern = (tokenName) => TOKEN_PATTERNS[tokenName];
 /**
@@ -571,7 +492,161 @@ const applyTokenMappings = (template, mappings) => {
 const stripTokenMappings = (template) => {
 	return template.replace(/\{\{([^:}]+):[^}]+\}\}/g, "{{$1}}");
 };
-
+//#endregion
+//#region src/utils/textUtils.ts
+/**
+* Normalizes line endings to Unix-style (`\n`).
+*
+* Converts Windows (`\r\n`) and old Mac (`\r`) line endings to Unix style
+* for consistent pattern matching across platforms.
+*
+* @param content - Raw content with potentially mixed line endings
+* @returns Content with all line endings normalized to `\n`
+*/
+const normalizeLineEndings = (content) => {
+	return content.includes("\r") ? content.replace(/\r\n?/g, "\n") : content;
+};
+/**
+* Escapes regex metacharacters (parentheses and brackets) in template patterns,
+* but preserves content inside `{{...}}` token delimiters.
+*
+* This allows users to write intuitive patterns like `({{harf}}):` instead of
+* the verbose `\\({{harf}}\\):`. The escaping is applied BEFORE token expansion,
+* so tokens like `{{harf}}` which expand to `[أ-ي]` work correctly.
+*
+* @param pattern - Template pattern that may contain `()[]` and `{{tokens}}`
+* @returns Pattern with `()[]` escaped outside of `{{...}}` delimiters
+*
+* @example
+* escapeTemplateBrackets('({{harf}}): ')
+* // → '\\({{harf}}\\): '
+*
+* @example
+* escapeTemplateBrackets('[{{raqm}}] ')
+* // → '\\[{{raqm}}\\] '
+*
+* @example
+* escapeTemplateBrackets('{{harf}}')
+* // → '{{harf}}' (unchanged - no brackets outside tokens)
+*/
+const escapeTemplateBrackets = (pattern) => {
+	return pattern.replace(/(\{\{[^}]*\}\})|([()[\]])/g, (_match, token, bracket) => token || `\\${bracket}`);
+};
+/**
+* Character class matching all Arabic diacritics (Tashkeel/Harakat).
+*
+* Includes the following diacritical marks:
+* - U+0640: ـ (tatweel / kashida)
+* - U+064B: ً (fathatan - double fatha)
+* - U+064C: ٌ (dammatan - double damma)
+* - U+064D: ٍ (kasratan - double kasra)
+* - U+064E: َ (fatha - short a)
+* - U+064F: ُ (damma - short u)
+* - U+0650: ِ (kasra - short i)
+* - U+0651: ّ (shadda - gemination)
+* - U+0652: ْ (sukun - no vowel)
+*
+* @internal
+*/
+const DIACRITICS_CLASS = "[ـًٌٍَُِّْ]";
+/**
+* Groups of equivalent Arabic characters.
+*
+* Characters within the same group are considered equivalent for matching purposes.
+* This handles common variations in Arabic text where different characters are
+* used interchangeably or have the same underlying meaning.
+*
+* Equivalence groups:
+* - Alef variants: ا (bare), آ (with madda), أ (with hamza above), إ (with hamza below)
+* - Ta marbuta and Ha: ة ↔ ه (often interchangeable at word endings)
+* - Alef maqsura and Ya: ى ↔ ي (often interchangeable at word endings)
+*
+* @internal
+*/
+const EQUIV_GROUPS = [
+	[
+		"ا",
+		"آ",
+		"أ",
+		"إ"
+	],
+	["ة", "ه"],
+	["ى", "ي"]
+];
+const DIACRITICS_AND_MARKS_REGEX = new RegExp(ARABIC_MARKS_CLASS, "g");
+/**
+* Escapes a string for safe inclusion in a regular expression.
+*
+* Escapes all regex metacharacters: `.*+?^${}()|[\]\\`
+*
+* @param s - Any string to escape
+* @returns String with regex metacharacters escaped
+*
+* @example
+* escapeRegex('hello.world')   // → 'hello\\.world'
+* escapeRegex('[test]')        // → '\\[test\\]'
+* escapeRegex('a+b*c?')        // → 'a\\+b\\*c\\?'
+*/
+const escapeRegex = (s) => s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+const getEquivClass = (ch) => {
+	const group = EQUIV_GROUPS.find((g) => g.includes(ch));
+	return group ? `[${group.map(escapeRegex).join("")}]` : escapeRegex(ch);
+};
+const normalizeArabicLight = (str) => {
+	return str.normalize("NFC").replace(/[\u200C\u200D]/g, "").replace(/\s+/g, " ").trim();
+};
+/**
+* Normalizes Arabic text for exact comparisons while tolerating common variants.
+*
+* This removes Arabic diacritics, collapses whitespace, removes joiners, and
+* maps common equivalent letters to a shared canonical form:
+* - ا/آ/أ/إ -> ا
+* - ة/ه -> ه
+* - ى/ي -> ي
+*/
+const normalizeArabicForComparison = (text) => {
+	return Array.from(normalizeArabicLight(text).replace(DIACRITICS_AND_MARKS_REGEX, "")).map((ch) => {
+		if (ch === "آ" || ch === "أ" || ch === "إ") return "ا";
+		if (ch === "ة") return "ه";
+		if (ch === "ى") return "ي";
+		return ch;
+	}).join("");
+};
+const makeDiacriticInsensitive = (text) => {
+	const diacriticsMatcher = `${DIACRITICS_CLASS}*`;
+	return Array.from(normalizeArabicLight(text)).map((ch) => getEquivClass(ch) + diacriticsMatcher).join("");
+};
+const isCombiningMarkOrSelector = (char) => {
+	if (!char) return false;
+	return /\p{M}/u.test(char) || char === "︎" || char === "️";
+};
+const isJoiner = (char) => char === "‌" || char === "‍";
+/**
+* Ensures the position does not split a grapheme cluster (surrogate pairs,
+* combining marks, or zero-width joiners / variation selectors).
+*
+* This is only used as a last-resort fallback when we are forced to split
+* near a hard limit (e.g. maxContentLength with no safe whitespace/punctuation).
+*/
+const adjustForUnicodeBoundary = (content, position) => {
+	let adjusted = position;
+	while (adjusted > 0) {
+		const high = content.charCodeAt(adjusted - 1);
+		const low = content.charCodeAt(adjusted);
+		if (high >= 55296 && high <= 56319 && low >= 56320 && low <= 57343) {
+			adjusted -= 1;
+			continue;
+		}
+		const nextChar = content[adjusted];
+		const prevChar = content[adjusted - 1];
+		if (isCombiningMarkOrSelector(nextChar) || isJoiner(nextChar) || isJoiner(prevChar)) {
+			adjusted -= 1;
+			continue;
+		}
+		break;
+	}
+	return adjusted;
+};
 //#endregion
 //#region src/analysis/shared.ts
 const escapeSignatureLiteral = (s) => s.replace(/[.*+?^${}|\\{}]/g, "\\$&");
@@ -632,7 +707,6 @@ const findBestTokenMatchAt = (s, pos, compiled, isArabicLetter) => {
 };
 const isArabicLetter = (ch) => /\p{Script=Arabic}/u.test(ch) && /\p{L}/u.test(ch);
 const isCommonDelimiter = (ch) => /[:：\-–—ـ،؛.?!؟()[\]{}]/u.test(ch);
-
 //#endregion
 //#region src/analysis/line-starts.ts
 const resolveOptions$1 = (options = {}) => ({
@@ -658,65 +732,141 @@ const compareBySpecificity = (a, b) => {
 	return sb.tokenCount - sa.tokenCount || sb.literalLen - sa.literalLen || b.count - a.count || a.pattern.localeCompare(b.pattern);
 };
 const compareByCount = (a, b) => b.count !== a.count ? b.count - a.count : compareBySpecificity(a, b);
-/** Remove trailing whitespace placeholders */
-const trimTrailingWs = (out, mode) => {
-	const suffix = mode === "regex" ? "\\s*" : " ";
-	while (out.endsWith(suffix)) out = out.slice(0, -suffix.length);
-	return out;
-};
-/** Try to extract first word for fallback */
-const extractFirstWord = (s) => (s.match(/^[^\s:،؛.?!؟]+/u) ?? [])[0] ?? null;
-/** Consume prefix matchers at current position */
-const consumePrefixes = (s, pos, out, matchers, ws) => {
-	let matched = false;
+const appendPrefix = (s, pos, out, matchers, ws) => {
 	for (const re of matchers) {
 		if (pos >= s.length) break;
 		const m = re.exec(s.slice(pos));
 		if (!m?.index && m?.[0]) {
 			out += escapeSignatureLiteral(m[0]);
 			pos += m[0].length;
-			matched = true;
 			const wsm = /^[ \t]+/u.exec(s.slice(pos));
 			if (wsm) {
 				pos += wsm[0].length;
 				out = appendWs(out, ws);
 			}
+			return {
+				matched: true,
+				out,
+				pos
+			};
 		}
 	}
 	return {
-		matched,
+		matched: false,
 		out,
 		pos
 	};
 };
-/** Try to match a token at current position and append to signature */
-const tryMatchToken = (s, pos, out, compiled) => {
+const appendToken = (s, pos, out, compiled) => {
 	const best = findBestTokenMatchAt(s, pos, compiled, isArabicLetter);
-	if (!best) return {
-		matched: false,
-		out,
-		pos
-	};
-	return {
+	return best ? {
 		matched: true,
 		out: `${out}{{${best.token}}}`,
 		pos: pos + best.text.length
+	} : {
+		matched: false,
+		out,
+		pos
 	};
 };
-/** Try to match a delimiter at current position */
-const tryMatchDelimiter = (s, pos, out) => {
+const appendDelimiter = (s, pos, out) => {
 	const ch = s[pos];
-	if (!ch || !isCommonDelimiter(ch)) return {
+	return ch && isCommonDelimiter(ch) ? {
+		matched: true,
+		out: `${out}${escapeSignatureLiteral(ch)}`,
+		pos: pos + 1
+	} : {
 		matched: false,
 		out,
 		pos
 	};
-	return {
-		matched: true,
-		out: out + escapeSignatureLiteral(ch),
-		pos: pos + 1
+};
+const appendFallbackWord = (s, pos, out) => {
+	const word = extractFirstWord(s.slice(pos));
+	return word ? `${out}${escapeSignatureLiteral(word)}` : null;
+};
+const consumeLineStartStep = (s, pos, out, compiled, opts, matchedAny, matchedToken) => {
+	const ws = skipWhitespace$1(s, pos, out, opts.whitespace);
+	if (ws.skipped) return {
+		done: false,
+		matchedAny,
+		matchedToken,
+		out: ws.out,
+		pos: ws.pos,
+		steps: 0
+	};
+	const tok = appendToken(s, pos, out, compiled);
+	if (tok.matched) return {
+		done: false,
+		matchedAny: true,
+		matchedToken: true,
+		out: tok.out,
+		pos: tok.pos,
+		steps: 1
+	};
+	if (matchedAny) {
+		const delim = appendDelimiter(s, pos, out);
+		if (delim.matched) return {
+			done: false,
+			matchedAny,
+			matchedToken,
+			out: delim.out,
+			pos: delim.pos,
+			steps: 0
+		};
+		if (opts.includeFirstWordFallback && !matchedToken) {
+			const fallback = appendFallbackWord(s, pos, out);
+			if (fallback) return {
+				done: true,
+				matchedAny,
+				matchedToken,
+				out: fallback,
+				pos,
+				steps: 1
+			};
+		}
+		return {
+			done: true,
+			matchedAny,
+			matchedToken,
+			out,
+			pos,
+			steps: 0
+		};
+	}
+	if (!opts.includeFirstWordFallback) return {
+		done: true,
+		matchedAny,
+		matchedToken,
+		out,
+		pos,
+		steps: 0
+	};
+	const fallback = appendFallbackWord(s, pos, out);
+	return fallback ? {
+		done: true,
+		matchedAny: true,
+		matchedToken,
+		out: fallback,
+		pos,
+		steps: 0
+	} : {
+		done: true,
+		matchedAny,
+		matchedToken,
+		out,
+		pos,
+		steps: 0
 	};
 };
+/** Remove trailing whitespace placeholders */
+const trimTrailingWs = (out, mode) => {
+	const suffix = mode === "regex" ? "\\s*" : " ";
+	while (out.endsWith(suffix)) out = out.slice(0, -suffix.length);
+	return out;
+};
+/** Try to extract first word for fallback */
+const extractFirstWord = (s) => (s.match(/^[^\s:،؛.?!؟]+/u) ?? [])[0] ?? null;
 /** Skip whitespace at position */
 const skipWhitespace$1 = (s, pos, out, ws) => {
 	const m = /^[ \t]+/u.exec(s.slice(pos));
@@ -737,47 +887,25 @@ const tokenizeLineStart = (line, tokenNames, opts) => {
 	const s = (opts.normalizeArabicDiacritics ? stripArabicDiacritics(trimmed) : trimmed).slice(0, opts.prefixChars);
 	const compiled = compileTokenRegexes(tokenNames);
 	let pos = 0, out = "", matchedAny = false, matchedToken = false, steps = 0;
-	const prefix = consumePrefixes(s, pos, out, opts.prefixMatchers, opts.whitespace);
+	const prefix = appendPrefix(s, pos, out, opts.prefixMatchers, opts.whitespace);
 	pos = prefix.pos;
 	out = prefix.out;
 	matchedAny = prefix.matched;
 	while (steps < 6 && pos < s.length) {
-		const ws = skipWhitespace$1(s, pos, out, opts.whitespace);
-		if (ws.skipped) {
-			pos = ws.pos;
-			out = ws.out;
-			continue;
-		}
-		const tok = tryMatchToken(s, pos, out, compiled);
-		if (tok.matched) {
-			pos = tok.pos;
-			out = tok.out;
-			matchedAny = matchedToken = true;
-			steps++;
-			continue;
-		}
-		if (matchedAny) {
-			const delim = tryMatchDelimiter(s, pos, out);
-			if (delim.matched) {
-				pos = delim.pos;
-				out = delim.out;
-				continue;
-			}
-		}
-		if (matchedAny) {
-			if (opts.includeFirstWordFallback && !matchedToken) {
-				const word = extractFirstWord(s.slice(pos));
-				if (word) {
-					out += escapeSignatureLiteral(word);
-					steps++;
-				}
-			}
+		const next = consumeLineStartStep(s, pos, out, compiled, opts, matchedAny, matchedToken);
+		if (next.done) {
+			if (!next.matchedAny && !next.matchedToken && next.out === out && next.pos === pos) return null;
+			if (next.steps > 0) steps += next.steps;
+			matchedAny = next.matchedAny;
+			matchedToken = next.matchedToken;
+			out = next.out;
 			break;
 		}
-		if (!opts.includeFirstWordFallback) return null;
-		const word = extractFirstWord(s.slice(pos));
-		if (!word) return null;
-		return escapeSignatureLiteral(word);
+		pos = next.pos;
+		out = next.out;
+		matchedAny = next.matchedAny;
+		matchedToken = next.matchedToken;
+		steps += next.steps;
 	}
 	return matchedAny ? trimTrailingWs(out, opts.whitespace) : null;
 };
@@ -821,7 +949,6 @@ const analyzeCommonLineStarts = (pages, options = {}) => {
 		pattern
 	})).filter((p) => p.count >= opts.minCount).sort(comparator).slice(0, opts.topK);
 };
-
 //#endregion
 //#region src/analysis/repeating-sequences.ts
 const resolveOptions = (options) => {
@@ -843,6 +970,7 @@ const resolveOptions = (options) => {
 const createRawCursor = (text, normalize) => {
 	let rawPos = 0;
 	return {
+		/** Advance cursor, returning the raw text chunk consumed */
 		advance(normalizedLen) {
 			if (!normalize) {
 				const chunk = text.slice(rawPos, rawPos + normalizedLen);
@@ -947,23 +1075,27 @@ const buildExample = (page, window, contextChars) => {
 		text: page.content.slice(start, end)
 	};
 };
+const recordPattern = (page, window, opts, stats) => {
+	if (opts.requireToken && !hasTokenInWindow(window)) return;
+	const pattern = buildPattern(window, opts.whitespace);
+	let entry = stats.get(pattern);
+	if (!entry) {
+		if (stats.size >= opts.maxUniquePatterns) return;
+		entry = {
+			count: 0,
+			examples: [],
+			...computeWindowStats(window)
+		};
+		stats.set(pattern, entry);
+	}
+	entry.count++;
+	if (entry.examples.length < opts.maxExamples) entry.examples.push(buildExample(page, window, opts.contextChars));
+};
 /** Extract N-grams from a single page */
 const extractPageNgrams = (page, items, opts, stats) => {
-	for (let i = 0; i <= items.length - opts.minElements; i++) for (let n = opts.minElements; n <= Math.min(opts.maxElements, items.length - i); n++) {
-		const window = items.slice(i, i + n);
-		if (opts.requireToken && !hasTokenInWindow(window)) continue;
-		const pattern = buildPattern(window, opts.whitespace);
-		if (!stats.has(pattern)) {
-			if (stats.size >= opts.maxUniquePatterns) continue;
-			stats.set(pattern, {
-				count: 0,
-				examples: [],
-				...computeWindowStats(window)
-			});
-		}
-		const entry = stats.get(pattern);
-		entry.count++;
-		if (entry.examples.length < opts.maxExamples) entry.examples.push(buildExample(page, window, opts.contextChars));
+	for (let i = 0; i <= items.length - opts.minElements; i++) {
+		const maxWindowSize = Math.min(opts.maxElements, items.length - i);
+		for (let n = opts.minElements; n <= maxWindowSize; n++) recordPattern(page, items.slice(i, i + n), opts, stats);
 	}
 };
 /**
@@ -985,7 +1117,6 @@ const analyzeRepeatingSequences = (pages, options) => {
 		pattern
 	}));
 };
-
 //#endregion
 //#region src/detection.ts
 /**
@@ -1147,7 +1278,6 @@ const analyzeTextForRule = (text) => {
 		...suggestPatternConfig(detected)
 	};
 };
-
 //#endregion
 //#region src/types/rules.ts
 /**
@@ -1170,9 +1300,9 @@ const PATTERN_TYPE_KEYS = [
 	"lineStartsAfter",
 	"lineEndsWith",
 	"template",
-	"regex"
+	"regex",
+	"dictionaryEntry"
 ];
-
 //#endregion
 //#region src/optimization/optimize-rules.ts
 const MERGEABLE_KEYS = new Set([
@@ -1190,11 +1320,17 @@ const getPatternArray = (rule, key) => {
 };
 const getPatternString = (rule, key) => {
 	const value = rule[key];
-	return typeof value === "string" ? value : Array.isArray(value) ? value.join("\n") : "";
+	return typeof value === "string" ? value : Array.isArray(value) ? value.join("\n") : value ? JSON.stringify(value) : "";
 };
 const normalizePatterns = (patterns) => [...new Set(patterns)].sort((a, b) => b.length - a.length || a.localeCompare(b));
+const getDictionaryEntrySpecificityScore = (rule) => {
+	if (!("dictionaryEntry" in rule)) return 0;
+	const { allowCommaSeparated = false, allowParenthesized = false, allowWhitespaceBeforeColon = false, maxLetters = 10, midLineSubentries = true, minLetters = 2, stopWords } = rule.dictionaryEntry;
+	return minLetters * 20 + maxLetters + (allowCommaSeparated ? 0 : 120) + (allowParenthesized ? 0 : 60) + (allowWhitespaceBeforeColon ? 0 : 20) + (midLineSubentries ? 0 : 160) + Math.min(stopWords.length, 25);
+};
 const getSpecificityScore = (rule) => {
 	const key = getPatternKey(rule);
+	if (key === "dictionaryEntry") return getDictionaryEntrySpecificityScore(rule);
 	return MERGEABLE_KEYS.has(key) ? getPatternArray(rule, key).reduce((max, p) => Math.max(max, p.length), 0) : getPatternString(rule, key).length;
 };
 const createMergeKey = (rule) => {
@@ -1231,7 +1367,6 @@ const optimizeRules = (rules) => {
 		rules: output.sort((a, b) => getSpecificityScore(b) - getSpecificityScore(a))
 	};
 };
-
 //#endregion
 //#region src/preprocessing/transforms.ts
 /** Helper for exhaustive switch checking - TypeScript will error if a case is missed */
@@ -1340,170 +1475,115 @@ const applyPreprocessToPage = (content, pageId, transforms) => {
 	}
 	return result;
 };
-
 //#endregion
-//#region src/segmentation/rule-regex.ts
-/**
-* Checks if a regex pattern contains standard (anonymous) capturing groups.
-*
-* Detects standard capturing groups `(...)` while excluding:
-* - Non-capturing groups `(?:...)`
-* - Lookahead assertions `(?=...)` and `(?!...)`
-* - Lookbehind assertions `(?<=...)` and `(?<!...)`
-* - Named groups `(?<name>...)` (start with `(?` so excluded here)
-*
-* NOTE: Named capture groups are still captures, but they're tracked via `captureNames`.
-*/
-const hasCapturingGroup = (pattern) => /\((?!\?)/.test(pattern);
-/**
-* Extracts named capture group names from a regex pattern.
-*
-* Parses patterns like `(?<num>[0-9]+)` and returns `['num']`.
-*
-* @example
-* extractNamedCaptureNames('^(?<num>[٠-٩]+)\\s+') // ['num']
-* extractNamedCaptureNames('^(?<a>\\d+)(?<b>\\w+)') // ['a', 'b']
-* extractNamedCaptureNames('^\\d+') // []
-*/
-const extractNamedCaptureNames = (pattern) => [...pattern.matchAll(/\(\?<([^>]+)>/g)].map((m) => m[1]).filter((n) => !n.startsWith("_r") && !n.startsWith("_w"));
-/**
-* Safely compiles a regex pattern, throwing a helpful error if invalid.
-*/
-const compileRuleRegex = (pattern) => {
-	try {
-		return new RegExp(pattern, "gmu");
-	} catch (error) {
-		throw new Error(`Invalid regex pattern: ${pattern}\n  Cause: ${error instanceof Error ? error.message : String(error)}`);
+//#region src/segmentation/arabic-dictionary-rule.ts
+const uniqueCanonicalWords = (words) => {
+	const seen = /* @__PURE__ */ new Set();
+	const result = [];
+	for (const word of words) {
+		const normalized = normalizeArabicForComparison(word);
+		if (!normalized || seen.has(normalized)) continue;
+		seen.add(normalized);
+		result.push(word);
 	}
+	return result;
 };
-/**
-* Processes a pattern string by expanding tokens and optionally applying fuzzy matching.
-*
-* Brackets `()[]` outside `{{tokens}}` are auto-escaped.
-*/
-const processPattern = (pattern, fuzzy, capturePrefix) => {
-	const { pattern: expanded, captureNames } = expandTokensWithCaptures(escapeTemplateBrackets(pattern), fuzzy ? makeDiacriticInsensitive : void 0, capturePrefix);
+const buildStopAlternation = (stopWords) => {
+	const unique = uniqueCanonicalWords(stopWords);
+	if (unique.length === 0) return "";
+	return unique.map((word) => makeDiacriticInsensitive(normalizeArabicForComparison(word))).join("|");
+};
+const buildHeadwordBody = ({ allowCommaSeparated, colonPattern, stopAlternation, stopwordBody, unit }) => {
+	if (!stopAlternation) return allowCommaSeparated ? `${unit}(?:\\s*[،,]\\s*${unit})*` : unit;
+	const guardedUnit = `(?!(?:${stopwordBody})${allowCommaSeparated ? `(?:\\s*[،,]\\s*|${colonPattern})` : colonPattern})${unit}`;
+	return allowCommaSeparated ? `${guardedUnit}(?:\\s*[،,]\\s*${guardedUnit})*` : guardedUnit;
+};
+const buildBalancedMarker = ({ allowParenthesized, allowWhitespaceBeforeColon, captureName, headwordBody }) => {
+	const colon = allowWhitespaceBeforeColon ? "\\s*:" : ":";
+	const withCapture = `(?<${captureName}>${headwordBody})`;
+	if (!allowParenthesized) return `${withCapture}${colon}`;
+	return `(?:\\(\\s*${withCapture}\\s*\\)|${withCapture})${colon}`;
+};
+const validateDictionaryEntryOptions = ({ captureName = "lemma", maxLetters = 10, minLetters = 2 }) => {
+	if (!Number.isInteger(minLetters) || minLetters < 1) throw new Error(`createArabicDictionaryEntryRule: minLetters must be an integer >= 1, got ${minLetters}`);
+	if (!Number.isInteger(maxLetters) || maxLetters < minLetters) throw new Error(`createArabicDictionaryEntryRule: maxLetters must be an integer >= minLetters, got ${maxLetters}`);
+	if (!captureName.match(/^[A-Za-z_]\w*$/)) throw new Error(`createArabicDictionaryEntryRule: invalid captureName "${captureName}"`);
+};
+const buildArabicDictionaryEntryRegexSource = ({ allowCommaSeparated = false, allowParenthesized = false, allowWhitespaceBeforeColon = false, captureName = "lemma", maxLetters = 10, midLineSubentries = true, minLetters = 2, stopWords }, capturePrefix) => {
+	validateDictionaryEntryOptions({
+		captureName,
+		maxLetters,
+		minLetters
+	});
+	const zeroWidthPrefix = "[\\u200E\\u200F\\u061C\\u200B\\u200C\\u200D\\uFEFF]*";
+	const wawWithMarks = `و${ARABIC_MARKS_CLASS}*`;
+	const alWithMarks = `ا${ARABIC_MARKS_CLASS}*ل${ARABIC_MARKS_CLASS}*`;
+	const lemmaUnit = `(?:${wawWithMarks})?(?:${alWithMarks})?${`${ARABIC_LETTER_WITH_OPTIONAL_MARKS_PATTERN}(?:${ARABIC_LETTER_WITH_OPTIONAL_MARKS_PATTERN}){${minLetters - 1},${maxLetters - 1}}`}`;
+	const stopAlternation = buildStopAlternation(stopWords);
+	const lemmaBody = buildHeadwordBody({
+		allowCommaSeparated,
+		colonPattern: allowWhitespaceBeforeColon ? "\\s*:" : ":",
+		stopAlternation,
+		stopwordBody: stopAlternation ? `(?:${wawWithMarks})?(?:${stopAlternation})` : "",
+		unit: lemmaUnit
+	});
+	const lineStartBoundary = `(?:(?<=^)|(?<=\\n))${zeroWidthPrefix}`;
+	const midLineTrigger = allowParenthesized ? `(?<=\\s)(?=(?:\\(\\s*)?${wawWithMarks}(?:${alWithMarks})?)` : `(?<=\\s)(?=${wawWithMarks}(?:${alWithMarks})?)`;
+	const prefixedCaptureName = capturePrefix ? `${capturePrefix}${captureName}` : captureName;
+	const regex = `(?:${lineStartBoundary}${midLineSubentries ? `|${midLineTrigger}` : ""})` + buildBalancedMarker({
+		allowParenthesized,
+		allowWhitespaceBeforeColon,
+		captureName: prefixedCaptureName,
+		headwordBody: lemmaBody
+	});
 	return {
-		captureNames,
-		pattern: expanded
+		captureNames: [prefixedCaptureName],
+		regex
 	};
 };
 /**
-* Processes a breakpoint pattern by expanding tokens only.
+* Creates a reusable split rule for Arabic dictionary entries.
 *
-* Unlike `processPattern`, this does NOT escape brackets because breakpoints
-* are treated as raw regex patterns (like the `regex` rule type).
-* Users have full control over regex syntax including `(?:...)` groups.
-*/
-const processBreakpointPattern = (pattern) => {
-	const { pattern: expanded } = expandTokensWithCaptures(pattern);
-	return expanded;
-};
-const buildLineStartsAfterRegexSource = (patterns, fuzzy, capturePrefix) => {
-	const processed = patterns.map((p) => processPattern(p, fuzzy, capturePrefix));
-	const alternatives = processed.map((p, i) => `(?<_r${i}>${p.pattern})`).join("|");
-	return {
-		captureNames: processed.flatMap((p) => p.captureNames),
-		regex: `^[\\u200E\\u200F\\u061C\\u200B\\u200C\\u200D\\uFEFF]*(?:${alternatives})${capturePrefix ? `(?<${capturePrefix}__content>.*)` : "(.*)"}`
-	};
-};
-const buildLineStartsWithRegexSource = (patterns, fuzzy, capturePrefix) => {
-	const processed = patterns.map((p) => processPattern(p, fuzzy, capturePrefix));
-	const alternatives = processed.map((p, i) => `(?<_r${i}>${p.pattern})`).join("|");
-	return {
-		captureNames: processed.flatMap((p) => p.captureNames),
-		regex: `^[\\u200E\\u200F\\u061C\\u200B\\u200C\\u200D\\uFEFF]*(?:${alternatives})`
-	};
-};
-const buildLineEndsWithRegexSource = (patterns, fuzzy, capturePrefix) => {
-	const processed = patterns.map((p) => processPattern(p, fuzzy, capturePrefix));
-	const alternatives = processed.map((p, i) => `(?<_r${i}>${p.pattern})`).join("|");
-	return {
-		captureNames: processed.flatMap((p) => p.captureNames),
-		regex: `(?:${alternatives})$`
-	};
-};
-const buildTemplateRegexSource = (template, capturePrefix) => {
-	const { pattern, captureNames } = expandTokensWithCaptures(escapeTemplateBrackets(template), void 0, capturePrefix);
-	return {
-		captureNames,
-		regex: pattern
-	};
-};
-/**
-* Builds a compiled regex and metadata from a split rule.
+* The returned rule preserves authoring intent as a serializable
+* `{ dictionaryEntry: ... }` pattern rather than eagerly compiling to a raw
+* regex string.
 *
-* Behavior mirrors the previous implementation in `segmenter.ts`.
+* @example
+* createArabicDictionaryEntryRule({
+*   stopWords: ['وقيل', 'ويقال', 'قال'],
+*   pageStartPrevWordStoplist: ['قال', 'وقيل', 'ويقال'],
+* })
+*
+* @example
+* createArabicDictionaryEntryRule({
+*   allowParenthesized: true,
+*   allowWhitespaceBeforeColon: true,
+*   allowCommaSeparated: true,
+*   stopWords: ['الليث', 'العجاج'],
+* })
 */
-const buildRuleRegex = (rule, capturePrefix) => {
-	const { lineStartsWith, lineStartsAfter, lineEndsWith, template, regex } = rule;
-	const fuzzy = rule.fuzzy ?? shouldDefaultToFuzzy([
-		...lineStartsWith ?? [],
-		...lineStartsAfter ?? [],
-		...lineEndsWith ?? []
-	]);
-	if (lineStartsAfter?.length) {
-		const { regex: lsaRegex, captureNames } = buildLineStartsAfterRegexSource(lineStartsAfter, fuzzy, capturePrefix);
-		return {
-			captureNames,
-			regex: compileRuleRegex(lsaRegex),
-			usesCapture: true,
-			usesLineStartsAfter: true
-		};
-	}
-	let finalRegex = regex;
-	let allCaptureNames = [];
-	if (lineStartsWith?.length) {
-		const res = buildLineStartsWithRegexSource(lineStartsWith, fuzzy, capturePrefix);
-		finalRegex = res.regex;
-		allCaptureNames = res.captureNames;
-	}
-	if (lineEndsWith?.length) {
-		const res = buildLineEndsWithRegexSource(lineEndsWith, fuzzy, capturePrefix);
-		finalRegex = res.regex;
-		allCaptureNames = res.captureNames;
-	}
-	if (template) {
-		const res = buildTemplateRegexSource(template, capturePrefix);
-		finalRegex = res.regex;
-		allCaptureNames = [...allCaptureNames, ...res.captureNames];
-	}
-	if (!finalRegex) throw new Error("Rule must specify exactly one pattern type: regex, template, lineStartsWith, lineStartsAfter, or lineEndsWith");
-	if (allCaptureNames.length === 0) allCaptureNames = extractNamedCaptureNames(finalRegex);
+const createArabicDictionaryEntryRule = ({ allowCommaSeparated = false, allowParenthesized = false, allowWhitespaceBeforeColon = false, captureName = "lemma", maxLetters = 10, meta, midLineSubentries = true, minLetters = 2, pageStartPrevWordStoplist, samePagePrevWordStoplist, stopWords }) => {
+	validateDictionaryEntryOptions({
+		captureName,
+		maxLetters,
+		minLetters
+	});
 	return {
-		captureNames: allCaptureNames,
-		regex: compileRuleRegex(finalRegex),
-		usesCapture: hasCapturingGroup(finalRegex),
-		usesLineStartsAfter: false
+		dictionaryEntry: {
+			allowCommaSeparated,
+			allowParenthesized,
+			allowWhitespaceBeforeColon,
+			captureName,
+			maxLetters,
+			midLineSubentries,
+			minLetters,
+			stopWords: uniqueCanonicalWords(stopWords)
+		},
+		meta,
+		pageStartPrevWordStoplist,
+		samePagePrevWordStoplist
 	};
 };
-
-//#endregion
-//#region src/segmentation/breakpoint-constants.ts
-/**
-* Shared constants for segmentation breakpoint processing.
-*/
-/**
-* Threshold for using offset-based fast path in boundary processing.
-*
-* Below this: accurate string-search (handles offset drift from structural rules).
-* At or above this: O(n) arithmetic (performance critical for large books).
-*
-* The value of 1000 is chosen based on typical Arabic book sizes:
-* - Sahih al-Bukhari: ~1000-3000 pages
-* - Standard hadith collections: 1000-7000 pages
-* - Large aggregated corpora: 10k-50k pages
-*
-* For segments ≥1000 pages, the performance gain from offset-based slicing
-* outweighs the minor accuracy loss from potential offset drift.
-*
-* @remarks
-* Fast path is skipped when:
-* - `maxContentLength` is set (requires character-accurate splitting)
-* - `debugMetaKey` is set (requires proper provenance tracking)
-* - Content was structurally modified by marker stripping (offsets may drift)
-*/
-const FAST_PATH_THRESHOLD = 1e3;
 const WINDOW_PREFIX_LENGTHS = [
 	80,
 	60,
@@ -1530,23 +1610,6 @@ const STOP_CHARACTERS = /[\s\n.,;!?؛،۔۝۞]/;
 * Matches outside this range are rejected unless `ignoreDeviation` is active.
 */
 const MAX_DEVIATION = 2e3;
-/**
-* Penalty score applied to non-newline anchor candidates.
-*
-* Designed to prioritize newline-aligned boundaries unless a whitespace match is
-* significantly closer (within 20 chars). Handles cases where marker stripping
-* shifts the boundary slightly.
-*/
-const NON_NEWLINE_PENALTY = 20;
-/**
-* Limit for inferring start offset from a relaxed search (characters).
-*
-* If the relaxed search finds a match more than this distance away from the
-* expected position, we assume it's a false positive (e.g. repeated content)
-* and do not use it to infer the start offset.
-*/
-const INFERENCE_PROXIMITY_LIMIT = 500;
-
 //#endregion
 //#region src/segmentation/match-utils.ts
 /**
@@ -1665,7 +1728,6 @@ const extractDebugIndex = (groups, prefix) => {
 		if (!Number.isNaN(idx)) return idx;
 	}
 };
-
 //#endregion
 //#region src/segmentation/breakpoint-utils.ts
 /**
@@ -2067,8 +2129,8 @@ const findAnchorCandidates = (content, prefix, start, end) => {
 /** Selects the best anchor candidate, prioritizing newlines then proximity to boundary */
 const selectBestAnchor = (candidates, expectedBoundary) => {
 	return candidates.reduce((best, curr) => {
-		const bestScore = Math.abs(best.pos - expectedBoundary) + (best.isNewline ? 0 : NON_NEWLINE_PENALTY);
-		return Math.abs(curr.pos - expectedBoundary) + (curr.isNewline ? 0 : NON_NEWLINE_PENALTY) < bestScore ? curr : best;
+		const bestScore = Math.abs(best.pos - expectedBoundary) + (best.isNewline ? 0 : 20);
+		return Math.abs(curr.pos - expectedBoundary) + (curr.isNewline ? 0 : 20) < bestScore ? curr : best;
 	});
 };
 /**
@@ -2122,7 +2184,7 @@ const resolveBoundaryMatch = (segmentContent, pageIdx, rawBoundary, startOffsetI
 		if (relaxedPos > 0) {
 			const inferredStartOffset = rawBoundary - relaxedPos;
 			const currentExpected = Math.max(0, rawBoundary - startOffsetInFromPage);
-			if (inferredStartOffset >= 0 && Math.abs(relaxedPos - currentExpected) < INFERENCE_PROXIMITY_LIMIT) {
+			if (inferredStartOffset >= 0 && Math.abs(relaxedPos - currentExpected) < 500) {
 				startOffsetInFromPage = inferredStartOffset;
 				expectedBoundary = Math.max(0, rawBoundary - startOffsetInFromPage);
 				pos = relaxedPos;
@@ -2196,7 +2258,7 @@ const buildBoundaryPositionsAccurate = (segmentContent, fromIdx, toIdx, pageCoun
 const buildBoundaryPositions = (segmentContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, logger) => {
 	const pageCount = toIdx - fromIdx + 1;
 	const expectedLength = (cumulativeOffsets[toIdx + 1] ?? 0) - (cumulativeOffsets[fromIdx] ?? 0);
-	if (pageCount >= FAST_PATH_THRESHOLD && segmentContent.length === expectedLength) return buildBoundaryPositionsFastPath(segmentContent, fromIdx, toIdx, pageCount, cumulativeOffsets, logger);
+	if (pageCount >= 1e3 && segmentContent.length === expectedLength) return buildBoundaryPositionsFastPath(segmentContent, fromIdx, toIdx, pageCount, cumulativeOffsets, logger);
 	return buildBoundaryPositionsAccurate(segmentContent, fromIdx, toIdx, pageCount, pageIds, normalizedPages, cumulativeOffsets, logger);
 };
 /**
@@ -2428,7 +2490,6 @@ const findSafeBreakPosition = (content, targetPosition, lookbackChars = 100) =>
 	}
 	return -1;
 };
-
 //#endregion
 //#region src/segmentation/debug-meta.ts
 const resolveDebugConfig = (debug) => {
@@ -2470,59 +2531,222 @@ const buildRuleDebugPatch = (ruleIndex, rule, wordIndex) => {
 		...word !== void 0 ? { word } : {}
 	} };
 };
-const buildBreakpointDebugPatch = (breakpointIndex, rule, wordIndex) => ({ breakpoint: {
-	index: breakpointIndex,
-	kind: rule.pattern === "" ? "pageBoundary" : "pattern",
-	pattern: rule.pattern ?? rule.regex,
-	...wordIndex !== void 0 ? { wordIndex } : {},
-	...wordIndex !== void 0 && rule.words ? { word: rule.words[wordIndex] } : {}
-} });
+const buildBreakpointDebugPatch = (breakpointIndex, rule, wordIndex) => ({ breakpoint: {
+	index: breakpointIndex,
+	kind: rule.pattern === "" ? "pageBoundary" : rule.regex ? "regex" : "pattern",
+	pattern: rule.pattern ?? rule.regex,
+	...wordIndex !== void 0 ? { wordIndex } : {},
+	...wordIndex !== void 0 && rule.words ? { word: rule.words[wordIndex] } : {}
+} });
+/**
+* Helper to format the debug info into a human-readable string.
+* @param meta - The segment metadata object
+* @param options - Formatting options
+*/
+const formatRuleReason = (rule, concise) => {
+	const { index, patternType, wordIndex, word } = rule;
+	if (concise) return `Rule: ${word ? `"${word}"` : patternType}`;
+	const wordInfo = word ? ` (Matched: "${word}")` : "";
+	return `Rule #${index} (${patternType})${wordIndex !== void 0 ? ` [idx:${wordIndex}]` : ""}${wordInfo}`;
+};
+const formatBreakpointReason = (breakpoint, concise) => {
+	const { index, kind, pattern, wordIndex, word } = breakpoint;
+	if (kind === "pageBoundary") return concise ? "Breakpoint: <page-boundary>" : "Page Boundary (Fallback)";
+	if (concise) return `Breakpoint: ${word ? `"${word}"` : `"${pattern}"`}`;
+	if (word) return `Breakpoint #${index} (Words) [idx:${wordIndex}] - "${word}"`;
+	return `Breakpoint #${index} (${kind}) - "${pattern}"`;
+};
+const formatContentLengthReason = (split, concise) => {
+	const { maxContentLength, splitReason } = split;
+	if (concise) return `> ${maxContentLength} (${splitReason})`;
+	return `Safety Split (${splitReason}) > ${maxContentLength}`;
+};
+/**
+* Helper to format the debug info into a human-readable string.
+* @param meta - The segment metadata object
+* @param options - Formatting options
+*/
+const getDebugReason = (meta, options) => {
+	const debug = meta?._flappa;
+	if (!debug) return "-";
+	const concise = options?.concise;
+	if (debug.rule) return formatRuleReason(debug.rule, concise);
+	if (debug.breakpoint) return formatBreakpointReason(debug.breakpoint, concise);
+	if (debug.contentLengthSplit) return formatContentLengthReason(debug.contentLengthSplit, concise);
+	return "Unknown";
+};
+/**
+* Convenience helper to get the formatted debug reason directly from a segment.
+* @param segment - The segment object
+* @param options - Formatting options
+*/
+const getSegmentDebugReason = (segment, options) => {
+	return getDebugReason(segment.meta, options);
+};
+//#endregion
+//#region src/segmentation/pattern-validator.ts
+const KNOWN_TOKENS = new Set(getAvailableTokens());
+const TOKEN_INSIDE_BRACES = /\{\{(\w+)(?::\w+)?\}\}/g;
+const buildBareTokenRegex = () => {
+	const tokens = [...KNOWN_TOKENS].sort((a, b) => b.length - a.length);
+	return new RegExp(`(?<!\\{\\{)(${tokens.join("|")})(?::\\w+)?(?!\\}\\})`, "g");
+};
+/**
+* Validates a single pattern for common issues.
+*/
+const validatePattern = (pattern, seenPatterns) => {
+	if (!pattern.trim()) return {
+		message: "Empty pattern is not allowed",
+		type: "empty_pattern"
+	};
+	if (seenPatterns.has(pattern)) return {
+		message: `Duplicate pattern: "${pattern}"`,
+		pattern,
+		type: "duplicate"
+	};
+	seenPatterns.add(pattern);
+	TOKEN_INSIDE_BRACES.lastIndex = 0;
+	for (const match of pattern.matchAll(TOKEN_INSIDE_BRACES)) {
+		const name = match[1];
+		if (!KNOWN_TOKENS.has(name)) return {
+			message: `Unknown token: {{${name}}}. Available tokens: ${[...KNOWN_TOKENS].slice(0, 5).join(", ")}...`,
+			suggestion: "Check spelling or use a known token",
+			token: name,
+			type: "unknown_token"
+		};
+	}
+	for (const match of pattern.matchAll(buildBareTokenRegex())) {
+		const [full, name] = match;
+		const idx = match.index;
+		if (pattern.slice(Math.max(0, idx - 2), idx) !== "{{" || pattern.slice(idx + full.length, idx + full.length + 2) !== "}}") return {
+			message: `Token "${name}" appears to be missing {{}}. Did you mean "{{${full}}}"?`,
+			suggestion: `{{${full}}}`,
+			token: name,
+			type: "missing_braces"
+		};
+	}
+};
 /**
-* Helper to format the debug info into a human-readable string.
-* @param meta - The segment metadata object
-* @param options - Formatting options
+* Validates an array of patterns, returning parallel array of issues.
 */
-const formatRuleReason = (rule, concise) => {
-	const { index, patternType, wordIndex, word } = rule;
-	if (concise) return `Rule: ${word ? `"${word}"` : patternType}`;
-	const wordInfo = word ? ` (Matched: "${word}")` : "";
-	return `Rule #${index} (${patternType})${wordIndex !== void 0 ? ` [idx:${wordIndex}]` : ""}${wordInfo}`;
+const validatePatternArray = (patterns) => {
+	const seen = /* @__PURE__ */ new Set();
+	const issues = patterns.map((p) => validatePattern(p, seen));
+	return issues.some(Boolean) ? issues : void 0;
 };
-const formatBreakpointReason = (breakpoint, concise) => {
-	const { index, kind, pattern, wordIndex, word } = breakpoint;
-	if (kind === "pageBoundary") return concise ? "Breakpoint: <page-boundary>" : "Page Boundary (Fallback)";
-	if (concise) return `Breakpoint: ${word ? `"${word}"` : `"${pattern}"`}`;
-	if (word) return `Breakpoint #${index} (Words) [idx:${wordIndex}] - "${word}"`;
-	return `Breakpoint #${index} (${kind}) - "${pattern}"`;
+const applyRulePatternValidation = (result, key, patterns) => {
+	if (!patterns) return false;
+	const issues = validatePatternArray(patterns);
+	if (!issues) return false;
+	result[key] = issues;
+	return true;
 };
-const formatContentLengthReason = (split, concise) => {
-	const { maxContentLength, splitReason } = split;
-	if (concise) return `> ${maxContentLength} (${splitReason})`;
-	return `Safety Split (${splitReason}) > ${maxContentLength}`;
+const validateTemplateRule = (rule, result) => {
+	if (rule.template === void 0) return false;
+	const issue = validatePattern(rule.template, /* @__PURE__ */ new Set());
+	if (!issue) return false;
+	result.template = issue;
+	return true;
+};
+const validateRegexRule = (rule, result) => {
+	if (rule.regex === void 0) return false;
+	if (!rule.regex.trim()) {
+		result.regex = {
+			message: "Empty pattern is not allowed",
+			type: "empty_pattern"
+		};
+		return true;
+	}
+	try {
+		new RegExp(rule.regex, "u");
+		return false;
+	} catch (error) {
+		result.regex = {
+			message: error instanceof Error ? error.message : String(error),
+			pattern: rule.regex,
+			type: "invalid_regex"
+		};
+		return true;
+	}
+};
+const invalidDictionaryEntryIssue = (message) => ({
+	message,
+	type: "invalid_option"
+});
+const validateDictionaryEntryRule = (rule, result) => {
+	if (!("dictionaryEntry" in rule) || !rule.dictionaryEntry) return false;
+	const issues = {};
+	const { allowCommaSeparated, allowParenthesized, allowWhitespaceBeforeColon, captureName, maxLetters, midLineSubentries, minLetters, stopWords } = rule.dictionaryEntry;
+	if (!Array.isArray(stopWords) || stopWords.some((word) => typeof word !== "string" || !word.trim())) issues.stopWords = invalidDictionaryEntryIssue("stopWords must be a string[] with non-empty entries");
+	if (allowCommaSeparated !== void 0 && typeof allowCommaSeparated !== "boolean") issues.allowCommaSeparated = invalidDictionaryEntryIssue("allowCommaSeparated must be a boolean");
+	if (allowParenthesized !== void 0 && typeof allowParenthesized !== "boolean") issues.allowParenthesized = invalidDictionaryEntryIssue("allowParenthesized must be a boolean");
+	if (allowWhitespaceBeforeColon !== void 0 && typeof allowWhitespaceBeforeColon !== "boolean") issues.allowWhitespaceBeforeColon = invalidDictionaryEntryIssue("allowWhitespaceBeforeColon must be a boolean");
+	if (midLineSubentries !== void 0 && typeof midLineSubentries !== "boolean") issues.midLineSubentries = invalidDictionaryEntryIssue("midLineSubentries must be a boolean");
+	if (captureName !== void 0 && !captureName.match(/^[A-Za-z_]\w*$/)) issues.captureName = invalidDictionaryEntryIssue(`captureName must match /^[A-Za-z_]\\w*$/, got "${captureName}"`);
+	if (minLetters !== void 0 && (!Number.isInteger(minLetters) || minLetters < 1)) issues.minLetters = invalidDictionaryEntryIssue("minLetters must be an integer >= 1");
+	if (maxLetters !== void 0 && (!Number.isInteger(maxLetters) || maxLetters < (minLetters ?? 2))) issues.maxLetters = invalidDictionaryEntryIssue(`maxLetters must be an integer >= ${minLetters ?? 2}`);
+	if (Object.keys(issues).length === 0) return false;
+	result.dictionaryEntry = issues;
+	return true;
+};
+const formatValidationIssue = (_type, issue, loc) => {
+	if (!issue) return null;
+	if (issue.type === "missing_braces") return `${loc}: Missing {{}} around token "${issue.token}"`;
+	if (issue.type === "unknown_token") return `${loc}: Unknown token "{{${issue.token}}}"`;
+	if (issue.type === "duplicate") return `${loc}: Duplicate pattern "${issue.pattern}"`;
+	if (issue.type === "invalid_regex") return `${loc}: Invalid regex (${issue.message})`;
+	return `${loc}: ${issue.message || issue.type}`;
 };
 /**
-* Helper to format the debug info into a human-readable string.
-* @param meta - The segment metadata object
-* @param options - Formatting options
+* Validates split rules for common pattern issues.
+*
+* Checks for:
+* - Missing `{{}}` around known token names (e.g., `raqms:num` instead of `{{raqms:num}}`)
+* - Unknown token names inside `{{}}` (e.g., `{{nonexistent}}`)
+* - Duplicate patterns within the same rule
+*
+* @param rules - Array of split rules to validate
+* @returns Array parallel to input with validation results (undefined if no issues)
+*
+* @example
+* const issues = validateRules([
+*   { lineStartsAfter: ['raqms:num'] },  // Missing braces
+*   { lineStartsWith: ['{{unknown}}'] }, // Unknown token
+* ]);
+* // issues[0]?.lineStartsAfter?.[0]?.type === 'missing_braces'
+* // issues[1]?.lineStartsWith?.[0]?.type === 'unknown_token'
 */
-const getDebugReason = (meta, options) => {
-	const debug = meta?._flappa;
-	if (!debug) return "-";
-	const concise = options?.concise;
-	if (debug.rule) return formatRuleReason(debug.rule, concise);
-	if (debug.breakpoint) return formatBreakpointReason(debug.breakpoint, concise);
-	if (debug.contentLengthSplit) return formatContentLengthReason(debug.contentLengthSplit, concise);
-	return "Unknown";
-};
+const validateRules = (rules) => rules.map((rule) => {
+	const result = {};
+	const startsWithIssues = applyRulePatternValidation(result, "lineStartsWith", rule.lineStartsWith);
+	const startsAfterIssues = applyRulePatternValidation(result, "lineStartsAfter", rule.lineStartsAfter);
+	const endsWithIssues = applyRulePatternValidation(result, "lineEndsWith", rule.lineEndsWith);
+	const templateIssues = validateTemplateRule(rule, result);
+	const regexIssues = validateRegexRule(rule, result);
+	const dictionaryEntryIssues = validateDictionaryEntryRule(rule, result);
+	return startsWithIssues || startsAfterIssues || endsWithIssues || templateIssues || regexIssues || dictionaryEntryIssues ? result : void 0;
+});
 /**
-* Convenience helper to get the formatted debug reason directly from a segment.
-* @param segment - The segment object
-* @param options - Formatting options
+* Formats a validation result array into a list of human-readable error messages.
+*
+* Useful for displaying validation errors in UIs.
+*
+* @param results - The result array from `validateRules()`
+* @returns Array of formatted error strings
+*
+* @example
+* const issues = validateRules(rules);
+* const errors = formatValidationReport(issues);
+* // ["Rule 1, lineStartsWith: Missing {{}} around token..."]
 */
-const getSegmentDebugReason = (segment, options) => {
-	return getDebugReason(segment.meta, options);
+const formatValidationReport = (results) => results.flatMap((result, i) => {
+	if (!result) return [];
+	return Object.entries(result).flatMap(([type, issues]) => formatValidationIssues(type, issues, i + 1));
+});
+const formatValidationIssues = (type, issues, ruleNumber) => {
+	if (type === "dictionaryEntry" && issues && typeof issues === "object" && !Array.isArray(issues)) return Object.entries(issues).map(([field, issue]) => formatValidationIssue(type, issue, `Rule ${ruleNumber}, ${type}.${field}`)).filter((msg) => msg !== null);
+	return (Array.isArray(issues) ? issues : [issues]).map((issue) => formatValidationIssue(type, issue, `Rule ${ruleNumber}, ${type}`)).filter((msg) => msg !== null);
 };
-
 //#endregion
 //#region src/segmentation/breakpoint-processor.ts
 const buildPageIdToIndexMap = (pageIds) => new Map(pageIds.map((id, i) => [id, i]));
@@ -2650,7 +2874,7 @@ const checkFastPathAlignment = (cumulativeOffsets, fullContent, fromIdx, toIdx,
 	const expectedLength = (cumulativeOffsets[toIdx + 1] ?? fullContent.length) - (cumulativeOffsets[fromIdx] ?? 0);
 	const driftTolerance = Math.max(100, fullContent.length * .01);
 	const isAligned = Math.abs(expectedLength - fullContent.length) <= driftTolerance;
-	if (!isAligned && pageCount >= FAST_PATH_THRESHOLD) logger?.warn?.("[breakpoints] Offset drift detected in fast-path candidate, falling back to slow path", {
+	if (!isAligned && pageCount >= 1e3) logger?.warn?.("[breakpoints] Offset drift detected in fast-path candidate, falling back to slow path", {
 		actualLength: fullContent.length,
 		drift: Math.abs(expectedLength - fullContent.length),
 		expectedLength,
@@ -2791,8 +3015,7 @@ const computeWindowEndPositionForIteration = (remainingContent, cursorPos, curre
 	if (maxPages === 0) {
 		const nextPageStartPos = boundaryPositions[currentFromIdx - fromIdx + 1] ?? Number.POSITIVE_INFINITY;
 		const remainingInCurrentPage = Math.max(0, nextPageStartPos - cursorPos);
-		const capped = maxContentLength ? Math.min(remainingInCurrentPage, maxContentLength) : remainingInCurrentPage;
-		return Math.min(capped, remainingContent.length);
+		return Math.min(maxContentLength ? Math.min(remainingInCurrentPage, maxContentLength) : remainingInCurrentPage, remainingContent.length);
 	}
 	const pos = getWindowEndPosition(remainingContent, currentFromIdx, windowEndIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, maxContentLength, logger);
 	return Math.min(pos, remainingContent.length);
@@ -2847,7 +3070,7 @@ const tryProcessOversizedSegmentFastPath = (segment, fromIdx, toIdx, pageIds, no
 	const pageCount = toIdx - fromIdx + 1;
 	const isAligned = checkFastPathAlignment(cumulativeOffsets, fullContent, fromIdx, toIdx, pageCount, logger);
 	const isPageBoundaryOnly = expandedBreakpoints.every((bp) => bp.regex === null && bp.excludeSet.size === 0 && bp.skipWhenRegex === null);
-	if (pageCount < FAST_PATH_THRESHOLD || !isAligned || !isPageBoundaryOnly || maxContentLength || debugMetaKey) return null;
+	if (pageCount < 1e3 || !isAligned || !isPageBoundaryOnly || maxContentLength || debugMetaKey) return null;
 	if (maxPages === 0) return processTrivialFastPath(fromIdx, toIdx, pageIds, normalizedPages, pageCount, segment.meta, debugMetaKey, logger);
 	return processOffsetFastPath(fullContent, fromIdx, toIdx, pageIds, cumulativeOffsets, maxPages, segment.meta, debugMetaKey, logger);
 };
@@ -3030,7 +3253,179 @@ const applyBreakpoints = (segments, pages, normalizedContent, maxPages, breakpoi
 	logger?.info?.("Breakpoint processing completed", { resultCount: result.length });
 	return result;
 };
-
+//#endregion
+//#region src/segmentation/rule-regex.ts
+/**
+* Checks if a regex pattern contains standard (anonymous) capturing groups.
+*
+* Detects standard capturing groups `(...)` while excluding:
+* - Non-capturing groups `(?:...)`
+* - Lookahead assertions `(?=...)` and `(?!...)`
+* - Lookbehind assertions `(?<=...)` and `(?<!...)`
+* - Named groups `(?<name>...)` (start with `(?` so excluded here)
+*
+* NOTE: Named capture groups are still captures, but they're tracked via `captureNames`.
+*/
+const hasCapturingGroup = (pattern) => /\((?!\?)/.test(pattern);
+/**
+* Extracts named capture group names from a regex pattern.
+*
+* Parses patterns like `(?<num>[0-9]+)` and returns `['num']`.
+*
+* @example
+* extractNamedCaptureNames('^(?<num>[٠-٩]+)\\s+') // ['num']
+* extractNamedCaptureNames('^(?<a>\\d+)(?<b>\\w+)') // ['a', 'b']
+* extractNamedCaptureNames('^\\d+') // []
+*/
+const extractNamedCaptureNames = (pattern) => [...pattern.matchAll(/\(\?<([A-Za-z_]\w*)>/g)].map((m) => m[1]).filter((n) => !n.startsWith("_r") && !n.startsWith("_w"));
+/**
+* Safely compiles a regex pattern, throwing a helpful error if invalid.
+*/
+const compileRuleRegex = (pattern) => {
+	try {
+		return new RegExp(pattern, "gmu");
+	} catch (error) {
+		throw new Error(`Invalid regex pattern: ${pattern}\n  Cause: ${error instanceof Error ? error.message : String(error)}`);
+	}
+};
+/**
+* Processes a pattern string by expanding tokens and optionally applying fuzzy matching.
+*
+* Brackets `()[]` outside `{{tokens}}` are auto-escaped.
+*/
+const processPattern = (pattern, fuzzy, capturePrefix) => {
+	const { pattern: expanded, captureNames } = expandTokensWithCaptures(escapeTemplateBrackets(pattern), fuzzy ? makeDiacriticInsensitive : void 0, capturePrefix);
+	return {
+		captureNames,
+		pattern: expanded
+	};
+};
+/**
+* Processes a breakpoint pattern by expanding tokens only.
+*
+* Unlike `processPattern`, this does NOT escape brackets because breakpoints
+* are treated as raw regex patterns (like the `regex` rule type).
+* Users have full control over regex syntax including `(?:...)` groups.
+*/
+const processBreakpointPattern = (pattern) => {
+	const { pattern: expanded } = expandTokensWithCaptures(pattern);
+	return expanded;
+};
+/**
+* Builds the raw regex source for a `lineStartsAfter` rule.
+*
+* Expands each pattern through `processPattern()`, combines them into an
+* alternation at the start of a line, and appends a trailing content capture.
+*
+* @param patterns - Template-like line-start markers to match
+* @param fuzzy - Whether Arabic fuzzy matching should be applied during expansion
+* @param capturePrefix - Optional prefix used for internal named captures
+* @returns Regex source plus the named captures extracted from the patterns
+*/
+const buildLineStartsAfterRegexSource = (patterns, fuzzy, capturePrefix) => {
+	const processed = patterns.map((p) => processPattern(p, fuzzy, capturePrefix));
+	const alternatives = processed.map((p, i) => `(?<_r${i}>${p.pattern})`).join("|");
+	return {
+		captureNames: processed.flatMap((p) => p.captureNames),
+		regex: `^[\\u200E\\u200F\\u061C\\u200B\\u200C\\u200D\\uFEFF]*(?:${alternatives})${capturePrefix ? `(?<${capturePrefix}__content>.*)` : "(.*)"}`
+	};
+};
+/**
+* Builds the raw regex source for a `lineStartsWith` rule.
+*
+* Expands each pattern through `processPattern()` and combines them into an
+* alternation anchored at the start of a line.
+*
+* @param patterns - Template-like line-start markers to match
+* @param fuzzy - Whether Arabic fuzzy matching should be applied during expansion
+* @param capturePrefix - Optional prefix used for internal named captures
+* @returns Regex source plus the named captures extracted from the patterns
+*/
+const buildLineStartsWithRegexSource = (patterns, fuzzy, capturePrefix) => {
+	const processed = patterns.map((p) => processPattern(p, fuzzy, capturePrefix));
+	const alternatives = processed.map((p, i) => `(?<_r${i}>${p.pattern})`).join("|");
+	return {
+		captureNames: processed.flatMap((p) => p.captureNames),
+		regex: `^[\\u200E\\u200F\\u061C\\u200B\\u200C\\u200D\\uFEFF]*(?:${alternatives})`
+	};
+};
+/**
+* Builds the raw regex source for a `lineEndsWith` rule.
+*
+* Expands each pattern through `processPattern()` and combines them into an
+* end-anchored alternation.
+*
+* @param patterns - Template-like line-end markers to match
+* @param fuzzy - Whether Arabic fuzzy matching should be applied during expansion
+* @param capturePrefix - Optional prefix used for internal named captures
+* @returns Regex source plus the named captures extracted from the patterns
+*/
+const buildLineEndsWithRegexSource = (patterns, fuzzy, capturePrefix) => {
+	const processed = patterns.map((p) => processPattern(p, fuzzy, capturePrefix));
+	const alternatives = processed.map((p, i) => `(?<_r${i}>${p.pattern})`).join("|");
+	return {
+		captureNames: processed.flatMap((p) => p.captureNames),
+		regex: `(?:${alternatives})$`
+	};
+};
+/**
+* Builds the raw regex source for a `template` rule.
+*
+* Expands tokens and named captures via `expandTokensWithCaptures()` after
+* applying `escapeTemplateBrackets()` to non-token brackets.
+*
+* @param template - Template string containing optional `{{token}}` markers
+* @param capturePrefix - Optional prefix used for internal named captures
+* @returns Regex source plus the named captures extracted from the template
+*/
+const buildTemplateRegexSource = (template, capturePrefix) => {
+	const { pattern, captureNames } = expandTokensWithCaptures(escapeTemplateBrackets(template), void 0, capturePrefix);
+	return {
+		captureNames,
+		regex: pattern
+	};
+};
+const getFuzzyCandidatePatterns = (rule) => [
+	..."lineStartsWith" in rule && Array.isArray(rule.lineStartsWith) ? rule.lineStartsWith : [],
+	..."lineStartsAfter" in rule && Array.isArray(rule.lineStartsAfter) ? rule.lineStartsAfter : [],
+	..."lineEndsWith" in rule && Array.isArray(rule.lineEndsWith) ? rule.lineEndsWith : []
+];
+const buildLineBasedRuleRegex = (rule, fuzzy, capturePrefix) => {
+	if ("lineStartsWith" in rule && Array.isArray(rule.lineStartsWith) && rule.lineStartsWith.length > 0) return buildLineStartsWithRegexSource(rule.lineStartsWith, fuzzy, capturePrefix);
+	if ("lineEndsWith" in rule && Array.isArray(rule.lineEndsWith) && rule.lineEndsWith.length > 0) return buildLineEndsWithRegexSource(rule.lineEndsWith, fuzzy, capturePrefix);
+	if ("template" in rule && typeof rule.template === "string") return buildTemplateRegexSource(rule.template, capturePrefix);
+	if ("dictionaryEntry" in rule && rule.dictionaryEntry) return buildArabicDictionaryEntryRegexSource(rule.dictionaryEntry, capturePrefix);
+	return null;
+};
+/**
+* Builds a compiled regex and metadata from a split rule.
+*
+* Behavior mirrors the previous implementation in `segmenter.ts`.
+*/
+const buildRuleRegex = (rule, capturePrefix) => {
+	const fuzzy = rule.fuzzy ?? shouldDefaultToFuzzy(getFuzzyCandidatePatterns(rule));
+	if ("lineStartsAfter" in rule && Array.isArray(rule.lineStartsAfter) && rule.lineStartsAfter.length > 0) {
+		const { regex: lsaRegex, captureNames } = buildLineStartsAfterRegexSource(rule.lineStartsAfter, fuzzy, capturePrefix);
+		return {
+			captureNames,
+			regex: compileRuleRegex(lsaRegex),
+			usesCapture: true,
+			usesLineStartsAfter: true
+		};
+	}
+	const ruleRegexSource = buildLineBasedRuleRegex(rule, fuzzy, capturePrefix);
+	let finalRegex = ruleRegexSource?.regex;
+	let allCaptureNames = ruleRegexSource?.captureNames ?? [];
+	if (!finalRegex && "regex" in rule && typeof rule.regex === "string") finalRegex = rule.regex;
+	if (!finalRegex) throw new Error("Rule must specify exactly one pattern type: regex, template, lineStartsWith, lineStartsAfter, lineEndsWith, or dictionaryEntry");
+	if (allCaptureNames.length === 0) allCaptureNames = extractNamedCaptureNames(finalRegex);
+	return {
+		captureNames: allCaptureNames,
+		regex: compileRuleRegex(finalRegex),
+		usesCapture: hasCapturingGroup(finalRegex),
+		usesLineStartsAfter: false
+	};
+};
 //#endregion
 //#region src/segmentation/fast-fuzzy-prefix.ts
 /**
@@ -3078,9 +3473,8 @@ const compileFastFuzzyTokenRule = (tokenTemplate) => {
 	const m = tokenTemplate.match(/^\{\{(\w+)\}\}$/);
 	if (!m) return null;
 	const token = m[1];
-	const tokenPattern = getTokenPattern(token);
-	if (!tokenPattern) return null;
-	const compiled = compileLiteralAlternation(tokenPattern);
+	if (!(token in TOKEN_PATTERNS)) return null;
+	const compiled = compileLiteralAlternation(getTokenPattern(token));
 	return compiled ? {
 		alternatives: compiled.alternatives,
 		token
@@ -3093,11 +3487,11 @@ const matchFastFuzzyTokenAt = (content, offset, compiled) => {
 	}
 	return null;
 };
-
 //#endregion
 //#region src/segmentation/segmenter-rule-utils.ts
 const tryCompileFastFuzzyRule = (rule) => {
-	if (!rule.fuzzy) return null;
+	const fuzzyCandidatePatterns = [..."lineStartsWith" in rule ? rule.lineStartsWith : [], ..."lineStartsAfter" in rule ? rule.lineStartsAfter : []];
+	if (!(rule.fuzzy ?? shouldDefaultToFuzzy(fuzzyCandidatePatterns))) return null;
 	if ("lineStartsWith" in rule && rule.lineStartsWith?.length === 1) {
 		const compiled = compileFastFuzzyTokenRule(rule.lineStartsWith[0]);
 		if (compiled) return {
@@ -3139,7 +3533,10 @@ const partitionRulesForMatching = (rules) => {
 			prefix: `r${index}_`,
 			rule
 		});
-		else standaloneRules.push(rule);
+		else standaloneRules.push({
+			index,
+			rule
+		});
 	}
 	return {
 		combinableRules,
@@ -3147,9 +3544,37 @@ const partitionRulesForMatching = (rules) => {
 		standaloneRules
 	};
 };
+const STRONG_SENTENCE_TERMINATORS = /[.!?؟؛۔…]$/u;
+const TRAILING_PAGE_WRAP_NOISE = /[\s\u0660-\u0669\d«»"“”'‘’()[\]{}<>]+$/u;
+const TRAILING_WORD_DELIMITERS = /[\s\u0660-\u0669\d«»"“”'‘’()[\]{}<>.,!?؟؛،:]+$/u;
+const ARABIC_WORD_REGEX = new RegExp(ARABIC_WORD_WITH_OPTIONAL_MARKS_PATTERN, "gu");
+const trimTrailingPageWrapNoise = (text) => {
+	let trimmed = text.trimEnd();
+	while (trimmed !== trimmed.replace(TRAILING_PAGE_WRAP_NOISE, "")) trimmed = trimmed.replace(TRAILING_PAGE_WRAP_NOISE, "");
+	return trimmed;
+};
+const endsWithStrongSentenceTerminator = (pageContent) => {
+	return STRONG_SENTENCE_TERMINATORS.test(trimTrailingPageWrapNoise(pageContent));
+};
+const extractLastArabicWord = (pageContent) => {
+	return [...trimTrailingPageWrapNoise(pageContent).replace(TRAILING_WORD_DELIMITERS, "").matchAll(ARABIC_WORD_REGEX)].at(-1)?.[0] ?? "";
+};
+const shouldAllowPageStartMatch = (previousPageContent, prevWordStoplist) => {
+	if (!prevWordStoplist || endsWithStrongSentenceTerminator(previousPageContent)) return true;
+	const lastWord = extractLastArabicWord(previousPageContent);
+	return !lastWord || !prevWordStoplist.has(normalizeArabicForComparison(lastWord));
+};
+const shouldAllowSamePageMatch = (contentBeforeMatch, stoplist) => {
+	if (!stoplist) return true;
+	const lastWord = extractLastArabicWord(contentBeforeMatch);
+	return !lastWord || !stoplist.has(normalizeArabicForComparison(lastWord));
+};
 const createPageStartGuardChecker = (matchContent, pageMap) => {
 	const pageStartToBoundaryIndex = new Map(pageMap.boundaries.map((b, i) => [b.start, i]));
 	const compiledPageStartPrev = /* @__PURE__ */ new Map();
+	const compiledPrevWordStoplists = /* @__PURE__ */ new Map();
+	const compiledSamePagePrevWordStoplists = /* @__PURE__ */ new Map();
+	const pageIdToBoundaryIndex = new Map(pageMap.boundaries.map((b, i) => [b.id, i]));
 	const getPageStartPrevRegex = (rule, ruleIndex) => {
 		if (compiledPageStartPrev.has(ruleIndex)) return compiledPageStartPrev.get(ruleIndex) ?? null;
 		const pattern = rule.pageStartGuard;
@@ -3161,6 +3586,33 @@ const createPageStartGuardChecker = (matchContent, pageMap) => {
 		compiledPageStartPrev.set(ruleIndex, re);
 		return re;
 	};
+	const getPrevWordStoplist = (rule, ruleIndex) => {
+		if (compiledPrevWordStoplists.has(ruleIndex)) return compiledPrevWordStoplists.get(ruleIndex) ?? null;
+		const stoplist = rule.pageStartPrevWordStoplist;
+		if (!stoplist?.length) {
+			compiledPrevWordStoplists.set(ruleIndex, null);
+			return null;
+		}
+		const normalized = new Set(stoplist.map((word) => normalizeArabicForComparison(word)).filter(Boolean));
+		compiledPrevWordStoplists.set(ruleIndex, normalized);
+		return normalized;
+	};
+	const getSamePagePrevWordStoplist = (rule, ruleIndex) => {
+		if (compiledSamePagePrevWordStoplists.has(ruleIndex)) return compiledSamePagePrevWordStoplists.get(ruleIndex) ?? null;
+		const stoplist = rule.samePagePrevWordStoplist;
+		if (!stoplist?.length) {
+			compiledSamePagePrevWordStoplists.set(ruleIndex, null);
+			return null;
+		}
+		const normalized = new Set(stoplist.map((word) => normalizeArabicForComparison(word)).filter(Boolean));
+		compiledSamePagePrevWordStoplists.set(ruleIndex, normalized);
+		return normalized;
+	};
+	const getPreviousPageContent = (boundaryIndex) => {
+		if (boundaryIndex <= 0) return "";
+		const prevBoundary = pageMap.boundaries[boundaryIndex - 1];
+		return matchContent.slice(prevBoundary.start, prevBoundary.end);
+	};
 	const getPrevPageLastNonWsChar = (boundaryIndex) => {
 		if (boundaryIndex <= 0) return "";
 		const prevBoundary = pageMap.boundaries[boundaryIndex - 1];
@@ -3170,13 +3622,24 @@ const createPageStartGuardChecker = (matchContent, pageMap) => {
 		}
 		return "";
 	};
+	const getCurrentPageContentBeforeMatch = (matchStart) => {
+		const pageId = pageMap.getId(matchStart);
+		const boundaryIndex = pageIdToBoundaryIndex.get(pageId);
+		if (boundaryIndex === void 0) return "";
+		const boundary = pageMap.boundaries[boundaryIndex];
+		return matchContent.slice(boundary.start, matchStart);
+	};
 	return (rule, ruleIndex, matchStart) => {
 		const boundaryIndex = pageStartToBoundaryIndex.get(matchStart);
-		if (boundaryIndex === void 0 || boundaryIndex === 0) return true;
-		const prevReq = getPageStartPrevRegex(rule, ruleIndex);
-		if (!prevReq) return true;
-		const lastChar = getPrevPageLastNonWsChar(boundaryIndex);
-		return lastChar ? prevReq.test(lastChar) : false;
+		if (boundaryIndex !== void 0 && boundaryIndex !== 0) {
+			const prevReq = getPageStartPrevRegex(rule, ruleIndex);
+			if (prevReq) {
+				const lastChar = getPrevPageLastNonWsChar(boundaryIndex);
+				if (!lastChar || !prevReq.test(lastChar)) return false;
+			}
+			return shouldAllowPageStartMatch(getPreviousPageContent(boundaryIndex), getPrevWordStoplist(rule, ruleIndex));
+		}
+		return shouldAllowSamePageMatch(getCurrentPageContentBeforeMatch(matchStart), getSamePagePrevWordStoplist(rule, ruleIndex));
 	};
 };
 /**
@@ -3212,10 +3675,10 @@ const attemptFastFuzzyMatch = (matchContent, lineStart, { compiled, kind, rule,
 /**
 * Processes matches for all fast-fuzzy rules at a specific line start.
 */
-const processFastFuzzyMatchesAt = (matchContent, lineStart, pageId, fastFuzzyRules, passesPageStartGuard, isPageStart, splitPointsByRule) => {
+const processFastFuzzyMatchesAt = (matchContent, lineStart, pageId, fastFuzzyRules, passesPageStartGuard, splitPointsByRule) => {
 	for (const ffRule of fastFuzzyRules) {
 		if (!passesRuleConstraints$1(ffRule.rule, pageId)) continue;
-		if (isPageStart && !passesPageStartGuard(ffRule.rule, ffRule.ruleIndex, lineStart)) continue;
+		if (!passesPageStartGuard(ffRule.rule, ffRule.ruleIndex, lineStart)) continue;
 		attemptFastFuzzyMatch(matchContent, lineStart, ffRule, splitPointsByRule);
 	}
 };
@@ -3230,19 +3693,17 @@ const collectFastFuzzySplitPoints = (matchContent, pageMap, fastFuzzyRules, pass
 			currentBoundary = pageMap.boundaries[boundaryIdx];
 		}
 	};
-	const isPageStart = (offset) => offset === currentBoundary?.start;
 	for (let lineStart = 0; lineStart <= matchContent.length;) {
 		advanceBoundaryTo(lineStart);
 		const pageId = currentBoundary?.id ?? 0;
 		if (lineStart >= matchContent.length) break;
-		processFastFuzzyMatchesAt(matchContent, lineStart, pageId, fastFuzzyRules, passesPageStartGuard, isPageStart(lineStart), splitPointsByRule);
+		processFastFuzzyMatchesAt(matchContent, lineStart, pageId, fastFuzzyRules, passesPageStartGuard, splitPointsByRule);
 		const nextNl = matchContent.indexOf("\n", lineStart);
 		if (nextNl === -1) break;
 		lineStart = nextNl + 1;
 	}
 	return splitPointsByRule;
 };
-
 //#endregion
 //#region src/segmentation/split-point-helpers.ts
 const MAX_REGEX_ITERATIONS = 1e5;
@@ -3256,7 +3717,7 @@ const buildContentOffsets = (match, ruleInfo) => {
 	if (!ruleInfo.usesLineStartsAfter) return {};
 	const captured = match.groups?.[`${ruleInfo.prefix}__content`];
 	if (captured === void 0) return {};
-	return { contentStartOffset: (match.groups?.[ruleInfo.prefix] || match[0]).length - captured.length };
+	return { contentStartOffset: (match.groups?.[ruleInfo.prefix] ?? match[0]).length - captured.length };
 };
 const passesRuleConstraints = (rule, pageId) => (rule.min === void 0 || pageId >= rule.min) && (rule.max === void 0 || pageId <= rule.max) && !isPageExcluded(pageId, rule.exclude);
 const createSplitPointFromMatch = (match, rule, ruleInfo) => {
@@ -3271,7 +3732,32 @@ const createSplitPointFromMatch = (match, rule, ruleInfo) => {
 		wordIndex
 	};
 };
+const addSplitPoint = (splitPointsByRule, originalIndex, point) => {
+	const arr = splitPointsByRule.get(originalIndex);
+	if (!arr) {
+		splitPointsByRule.set(originalIndex, [point]);
+		return;
+	}
+	arr.push(point);
+};
+/**
+* Executes a combined regex over the content for combinable rules and records
+* any resulting split points into `splitPointsByRule`.
+*
+* This function mutates `splitPointsByRule` in place and throws if the regex
+* iteration guard is exceeded.
+*
+* @param matchContent - Concatenated content being segmented
+* @param combinableRules - Rules that can be combined into a single alternation
+* @param ruleRegexes - Compiled regex metadata aligned with `combinableRules`
+* @param pageMap - Page boundary mapping utilities for the content
+* @param passesPageStartGuard - Callback that decides whether a match is allowed
+* @param splitPointsByRule - Mutable map collecting split points by rule index
+* @param logger - Optional logger for iteration diagnostics
+* @returns Nothing; results are written into `splitPointsByRule`
+*/
 const processCombinedMatches = (matchContent, combinableRules, ruleRegexes, pageMap, passesPageStartGuard, splitPointsByRule, logger) => {
+	assertCombinedRuleAlignment(combinableRules, ruleRegexes);
 	const combinedSource = ruleRegexes.map((r) => r.source).join("|");
 	const combinedRegex = new RegExp(combinedSource, "gm");
 	logger?.debug?.("[segmenter] combined regex built", {
@@ -3286,19 +3772,29 @@ const processCombinedMatches = (matchContent, combinableRules, ruleRegexes, page
 			iterations,
 			position: m.index
 		});
-		const matchedIndex = combinableRules.findIndex(({ prefix }) => m?.groups?.[prefix] !== void 0);
-		if (matchedIndex !== -1) {
-			const { rule, index: originalIndex } = combinableRules[matchedIndex];
-			if (passesRuleConstraints(rule, pageMap.getId(m.index)) && passesPageStartGuard(rule, originalIndex, m.index)) {
-				const arr = splitPointsByRule.get(originalIndex);
-				if (!arr) splitPointsByRule.set(originalIndex, [createSplitPointFromMatch(m, rule, ruleRegexes[matchedIndex])]);
-				else arr.push(createSplitPointFromMatch(m, rule, ruleRegexes[matchedIndex]));
-			}
-		}
+		processCombinedMatch(combinableRules, ruleRegexes, pageMap, passesPageStartGuard, splitPointsByRule, m);
 		if (m[0].length === 0) combinedRegex.lastIndex++;
 		m = combinedRegex.exec(matchContent);
 	}
 };
+const assertCombinedRuleAlignment = (combinableRules, ruleRegexes) => {
+	if (combinableRules.length !== ruleRegexes.length) throw new Error(`processCombinedMatches: combinableRules/ruleRegexes length mismatch (${combinableRules.length} !== ${ruleRegexes.length})`);
+	for (let i = 0; i < combinableRules.length; i++) if (!ruleRegexes[i].source.includes(`(?<${combinableRules[i].prefix}>`)) throw new Error(`processCombinedMatches: regex alignment mismatch for prefix "${combinableRules[i].prefix}" at index ${i}`);
+};
+const processCombinedMatch = (combinableRules, ruleRegexes, pageMap, passesPageStartGuard, splitPointsByRule, match) => {
+	const matchedIndex = combinableRules.findIndex(({ prefix }) => match.groups?.[prefix] !== void 0);
+	if (matchedIndex === -1) return;
+	const { rule, index: originalIndex } = combinableRules[matchedIndex];
+	if (!passesRuleConstraints(rule, pageMap.getId(match.index)) || !passesPageStartGuard(rule, originalIndex, match.index)) return;
+	addSplitPoint(splitPointsByRule, originalIndex, createSplitPointFromMatch(match, rule, ruleRegexes[matchedIndex]));
+};
+/**
+* Builds compiled regex metadata for each combinable rule while preserving the
+* prefix used to identify the matching branch inside a combined alternation.
+*
+* @param combinableRules - Rules eligible for combined-regex processing
+* @returns Rule regex metadata aligned with the input order
+*/
 const buildRuleRegexes = (combinableRules) => combinableRules.map(({ rule, prefix }) => {
 	const built = buildRuleRegex(rule, prefix);
 	return {
@@ -3307,6 +3803,18 @@ const buildRuleRegexes = (combinableRules) => combinableRules.map(({ rule, prefi
 		source: `(?<${prefix}>${built.regex.source})`
 	};
 });
+/**
+* Processes a standalone rule by matching it independently and appending its
+* resulting split points into `splitPointsByRule`.
+*
+* @param rule - The standalone split rule to evaluate
+* @param ruleIndex - Original rule index in the caller's rules array
+* @param matchContent - Concatenated content being segmented
+* @param pageMap - Page boundary mapping utilities for the content
+* @param passesPageStartGuard - Callback that decides whether a match is allowed
+* @param splitPointsByRule - Mutable map collecting split points by rule index
+* @returns Nothing; results are written into `splitPointsByRule`
+*/
 const processStandaloneRule = (rule, ruleIndex, matchContent, pageMap, passesPageStartGuard, splitPointsByRule) => {
 	const { regex, usesCapture, captureNames, usesLineStartsAfter } = buildRuleRegex(rule);
 	const points = filterByConstraints(findMatchesInContent(matchContent, regex, usesCapture, captureNames), rule, pageMap.getId).filter((m) => passesPageStartGuard(rule, ruleIndex, m.start)).map((m) => {
@@ -3341,6 +3849,15 @@ const findMatchesInContent = (content, regex, usesCapture, captureNames) => {
 	}
 	return matches;
 };
+/**
+* Applies per-rule occurrence filtering and optional debug metadata patches to
+* the collected split points.
+*
+* @param rules - Full rule list in original order
+* @param splitPointsByRule - Split points grouped by originating rule index
+* @param debugMetaKey - Optional metadata key used for debug provenance patches
+* @returns Flattened split points after occurrence filtering and debug merging
+*/
 const applyOccurrenceFilter = (rules, splitPointsByRule, debugMetaKey) => {
 	const result = [];
 	rules.forEach((rule, index) => {
@@ -3358,7 +3875,6 @@ const applyOccurrenceFilter = (rules, splitPointsByRule, debugMetaKey) => {
 	});
 	return result;
 };
-
 //#endregion
 //#region src/segmentation/segmenter.ts
 /**
@@ -3432,10 +3948,30 @@ const dedupeSplitPoints = (splitPoints) => {
 	const byIndex = /* @__PURE__ */ new Map();
 	for (const p of splitPoints) {
 		const existing = byIndex.get(p.index);
-		if (!existing || p.contentStartOffset !== void 0 && existing.contentStartOffset === void 0 || p.meta !== void 0 && existing.meta === void 0) byIndex.set(p.index, p);
+		if (!existing) {
+			byIndex.set(p.index, p);
+			continue;
+		}
+		byIndex.set(p.index, mergeSplitPoints(existing, p));
 	}
 	return [...byIndex.values()].sort((a, b) => a.index - b.index);
 };
+const prefersIncomingSplitPoint = (existing, incoming) => incoming.contentStartOffset !== void 0 && existing.contentStartOffset === void 0 || incoming.meta !== void 0 && existing.meta === void 0;
+const mergeRecord = (existing, incoming) => existing || incoming ? {
+	...existing ?? {},
+	...incoming ?? {}
+} : void 0;
+const mergeSplitPoints = (existing, incoming) => {
+	const preferred = prefersIncomingSplitPoint(existing, incoming) ? incoming : existing;
+	const fallback = preferred === incoming ? existing : incoming;
+	return {
+		...fallback,
+		...preferred,
+		contentStartOffset: preferred.contentStartOffset ?? fallback.contentStartOffset,
+		meta: mergeRecord(existing.meta, incoming.meta),
+		namedCaptures: mergeRecord(existing.namedCaptures, incoming.namedCaptures)
+	};
+};
 /**
 * If no structural rules produced segments, create a single segment spanning all pages.
 * This allows breakpoint processing to still run.
@@ -3468,7 +4004,7 @@ const collectSplitPointsFromRules = (rules, matchContent, pageMap, debugMetaKey,
 	});
 	const splitPointsByRule = collectFastFuzzySplitPoints(matchContent, pageMap, fastFuzzyRules, passesPageStartGuard);
 	if (combinableRules.length > 0) processCombinedMatches(matchContent, combinableRules, buildRuleRegexes(combinableRules), pageMap, passesPageStartGuard, splitPointsByRule, logger);
-	for (const rule of standaloneRules) processStandaloneRule(rule, rules.indexOf(rule), matchContent, pageMap, passesPageStartGuard, splitPointsByRule);
+	for (const { rule, index } of standaloneRules) processStandaloneRule(rule, index, matchContent, pageMap, passesPageStartGuard, splitPointsByRule);
 	return applyOccurrenceFilter(rules, splitPointsByRule, debugMetaKey);
 };
 /**
@@ -3508,7 +4044,7 @@ const findBreaksInRange = (startOffset, endOffset, sortedBreaks) => {
 * @returns Content with page-break newlines converted to spaces (or left as-is for `newline`)
 */
 const convertPageBreaks = (content, startOffset, pageBreaks, pageJoiner) => {
-	if (!content || !content.includes("\n")) return content;
+	if (!content?.includes("\n")) return content;
 	if (pageJoiner === "newline") return content;
 	const breaksInRange = findBreaksInRange(startOffset, startOffset + content.length, pageBreaks);
 	if (breaksInRange.length === 0) return content;
@@ -3616,16 +4152,23 @@ const segmentPages = (pages, options) => {
 * @returns Array of segment objects
 */
 const buildSegments = (splitPoints, content, pageMap, rules, pageJoiner) => {
+	const getActualStart = (start, contentStartOffset) => start + (contentStartOffset ?? 0);
+	const trimSegmentText = (sliced, capturedContent, contentStartOffset) => capturedContent?.trim() ?? (contentStartOffset ? sliced.trim() : sliced.replace(/[\s\n]+$/, ""));
+	const getAdjustedStart = (actualStart, sliced, contentStartOffset) => actualStart + (contentStartOffset ? sliced.length - sliced.trimStart().length : 0);
+	const applyMeta = (meta, namedCaptures) => meta || namedCaptures ? {
+		...meta,
+		...namedCaptures
+	} : void 0;
 	/**
 	* Creates a single segment from a content range.
 	*/
 	const createSegment = (start, end, meta, capturedContent, namedCaptures, contentStartOffset) => {
-		const actualStart = start + (contentStartOffset ?? 0);
+		const actualStart = getActualStart(start, contentStartOffset);
 		const sliced = content.slice(actualStart, end);
-		let text = capturedContent?.trim() ?? (contentStartOffset ? sliced.trim() : sliced.replace(/[\s\n]+$/, ""));
+		let text = trimSegmentText(sliced, capturedContent, contentStartOffset);
 		if (!text) return null;
 		if (!capturedContent) text = convertPageBreaks(text, actualStart, pageMap.pageBreaks, pageJoiner);
-		const adjustedStart = actualStart + (contentStartOffset ? sliced.length - sliced.trimStart().length : 0);
+		const adjustedStart = getAdjustedStart(actualStart, sliced, contentStartOffset);
 		const from = pageMap.getId(adjustedStart);
 		const to = capturedContent ? pageMap.getId(end - 1) : pageMap.getId(adjustedStart + text.length - 1);
 		const seg = {
@@ -3633,10 +4176,8 @@ const buildSegments = (splitPoints, content, pageMap, rules, pageJoiner) => {
 			from
 		};
 		if (to !== from) seg.to = to;
-		if (meta || namedCaptures) seg.meta = {
-			...meta,
-			...namedCaptures
-		};
+		const mergedMeta = applyMeta(meta, namedCaptures);
+		if (mergedMeta) seg.meta = mergedMeta;
 		return seg;
 	};
 	/**
@@ -3668,659 +4209,6 @@ const buildSegments = (splitPoints, content, pageMap, rules, pageJoiner) => {
 	}
 	return [...segments, ...createSegmentsFromSplitPoints()];
 };
-
-//#endregion
-//#region src/recovery.ts
-const preview = (s, max = 40) => s.length <= max ? s : `${s.slice(0, max)}…`;
-const normalizeForCompare = (s, mode) => {
-	if (mode === "none") return s;
-	let out = s;
-	if (mode === "whitespace_and_nfkc") out = out.normalize("NFKC").replace(/(?:\u200C|\u200D|\uFEFF)/gu, "");
-	out = out.replace(/\r\n?/gu, "\n").replace(/\s+/gu, " ").trim();
-	return out;
-};
-const segmentRangeKey = (s) => `${s.from}|${s.to ?? s.from}`;
-const buildFixedOptions = (options, selectedRuleIndices) => {
-	const fixedRules = (options.rules ?? []).map((r, idx) => {
-		if (!selectedRuleIndices.has(idx)) return r;
-		if (!("lineStartsAfter" in r) || !r.lineStartsAfter) return r;
-		const { lineStartsAfter, ...rest } = r;
-		return {
-			...rest,
-			lineStartsWith: lineStartsAfter
-		};
-	});
-	return {
-		...options,
-		rules: fixedRules
-	};
-};
-const buildPageIdToIndex = (pages) => new Map(pages.map((p, i) => [p.id, i]));
-const buildRangeContent = (processedPages, fromIdx, toIdx, pageJoiner) => {
-	const parts = [];
-	for (let i = fromIdx; i <= toIdx; i++) parts.push(normalizeLineEndings(processedPages[i].content));
-	const matchContent = parts.join("\n");
-	if (pageJoiner === "newline") return {
-		matchContent,
-		outputContent: matchContent
-	};
-	return {
-		matchContent,
-		outputContent: parts.join(" ")
-	};
-};
-const compileMistakenRulesAsStartsWith = (options, selectedRuleIndices) => {
-	const rules = options.rules ?? [];
-	const compiled = [];
-	for (const idx of selectedRuleIndices) {
-		const r = rules[idx];
-		if (!r || !("lineStartsAfter" in r) || !r.lineStartsAfter?.length) continue;
-		const { lineStartsAfter, ...rest } = r;
-		const built = buildRuleRegex({
-			...rest,
-			lineStartsWith: lineStartsAfter
-		});
-		compiled.push({
-			ruleIndex: idx,
-			startsWithRegex: new RegExp(built.regex.source, "mu")
-		});
-	}
-	return compiled;
-};
-const findUniqueAnchorPos = (outputContent, segmentContent) => {
-	for (const len of [
-		80,
-		60,
-		40,
-		30,
-		20,
-		15
-	]) {
-		const needle = segmentContent.slice(0, Math.min(len, segmentContent.length));
-		if (!needle.trim()) continue;
-		const first = outputContent.indexOf(needle);
-		if (first === -1) continue;
-		if (outputContent.indexOf(needle, first + 1) === -1) return first;
-	}
-	return null;
-};
-const findRecoveredPrefixAtLineStart = (segmentContent, matchContent, lineStart, anchorPos, compiledMistaken) => {
-	const line = matchContent.slice(lineStart);
-	for (const mr of compiledMistaken) {
-		mr.startsWithRegex.lastIndex = 0;
-		const m = mr.startsWithRegex.exec(line);
-		if (!m || m.index !== 0) continue;
-		const markerMatch = m[0];
-		const markerEnd = lineStart + markerMatch.length;
-		if (anchorPos < markerEnd) continue;
-		const gap = matchContent.slice(markerEnd, anchorPos);
-		const recoveredPrefix = /^\s*$/u.test(gap) ? `${markerMatch}${gap}` : markerMatch;
-		if (segmentContent.startsWith(markerMatch) || segmentContent.startsWith(recoveredPrefix)) return { reason: "content already starts with selected marker" };
-		return { prefix: recoveredPrefix };
-	}
-	return { reason: "no selected marker pattern matched at anchored line start" };
-};
-const tryBestEffortRecoverOneSegment = (segment, processedPages, pageIdToIndex, compiledMistaken, pageJoiner) => {
-	const fromIdx = pageIdToIndex.get(segment.from);
-	const toIdx = pageIdToIndex.get(segment.to ?? segment.from) ?? fromIdx;
-	if (fromIdx === void 0 || toIdx === void 0 || fromIdx < 0 || toIdx < fromIdx) return {
-		kind: "unresolved",
-		reason: "segment page range not found in pages"
-	};
-	const { matchContent, outputContent } = buildRangeContent(processedPages, fromIdx, toIdx, pageJoiner);
-	if (!segment.content) return {
-		kind: "unresolved",
-		reason: "empty segment content"
-	};
-	const anchorPos = findUniqueAnchorPos(outputContent, segment.content);
-	if (anchorPos === null) return {
-		kind: "unresolved",
-		reason: "could not uniquely anchor segment content in page range"
-	};
-	const lineStart = matchContent.lastIndexOf("\n", Math.max(0, anchorPos - 1)) + 1;
-	const found = findRecoveredPrefixAtLineStart(segment.content, matchContent, lineStart, anchorPos, compiledMistaken);
-	if ("reason" in found) return found.reason.includes("already starts") ? { kind: "skipped_idempotent" } : {
-		kind: "unresolved",
-		reason: found.reason
-	};
-	return {
-		kind: "recovered",
-		recoveredContent: `${found.prefix}${segment.content}`,
-		recoveredPrefix: found.prefix
-	};
-};
-const resolveRuleIndicesSelector = (rules, indicesIn) => {
-	const errors = [];
-	const indices = /* @__PURE__ */ new Set();
-	for (const idx of indicesIn) {
-		if (!Number.isInteger(idx) || idx < 0 || idx >= rules.length) {
-			errors.push(`Selector index out of range: ${idx}`);
-			continue;
-		}
-		const rule = rules[idx];
-		if (!rule || !("lineStartsAfter" in rule)) {
-			errors.push(`Selector index ${idx} is not a lineStartsAfter rule`);
-			continue;
-		}
-		indices.add(idx);
-	}
-	return {
-		errors,
-		indices,
-		warnings: []
-	};
-};
-const resolvePredicateSelector = (rules, predicate) => {
-	const errors = [];
-	const warnings = [];
-	const indices = /* @__PURE__ */ new Set();
-	rules.forEach((r, i) => {
-		try {
-			if (!predicate(r, i)) return;
-			if ("lineStartsAfter" in r && r.lineStartsAfter?.length) {
-				indices.add(i);
-				return;
-			}
-			warnings.push(`Predicate selected rule ${i}, but it is not a lineStartsAfter rule; skipping`);
-		} catch (e) {
-			const msg = e instanceof Error ? e.message : String(e);
-			errors.push(`Predicate threw at rule ${i}: ${msg}`);
-		}
-	});
-	if (indices.size === 0) warnings.push("Predicate did not select any lineStartsAfter rules");
-	return {
-		errors,
-		indices,
-		warnings
-	};
-};
-const resolvePatternsSelector = (rules, patterns, matchMode) => {
-	const errors = [];
-	const warnings = [];
-	const indices = /* @__PURE__ */ new Set();
-	const normalizePattern = (p) => normalizeForCompare(p, (matchMode ?? "exact") === "normalized" ? "whitespace_and_nfkc" : "none");
-	const targets = patterns.map(normalizePattern);
-	for (let pi = 0; pi < patterns.length; pi++) {
-		const rawPattern = patterns[pi];
-		const pat = targets[pi];
-		const matched = [];
-		for (let i = 0; i < rules.length; i++) {
-			const r = rules[i];
-			if (!("lineStartsAfter" in r) || !r.lineStartsAfter?.length) continue;
-			if (r.lineStartsAfter.some((rp) => normalizePattern(rp) === pat)) matched.push(i);
-		}
-		if (matched.length === 0) {
-			errors.push(`Pattern "${rawPattern}" did not match any lineStartsAfter rule`);
-			continue;
-		}
-		if (matched.length > 1) warnings.push(`Pattern "${rawPattern}" matched multiple lineStartsAfter rules: [${matched.join(", ")}]`);
-		matched.forEach((i) => {
-			indices.add(i);
-		});
-	}
-	return {
-		errors,
-		indices,
-		warnings
-	};
-};
-const resolveSelectorToRuleIndices = (options, selector) => {
-	const rules = options.rules ?? [];
-	if (selector.type === "rule_indices") return resolveRuleIndicesSelector(rules, selector.indices);
-	if (selector.type === "predicate") return resolvePredicateSelector(rules, selector.predicate);
-	return resolvePatternsSelector(rules, selector.patterns, selector.match);
-};
-const longestCommonSuffixLength = (a, b) => {
-	const max = Math.min(a.length, b.length);
-	let i = 0;
-	while (i < max) {
-		if (a[a.length - 1 - i] !== b[b.length - 1 - i]) break;
-		i++;
-	}
-	return i;
-};
-const AMBIGUITY_SCORE_GAP = 5;
-const scoreCandidate = (orig, fixed, normalizeMode) => {
-	if (fixed.content === orig.content) return {
-		fixedIndex: -1,
-		kind: "exact",
-		score: 100
-	};
-	if (fixed.content.endsWith(orig.content)) {
-		const markerLen = fixed.content.length - orig.content.length;
-		return {
-			fixedIndex: -1,
-			kind: "exact_suffix",
-			score: 90 + Math.min(30, markerLen)
-		};
-	}
-	if (normalizeMode !== "none") {
-		const normFixed = normalizeForCompare(fixed.content, normalizeMode);
-		const normOrig = normalizeForCompare(orig.content, normalizeMode);
-		if (normFixed.endsWith(normOrig) && normOrig.length > 0) {
-			const overlap = longestCommonSuffixLength(normFixed, normOrig) / normOrig.length;
-			return {
-				fixedIndex: -1,
-				kind: "normalized_suffix",
-				score: 70 + Math.floor(overlap * 20)
-			};
-		}
-	}
-	return null;
-};
-const buildNoSelectionResult = (segments, reportBase, mode, selectorErrors) => {
-	const warnings = [...reportBase.warnings];
-	warnings.push("No lineStartsAfter rules selected for recovery; returning segments unchanged");
-	const details = segments.map((s, i) => {
-		const status = selectorErrors.length ? "unresolved_selector" : "unchanged";
-		return {
-			from: s.from,
-			notes: selectorErrors.length ? ["selector did not resolve"] : void 0,
-			originalStartPreview: preview(s.content),
-			segmentIndex: i,
-			status,
-			strategy: "none",
-			to: s.to
-		};
-	});
-	return {
-		report: {
-			...reportBase,
-			details,
-			summary: {
-				mode,
-				recovered: 0,
-				totalSegments: segments.length,
-				unchanged: segments.length,
-				unresolved: selectorErrors.length ? segments.length : 0
-			},
-			warnings
-		},
-		segments
-	};
-};
-const runStage1IfEnabled = (pages, segments, options, selectedRuleIndices, mode) => {
-	const recoveredAtIndex = /* @__PURE__ */ new Map();
-	const recoveredDetailAtIndex = /* @__PURE__ */ new Map();
-	if (mode !== "best_effort_then_rerun") return {
-		recoveredAtIndex,
-		recoveredDetailAtIndex
-	};
-	const pageIdToIndex = buildPageIdToIndex(pages);
-	const pageJoiner = options.pageJoiner ?? "space";
-	const compiledMistaken = compileMistakenRulesAsStartsWith(options, selectedRuleIndices);
-	for (let i = 0; i < segments.length; i++) {
-		const orig = segments[i];
-		const r = tryBestEffortRecoverOneSegment(orig, pages, pageIdToIndex, compiledMistaken, pageJoiner);
-		if (r.kind !== "recovered") continue;
-		const seg = {
-			...orig,
-			content: r.recoveredContent
-		};
-		recoveredAtIndex.set(i, seg);
-		recoveredDetailAtIndex.set(i, {
-			from: orig.from,
-			originalStartPreview: preview(orig.content),
-			recoveredPrefixPreview: preview(r.recoveredPrefix),
-			recoveredStartPreview: preview(seg.content),
-			segmentIndex: i,
-			status: "recovered",
-			strategy: "stage1",
-			to: orig.to
-		});
-	}
-	return {
-		recoveredAtIndex,
-		recoveredDetailAtIndex
-	};
-};
-const buildFixedBuckets = (fixedSegments) => {
-	const buckets = /* @__PURE__ */ new Map();
-	for (let i = 0; i < fixedSegments.length; i++) {
-		const k = segmentRangeKey(fixedSegments[i]);
-		const arr = buckets.get(k);
-		if (!arr) buckets.set(k, [i]);
-		else arr.push(i);
-	}
-	return buckets;
-};
-const findBestFixedMatch = (orig, candidates, fixedSegments, usedFixed, normalizeCompare) => {
-	let best = null;
-	let secondBestScore = -Infinity;
-	for (const fixedIdx of candidates) {
-		if (usedFixed.has(fixedIdx)) continue;
-		const fixed = fixedSegments[fixedIdx];
-		const scored = scoreCandidate(orig, fixed, normalizeCompare);
-		if (!scored) continue;
-		const candidateScore = scored.score;
-		if (!best || candidateScore > best.score) {
-			secondBestScore = best?.score ?? -Infinity;
-			best = {
-				fixedIdx,
-				score: candidateScore
-			};
-		} else if (candidateScore > secondBestScore) secondBestScore = candidateScore;
-	}
-	if (!best) return { kind: "none" };
-	if (best.score - secondBestScore < AMBIGUITY_SCORE_GAP && candidates.length > 1) return { kind: "ambiguous" };
-	return {
-		fixedIdx: best.fixedIdx,
-		kind: "match"
-	};
-};
-const detailUnresolved = (orig, segmentIndex, notes) => ({
-	from: orig.from,
-	notes,
-	originalStartPreview: preview(orig.content),
-	segmentIndex,
-	status: "unresolved_alignment",
-	strategy: "rerun",
-	to: orig.to
-});
-const detailSkippedIdempotent = (orig, segmentIndex, notes) => ({
-	from: orig.from,
-	notes,
-	originalStartPreview: preview(orig.content),
-	segmentIndex,
-	status: "skipped_idempotent",
-	strategy: "rerun",
-	to: orig.to
-});
-const detailRecoveredRerun = (orig, fixed, segmentIndex) => {
-	let recoveredPrefixPreview;
-	if (fixed.content.endsWith(orig.content)) recoveredPrefixPreview = preview(fixed.content.slice(0, fixed.content.length - orig.content.length));
-	return {
-		from: orig.from,
-		originalStartPreview: preview(orig.content),
-		recoveredPrefixPreview,
-		recoveredStartPreview: preview(fixed.content),
-		segmentIndex,
-		status: "recovered",
-		strategy: "rerun",
-		to: orig.to
-	};
-};
-const mergeWithRerun = (params) => {
-	const { fixedBuckets, fixedSegments, normalizeCompare, originalSegments, stage1RecoveredAtIndex, recoveredDetailAtIndex } = params;
-	const usedFixed = /* @__PURE__ */ new Set();
-	const out = [];
-	const details = [];
-	let recovered = 0;
-	let unresolved = 0;
-	let unchanged = 0;
-	for (let i = 0; i < originalSegments.length; i++) {
-		const stage1Recovered = stage1RecoveredAtIndex.get(i);
-		if (stage1Recovered) {
-			out.push(stage1Recovered);
-			recovered++;
-			details.push(recoveredDetailAtIndex.get(i) ?? {
-				from: stage1Recovered.from,
-				originalStartPreview: preview(originalSegments[i].content),
-				recoveredStartPreview: preview(stage1Recovered.content),
-				segmentIndex: i,
-				status: "recovered",
-				strategy: "stage1",
-				to: stage1Recovered.to
-			});
-			continue;
-		}
-		const orig = originalSegments[i];
-		const best = findBestFixedMatch(orig, fixedBuckets.get(segmentRangeKey(orig)) ?? [], fixedSegments, usedFixed, normalizeCompare);
-		if (best.kind === "none") {
-			out.push(orig);
-			unresolved++;
-			details.push(detailUnresolved(orig, i, ["no alignment candidate in rerun output for same (from,to)"]));
-			continue;
-		}
-		if (best.kind === "ambiguous") {
-			out.push(orig);
-			unresolved++;
-			details.push(detailUnresolved(orig, i, ["ambiguous alignment (score gap too small)"]));
-			continue;
-		}
-		usedFixed.add(best.fixedIdx);
-		const fixed = fixedSegments[best.fixedIdx];
-		if (fixed.content === orig.content) {
-			out.push(orig);
-			unchanged++;
-			details.push(detailSkippedIdempotent(orig, i, ["content already matches rerun output"]));
-			continue;
-		}
-		out.push({
-			...orig,
-			content: fixed.content
-		});
-		recovered++;
-		details.push(detailRecoveredRerun(orig, fixed, i));
-	}
-	return {
-		details,
-		segments: out,
-		summary: {
-			recovered,
-			unchanged,
-			unresolved
-		}
-	};
-};
-function recoverMistakenLineStartsAfterMarkers(pages, segments, options, selector, opts) {
-	const mode = opts?.mode ?? "rerun_only";
-	const normalizeCompare = opts?.normalizeCompare ?? "whitespace";
-	const resolved = resolveSelectorToRuleIndices(options, selector);
-	const reportBase = {
-		byRun: void 0,
-		errors: resolved.errors,
-		warnings: resolved.warnings
-	};
-	if (resolved.indices.size === 0) return buildNoSelectionResult(segments, reportBase, mode, resolved.errors);
-	const stage1 = runStage1IfEnabled(pages, segments, options, resolved.indices, mode);
-	const fixedSegments = segmentPages(pages, buildFixedOptions(options, resolved.indices));
-	const merged = mergeWithRerun({
-		fixedBuckets: buildFixedBuckets(fixedSegments),
-		fixedSegments,
-		normalizeCompare,
-		originalSegments: segments,
-		recoveredDetailAtIndex: stage1.recoveredDetailAtIndex,
-		stage1RecoveredAtIndex: stage1.recoveredAtIndex
-	});
-	return {
-		report: {
-			...reportBase,
-			details: merged.details,
-			summary: {
-				mode,
-				recovered: merged.summary.recovered,
-				totalSegments: segments.length,
-				unchanged: merged.summary.unchanged,
-				unresolved: merged.summary.unresolved
-			}
-		},
-		segments: merged.segments
-	};
-}
-function recoverMistakenMarkersForRuns(runs, opts) {
-	const allSegments = [];
-	const byRun = [];
-	const details = [];
-	const warnings = [];
-	const errors = [];
-	let recovered = 0;
-	let unchanged = 0;
-	let unresolved = 0;
-	let offset = 0;
-	for (let i = 0; i < runs.length; i++) {
-		const run = runs[i];
-		const res = recoverMistakenLineStartsAfterMarkers(run.pages, run.segments, run.options, run.selector, opts);
-		allSegments.push(...res.segments);
-		for (const d of res.report.details) details.push({
-			...d,
-			segmentIndex: d.segmentIndex + offset
-		});
-		offset += run.segments.length;
-		recovered += res.report.summary.recovered;
-		unchanged += res.report.summary.unchanged;
-		unresolved += res.report.summary.unresolved;
-		warnings.push(...res.report.warnings);
-		errors.push(...res.report.errors);
-		byRun.push({
-			recovered: res.report.summary.recovered,
-			runIndex: i,
-			totalSegments: run.segments.length,
-			unresolved: res.report.summary.unresolved
-		});
-	}
-	return {
-		report: {
-			byRun,
-			details,
-			errors,
-			summary: {
-				mode: opts?.mode ?? "rerun_only",
-				recovered,
-				totalSegments: offset,
-				unchanged,
-				unresolved
-			},
-			warnings
-		},
-		segments: allSegments
-	};
-}
-
-//#endregion
-//#region src/segmentation/pattern-validator.ts
-const KNOWN_TOKENS = new Set(getAvailableTokens());
-const TOKEN_INSIDE_BRACES = /\{\{(\w+)(?::\w+)?\}\}/g;
-const buildBareTokenRegex = () => {
-	const tokens = [...KNOWN_TOKENS].sort((a, b) => b.length - a.length);
-	return new RegExp(`(?<!\\{\\{)(${tokens.join("|")})(?::\\w+)?(?!\\}\\})`, "g");
-};
-/**
-* Validates a single pattern for common issues.
-*/
-const validatePattern = (pattern, seenPatterns) => {
-	if (!pattern.trim()) return {
-		message: "Empty pattern is not allowed",
-		type: "empty_pattern"
-	};
-	if (seenPatterns.has(pattern)) return {
-		message: `Duplicate pattern: "${pattern}"`,
-		pattern,
-		type: "duplicate"
-	};
-	seenPatterns.add(pattern);
-	TOKEN_INSIDE_BRACES.lastIndex = 0;
-	for (const match of pattern.matchAll(TOKEN_INSIDE_BRACES)) {
-		const name = match[1];
-		if (!KNOWN_TOKENS.has(name)) return {
-			message: `Unknown token: {{${name}}}. Available tokens: ${[...KNOWN_TOKENS].slice(0, 5).join(", ")}...`,
-			suggestion: "Check spelling or use a known token",
-			token: name,
-			type: "unknown_token"
-		};
-	}
-	for (const match of pattern.matchAll(buildBareTokenRegex())) {
-		const [full, name] = match;
-		const idx = match.index;
-		if (pattern.slice(Math.max(0, idx - 2), idx) !== "{{" || pattern.slice(idx + full.length, idx + full.length + 2) !== "}}") return {
-			message: `Token "${name}" appears to be missing {{}}. Did you mean "{{${full}}}"?`,
-			suggestion: `{{${full}}}`,
-			token: name,
-			type: "missing_braces"
-		};
-	}
-};
-/**
-* Validates an array of patterns, returning parallel array of issues.
-*/
-const validatePatternArray = (patterns) => {
-	const seen = /* @__PURE__ */ new Set();
-	const issues = patterns.map((p) => validatePattern(p, seen));
-	return issues.some(Boolean) ? issues : void 0;
-};
-/**
-* Validates split rules for common pattern issues.
-*
-* Checks for:
-* - Missing `{{}}` around known token names (e.g., `raqms:num` instead of `{{raqms:num}}`)
-* - Unknown token names inside `{{}}` (e.g., `{{nonexistent}}`)
-* - Duplicate patterns within the same rule
-*
-* @param rules - Array of split rules to validate
-* @returns Array parallel to input with validation results (undefined if no issues)
-*
-* @example
-* const issues = validateRules([
-*   { lineStartsAfter: ['raqms:num'] },  // Missing braces
-*   { lineStartsWith: ['{{unknown}}'] }, // Unknown token
-* ]);
-* // issues[0]?.lineStartsAfter?.[0]?.type === 'missing_braces'
-* // issues[1]?.lineStartsWith?.[0]?.type === 'unknown_token'
-*/
-const validateRules = (rules) => rules.map((rule) => {
-	const result = {};
-	let hasIssues = false;
-	for (const key of [
-		"lineStartsWith",
-		"lineStartsAfter",
-		"lineEndsWith"
-	]) if (key in rule && rule[key]) {
-		const issues = validatePatternArray(rule[key]);
-		if (issues) {
-			result[key] = issues;
-			hasIssues = true;
-		}
-	}
-	if ("template" in rule && rule.template !== void 0) {
-		const issue = validatePattern(rule.template, /* @__PURE__ */ new Set());
-		if (issue) {
-			result.template = issue;
-			hasIssues = true;
-		}
-	}
-	return hasIssues ? result : void 0;
-});
-/**
-* Formats a validation result array into a list of human-readable error messages.
-*
-* Useful for displaying validation errors in UIs.
-*
-* @param results - The result array from `validateRules()`
-* @returns Array of formatted error strings
-*
-* @example
-* const issues = validateRules(rules);
-* const errors = formatValidationReport(issues);
-* // ["Rule 1, lineStartsWith: Missing {{}} around token..."]
-*/
-const formatValidationReport = (results) => results.flatMap((result, i) => {
-	if (!result) return [];
-	return Object.entries(result).flatMap(([type, issues]) => (Array.isArray(issues) ? issues : [issues]).map((issue) => {
-		if (!issue) return null;
-		const loc = `Rule ${i + 1}, ${type}`;
-		if (issue.type === "missing_braces") return `${loc}: Missing {{}} around token "${issue.token}"`;
-		if (issue.type === "unknown_token") return `${loc}: Unknown token "{{${issue.token}}}"`;
-		if (issue.type === "duplicate") return `${loc}: Duplicate pattern "${issue.pattern}"`;
-		return `${loc}: ${issue.message || issue.type}`;
-	})).filter((msg) => msg !== null);
-});
-
-//#endregion
-//#region src/validation/validation-constants.ts
-/**
-* Validation-specific constants
-*/
-/**
-* Limit for validation issue preview length (characters).
-*/
-const PREVIEW_LIMIT = 140;
-/**
-* Threshold for short segment content (characters).
-* Segments shorter than this will trigger a full-document search fallback
-* if not found in the expected window.
-*/
-const FULL_SEARCH_THRESHOLD = 500;
-
 //#endregion
 //#region src/validation/validate-segments.ts
 /**
@@ -4329,8 +4217,8 @@ const FULL_SEARCH_THRESHOLD = 500;
 */
 const buildPreview = (text) => {
 	const normalized = text.replace(/\s+/g, " ").trim();
-	if (normalized.length <= PREVIEW_LIMIT) return normalized;
-	return `${normalized.slice(0, PREVIEW_LIMIT)}...`;
+	if (normalized.length <= 140) return normalized;
+	return `${normalized.slice(0, 140)}...`;
 };
 /**
 * Creates a lightweight snapshot of a segment for inclusion in validation checks.
@@ -4358,19 +4246,18 @@ const normalizePages = (pages, options) => {
 */
 const buildJoinedContent = (pages, joiner) => {
 	const boundaries = [];
-	const nonEmptyPages = pages.filter((p) => p.content);
-	const joined = nonEmptyPages.map((p) => p.content).join(joiner);
+	const joined = pages.map((p) => p.content).join(joiner);
 	let offset = 0;
-	for (let i = 0; i < nonEmptyPages.length; i++) {
-		const content = nonEmptyPages[i].content;
+	for (let i = 0; i < pages.length; i++) {
+		const content = pages[i].content;
 		const start = offset;
-		const end = start + content.length - 1;
+		const end = start + content.length;
 		boundaries.push({
 			end,
-			id: nonEmptyPages[i].id,
+			id: pages[i].id,
 			start
 		});
-		offset = end + 1 + (i < nonEmptyPages.length - 1 ? joiner.length : 0);
+		offset += content.length + (i < pages.length - 1 ? joiner.length : 0);
 	}
 	return {
 		boundaries,
@@ -4561,7 +4448,7 @@ const handleFallbackSearch = (segment, segmentIndex, joined, searchStart, search
 	const bufferSize = 1e3;
 	const rawMatches = findJoinedMatches(content, joined, Math.max(0, searchStart - bufferSize), Math.min(joined.length, searchEnd + bufferSize), 5);
 	if (rawMatches.length === 0) {
-		const threshold = validationOptions?.fullSearchThreshold ?? FULL_SEARCH_THRESHOLD;
+		const threshold = validationOptions?.fullSearchThreshold ?? 500;
 		if (content.length < threshold) {
 			const fullMatches = findJoinedMatches(content, joined, 0, joined.length, 50);
 			const validMatch = fullMatches.find((m) => {
@@ -4715,7 +4602,7 @@ const validateSegments = (pages, options, segments, validationOptions) => {
 		}
 	};
 };
-
 //#endregion
-export { PATTERN_TYPE_KEYS, TOKEN_PATTERNS, Token, analyzeCommonLineStarts, analyzeRepeatingSequences, analyzeTextForRule, applyPreprocessToPage, applyTokenMappings, condenseEllipsis, containsTokens, detectTokenPatterns, escapeRegex, escapeTemplateBrackets, escapeWordsOutsideTokens, expandCompositeTokensInTemplate, expandTokens, expandTokensWithCaptures, fixTrailingWaw, formatValidationReport, generateTemplateFromText, getAvailableTokens, getDebugReason, getSegmentDebugReason, getTokenPattern, makeDiacriticInsensitive, optimizeRules, recoverMistakenLineStartsAfterMarkers, recoverMistakenMarkersForRuns, removeZeroWidth, segmentPages, shouldDefaultToFuzzy, stripTokenMappings, suggestPatternConfig, templateToRegex, validateRules, validateSegments, withCapture };
+export { ARABIC_BASE_LETTER_CLASS, ARABIC_LETTER_WITH_OPTIONAL_MARKS_PATTERN, ARABIC_MARKS_CLASS, ARABIC_WORD_WITH_OPTIONAL_MARKS_PATTERN, PATTERN_TYPE_KEYS, TOKEN_PATTERNS, Token, analyzeCommonLineStarts, analyzeRepeatingSequences, analyzeTextForRule, applyPreprocessToPage, applyTokenMappings, condenseEllipsis, containsTokens, createArabicDictionaryEntryRule, detectTokenPatterns, escapeRegex, escapeTemplateBrackets, escapeWordsOutsideTokens, expandCompositeTokensInTemplate, expandTokens, expandTokensWithCaptures, fixTrailingWaw, formatValidationReport, generateTemplateFromText, getAvailableTokens, getDebugReason, getSegmentDebugReason, getTokenPattern, makeDiacriticInsensitive, normalizeArabicForComparison, optimizeRules, removeZeroWidth, segmentPages, shouldDefaultToFuzzy, stripTokenMappings, suggestPatternConfig, templateToRegex, validateRules, validateSegments, withCapture };
+
 //# sourceMappingURL=index.mjs.map