flappa-doormal 2.6.3 → 2.7.0

package/dist/index.mjs

@@ -156,1298 +156,1442 @@ const makeDiacriticInsensitive = (text) => {
 };
 
 //#endregion
-//#region src/segmentation/breakpoint-utils.ts
-const WINDOW_PREFIX_LENGTHS = [
-	80,
-	60,
-	40,
-	30,
-	20,
-	15
-];
-const JOINER_PREFIX_LENGTHS = [
-	80,
-	60,
-	40,
-	30,
-	20,
-	15,
-	12,
-	10,
-	8,
-	6
-];
+//#region src/segmentation/tokens.ts
 /**
-* Normalizes a breakpoint to the object form.
-* Strings are converted to { pattern: str } with no constraints.
+* Token-based template system for Arabic text pattern matching.
 *
-* @param bp - Breakpoint as string or object
-* @returns Normalized BreakpointRule object
+* This module provides a human-readable way to define regex patterns using
+* `{{token}}` placeholders that expand to their regex equivalents. It supports
+* named capture groups for extracting matched values into metadata.
+*
+* @module tokens
 *
 * @example
-* normalizeBreakpoint('\\n\\n')
-* // → { pattern: '\\n\\n' }
+* // Simple token expansion
+* expandTokens('{{raqms}} {{dash}}')
+* // → '[\\u0660-\\u0669]+ [-–—ـ]'
 *
-* normalizeBreakpoint({ pattern: '\\n', min: 10 })
-* // → { pattern: '\\n', min: 10 }
+* @example
+* // Named capture groups
+* expandTokensWithCaptures('{{raqms:num}} {{dash}}')
+* // → { pattern: '(?<num>[\\u0660-\\u0669]+) [-–—ـ]', captureNames: ['num'], hasCaptures: true }
 */
-const normalizeBreakpoint = (bp) => typeof bp === "string" ? { pattern: bp } : bp;
 /**
-* Checks if a page ID is in an excluded list (single pages or ranges).
+* Token definitions mapping human-readable token names to regex patterns.
 *
-* @param pageId - Page ID to check
-* @param excludeList - List of page IDs or [from, to] ranges to exclude
-* @returns True if page is excluded
+* Tokens are used in template strings with double-brace syntax:
+* - `{{token}}` - Expands to the pattern (non-capturing in context)
+* - `{{token:name}}` - Expands to a named capture group `(?<name>pattern)`
+* - `{{:name}}` - Captures any content with the given name `(?<name>.+)`
 *
-* @example
-* isPageExcluded(5, [1, 5, 10])
-* // → true
+* @remarks
+* These patterns are designed for Arabic text matching. For diacritic-insensitive
+* matching of Arabic patterns, use the `fuzzy: true` option in split rules,
+* which applies `makeDiacriticInsensitive()` to the expanded patterns.
 *
-* isPageExcluded(5, [[3, 7]])
-* // → true
+* @example
+* // Using tokens in a split rule
+* { lineStartsWith: ['{{kitab}}', '{{bab}}'], split: 'at', fuzzy: true }
 *
-* isPageExcluded(5, [[10, 20]])
-* // → false
+* @example
+* // Using tokens with named captures
+* { lineStartsAfter: ['{{raqms:hadithNum}} {{dash}} '], split: 'at' }
 */
-const isPageExcluded = (pageId, excludeList) => {
-	if (!excludeList || excludeList.length === 0) return false;
-	for (const item of excludeList) if (typeof item === "number") {
-		if (pageId === item) return true;
-	} else {
-		const [from, to] = item;
-		if (pageId >= from && pageId <= to) return true;
-	}
-	return false;
-};
 /**
-* Checks if a page ID is within a breakpoint's min/max range and not excluded.
+* Escapes regex metacharacters (parentheses and brackets) in template patterns,
+* but preserves content inside `{{...}}` token delimiters.
 *
-* @param pageId - Page ID to check
-* @param rule - Breakpoint rule with optional min/max/exclude constraints
-* @returns True if page is within valid range
+* 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
-* isInBreakpointRange(50, { pattern: '\\n', min: 10, max: 100 })
-* // → true
+* escapeTemplateBrackets('({{harf}}): ')
+* // → '\\({{harf}}\\): '
 *
-* isInBreakpointRange(5, { pattern: '\\n', min: 10 })
-* // → false (below min)
+* @example
+* escapeTemplateBrackets('[{{raqm}}] ')
+* // → '\\[{{raqm}}\\] '
+*
+* @example
+* escapeTemplateBrackets('{{harf}}')
+* // → '{{harf}}' (unchanged - no brackets outside tokens)
 */
-const isInBreakpointRange = (pageId, rule) => {
-	if (rule.min !== void 0 && pageId < rule.min) return false;
-	if (rule.max !== void 0 && pageId > rule.max) return false;
-	return !isPageExcluded(pageId, rule.exclude);
+const escapeTemplateBrackets = (pattern) => {
+	return pattern.replace(/(\{\{[^}]*\}\})|([()[\]])/g, (_match, token, bracket) => {
+		if (token) return token;
+		return `\\${bracket}`;
+	});
+};
+const RUMUZ_ATOM = `(?:${[
+	"تمييز(?![\\u064B-\\u0652\\u0670أ-ي])",
+	"خت",
+	"خغ",
+	"بخ",
+	"عخ",
+	"مق",
+	"مت",
+	"عس",
+	"سي",
+	"سن",
+	"كن",
+	"مد",
+	"قد",
+	"خد",
+	"فد",
+	"دل",
+	"كد",
+	"غد",
+	"صد",
+	"دت",
+	"دس",
+	"تم",
+	"فق",
+	"دق",
+	"[خرزيمنصسدفلتقع](?![\\u064B-\\u0652\\u0670أ-ي])",
+	"(?<![\\u0660-\\u0669])٤(?![\\u0660-\\u0669])"
+].join("|")})`;
+const RUMUZ_BLOCK = `${RUMUZ_ATOM}(?:\\s+${RUMUZ_ATOM})*`;
+const BASE_TOKENS = {
+	bab: "باب",
+	basmalah: ["بسم الله", "﷽"].join("|"),
+	bullet: "[•*°]",
+	dash: "[-–—ـ]",
+	fasl: ["مسألة", "فصل"].join("|"),
+	harf: "[أ-ي]",
+	harfs: "[أ-ي](?:\\s+[أ-ي])*",
+	kitab: "كتاب",
+	naql: [
+		"حدثني",
+		"وأخبرنا",
+		"حدثنا",
+		"سمعت",
+		"أنبأنا",
+		"وحدثنا",
+		"أخبرنا",
+		"وحدثني",
+		"وحدثنيه"
+	].join("|"),
+	raqm: "[\\u0660-\\u0669]",
+	raqms: "[\\u0660-\\u0669]+",
+	rumuz: RUMUZ_BLOCK,
+	tarqim: "[.!?؟؛]"
 };
 /**
-* Builds an exclude set from a PageRange array for O(1) lookups.
+* Composite token definitions using template syntax.
 *
-* @param excludeList - List of page IDs or [from, to] ranges
-* @returns Set of all excluded page IDs
+* These tokens reference base tokens using `{{token}}` syntax and are
+* automatically expanded to their final regex patterns at module load time.
 *
-* @remarks
-* This expands ranges into explicit page IDs for fast membership checks. For typical
-* book-scale inputs (thousands of pages), this is small and keeps downstream logic
-* simple and fast. If you expect extremely large ranges (e.g., millions of pages),
-* consider avoiding broad excludes or introducing a range-based membership structure.
+* This provides better abstraction - if base tokens change, composites
+* automatically update on the next build.
 *
-* @example
-* buildExcludeSet([1, 5, [10, 12]])
-* // → Set { 1, 5, 10, 11, 12 }
+* @internal
 */
-const buildExcludeSet = (excludeList) => {
-	const excludeSet = /* @__PURE__ */ new Set();
-	for (const item of excludeList || []) if (typeof item === "number") excludeSet.add(item);
-	else for (let i = item[0]; i <= item[1]; i++) excludeSet.add(i);
-	return excludeSet;
-};
+const COMPOSITE_TOKENS = { numbered: "{{raqms}} {{dash}} " };
 /**
-* Creates a segment with optional to and meta fields.
-* Returns null if content is empty after trimming.
+* Expands any *composite* tokens (like `{{numbered}}`) into their underlying template form
+* (like `{{raqms}} {{dash}} `).
 *
-* @param content - Segment content
-* @param fromPageId - Starting page ID
-* @param toPageId - Optional ending page ID (omitted if same as from)
-* @param meta - Optional metadata to attach
-* @returns Segment object or null if empty
+* This is useful when you want to take a signature produced by `analyzeCommonLineStarts()`
+* and turn it into an editable template where you can add named captures, e.g.:
 *
-* @example
-* createSegment('Hello world', 1, 3, { chapter: 1 })
-* // → { content: 'Hello world', from: 1, to: 3, meta: { chapter: 1 } }
+* - `{{numbered}}` → `{{raqms}} {{dash}} `
+* - then: `{{raqms:num}} {{dash}} ` to capture the number
 *
-* createSegment('   ', 1, undefined, undefined)
-* // → null (empty content)
+* Notes:
+* - This only expands the plain `{{token}}` form (not `{{token:name}}`).
+* - Expansion is repeated a few times to support nested composites.
 */
-const createSegment = (content, fromPageId, toPageId, meta) => {
-	const trimmed = content.trim();
-	if (!trimmed) return null;
-	const seg = {
-		content: trimmed,
-		from: fromPageId
-	};
-	if (toPageId !== void 0 && toPageId !== fromPageId) seg.to = toPageId;
-	if (meta) seg.meta = meta;
-	return seg;
+const expandCompositeTokensInTemplate = (template) => {
+	let out = template;
+	for (let i = 0; i < 10; i++) {
+		const next = out.replace(/\{\{(\w+)\}\}/g, (m, tokenName) => {
+			return COMPOSITE_TOKENS[tokenName] ?? m;
+		});
+		if (next === out) break;
+		out = next;
+	}
+	return out;
 };
 /**
-* Expands breakpoint patterns and pre-computes exclude sets.
+* Expands base tokens in a template string.
+* Used internally to pre-expand composite tokens.
 *
-* @param breakpoints - Array of breakpoint patterns or rules
-* @param processPattern - Function to expand tokens in patterns
-* @returns Array of expanded breakpoints with compiled regexes
+* @param template - Template string with `{{token}}` placeholders
+* @returns Expanded pattern with base tokens replaced
+* @internal
+*/
+const expandBaseTokens = (template) => {
+	return template.replace(/\{\{(\w+)\}\}/g, (_, tokenName) => {
+		return BASE_TOKENS[tokenName] ?? `{{${tokenName}}}`;
+	});
+};
+/**
+* Token definitions mapping human-readable token names to regex patterns.
+*
+* Tokens are used in template strings with double-brace syntax:
+* - `{{token}}` - Expands to the pattern (non-capturing in context)
+* - `{{token:name}}` - Expands to a named capture group `(?<name>pattern)`
+* - `{{:name}}` - Captures any content with the given name `(?<name>.+)`
 *
 * @remarks
-* This function compiles regex patterns dynamically. This can be a ReDoS vector
-* if patterns come from untrusted sources. In typical usage, breakpoint rules
-* are application configuration, not user input.
-*/
-const expandBreakpoints = (breakpoints, processPattern$1) => breakpoints.map((bp) => {
-	const rule = normalizeBreakpoint(bp);
-	const excludeSet = buildExcludeSet(rule.exclude);
-	const skipWhenRegex = rule.skipWhen !== void 0 ? (() => {
-		const expandedSkip = processPattern$1(rule.skipWhen);
-		try {
-			return new RegExp(expandedSkip, "mu");
-		} catch (error) {
-			const message = error instanceof Error ? error.message : String(error);
-			throw new Error(`Invalid breakpoint skipWhen regex: ${rule.skipWhen}\n  Cause: ${message}`);
-		}
-	})() : null;
-	if (rule.pattern === "") return {
-		excludeSet,
-		regex: null,
-		rule,
-		skipWhenRegex
-	};
-	const expanded = processPattern$1(rule.pattern);
-	try {
-		return {
-			excludeSet,
-			regex: new RegExp(expanded, "gmu"),
-			rule,
-			skipWhenRegex
-		};
-	} catch (error) {
-		const message = error instanceof Error ? error.message : String(error);
-		throw new Error(`Invalid breakpoint regex: ${rule.pattern}\n  Cause: ${message}`);
-	}
-});
-/**
-* Applies a configured joiner at detected page boundaries within a multi-page content chunk.
+* These patterns are designed for Arabic text matching. For diacritic-insensitive
+* matching of Arabic patterns, use the `fuzzy: true` option in split rules,
+* which applies `makeDiacriticInsensitive()` to the expanded patterns.
 *
-* This is used for breakpoint-generated segments which don't have access to the original
-* `pageMap.pageBreaks` offsets. We detect page starts sequentially by searching for each page's
-* prefix after the previous boundary, then replace ONLY the single newline immediately before
-* that page start.
+* @example
+* // Using tokens in a split rule
+* { lineStartsWith: ['{{kitab}}', '{{bab}}'], split: 'at', fuzzy: true }
 *
-* This avoids converting real in-page newlines, while still normalizing page joins consistently.
-*/
-const applyPageJoinerBetweenPages = (content, fromIdx, toIdx, pageIds, normalizedPages, joiner) => {
-	if (joiner === "newline" || fromIdx >= toIdx || !content.includes("\n")) return content;
-	let updated = content;
-	let searchFrom = 0;
-	for (let pi = fromIdx + 1; pi <= toIdx; pi++) {
-		const pageData = normalizedPages.get(pageIds[pi]);
-		if (!pageData) continue;
-		const found = findPrefixPositionInContent(updated, pageData.content.trimStart(), searchFrom);
-		if (found > 0 && updated[found - 1] === "\n") updated = `${updated.slice(0, found - 1)} ${updated.slice(found)}`;
-		if (found > 0) searchFrom = found;
-	}
-	return updated;
-};
-/**
-* Finds the position of a page prefix in content, trying multiple prefix lengths.
-*/
-const findPrefixPositionInContent = (content, trimmedPageContent, searchFrom) => {
-	for (const len of JOINER_PREFIX_LENGTHS) {
-		const prefix = trimmedPageContent.slice(0, Math.min(len, trimmedPageContent.length)).trim();
-		if (!prefix) continue;
-		const pos = content.indexOf(prefix, searchFrom);
-		if (pos > 0) return pos;
-	}
-	return -1;
-};
-/**
-* Estimates how far into the current page `remainingContent` begins.
+* @example
+* // Using tokens with named captures
+* { lineStartsAfter: ['{{raqms:hadithNum}} {{dash}} '], split: 'at' }
 *
-* During breakpoint processing, `remainingContent` can begin mid-page after a previous split.
-* When that happens, raw cumulative page offsets (computed from full page starts) can overestimate
-* expected boundary positions. This helper computes an approximate starting offset by matching
-* a short prefix of `remainingContent` inside the current page content.
+* @example
+* // Using the numbered convenience token
+* { lineStartsAfter: ['{{numbered}}'], split: 'at' }
 */
-const estimateStartOffsetInCurrentPage = (remainingContent, currentFromIdx, pageIds, normalizedPages) => {
-	const currentPageData = normalizedPages.get(pageIds[currentFromIdx]);
-	if (!currentPageData) return 0;
-	const remStart = remainingContent.trimStart().slice(0, Math.min(60, remainingContent.length));
-	const needle = remStart.slice(0, Math.min(30, remStart.length));
-	if (!needle) return 0;
-	const idx = currentPageData.content.indexOf(needle);
-	return idx > 0 ? idx : 0;
+const TOKEN_PATTERNS = {
+	...BASE_TOKENS,
+	...Object.fromEntries(Object.entries(COMPOSITE_TOKENS).map(([k, v]) => [k, expandBaseTokens(v)]))
 };
 /**
-* Attempts to find the start position of a target page within remainingContent,
-* anchored near an expected boundary position to reduce collisions.
+* Regex pattern for matching tokens with optional named capture syntax.
 *
-* This is used to define breakpoint windows in terms of actual content being split, rather than
-* raw per-page offsets which can desync when structural rules strip markers.
+* Matches:
+* - `{{token}}` - Simple token (group 1 = token name, group 2 = empty)
+* - `{{token:name}}` - Token with capture (group 1 = token, group 2 = name)
+* - `{{:name}}` - Capture-only (group 1 = empty, group 2 = name)
+*
+* @internal
 */
-const findPageStartNearExpectedBoundary = (remainingContent, _currentFromIdx, targetPageIdx, expectedBoundary, pageIds, normalizedPages) => {
-	const targetPageData = normalizedPages.get(pageIds[targetPageIdx]);
-	if (!targetPageData) return -1;
-	const approx = Math.min(Math.max(0, expectedBoundary), remainingContent.length);
-	const searchStart = Math.max(0, approx - 1e4);
-	const searchEnd = Math.min(remainingContent.length, approx + 2e3);
-	const targetTrimmed = targetPageData.content.trimStart();
-	for (const len of WINDOW_PREFIX_LENGTHS) {
-		const prefix = targetTrimmed.slice(0, Math.min(len, targetTrimmed.length)).trim();
-		if (!prefix) continue;
-		let pos = remainingContent.indexOf(prefix, searchStart);
-		while (pos !== -1 && pos <= searchEnd) {
-			if (pos > 0 && /\s/.test(remainingContent[pos - 1] ?? "")) return pos;
-			pos = remainingContent.indexOf(prefix, pos + 1);
-		}
-		const last = remainingContent.lastIndexOf(prefix, approx);
-		if (last > 0) return last;
-	}
-	return -1;
-};
+const TOKEN_WITH_CAPTURE_REGEX = /\{\{(\w*):?(\w*)\}\}/g;
 /**
-* Builds a boundary position map for pages within the given range.
-*
-* This function computes page boundaries once per segment and enables
-* O(log n) page lookups via binary search with `findPageIndexForPosition`.
-*
-* Boundaries are derived from segmentContent (post-structural-rules).
-* When the segment starts mid-page, an offset correction is applied to
-* keep boundary estimates aligned with the segment's actual content space.
+* Regex pattern for simple token matching (no capture syntax).
 *
-* @param segmentContent - Full segment content (already processed by structural rules)
-* @param fromIdx - Starting page index
-* @param toIdx - Ending page index
-* @param pageIds - Array of all page IDs
-* @param normalizedPages - Map of page ID to normalized content
-* @param cumulativeOffsets - Cumulative character offsets (for estimates)
-* @returns Array where boundaryPositions[i] = start position of page (fromIdx + i),
-*          with a sentinel boundary at segmentContent.length as the last element
+* Matches only `{{token}}` format where token is one or more word characters.
+* Used by `containsTokens()` for quick detection.
 *
-* @example
-* // For a 3-page segment:
-* buildBoundaryPositions(content, 0, 2, pageIds, normalizedPages, offsets)
-* // → [0, 23, 45, 67] where 67 is content.length (sentinel)
+* @internal
 */
-const buildBoundaryPositions = (segmentContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets) => {
-	const boundaryPositions = [0];
-	const startOffsetInFromPage = estimateStartOffsetInCurrentPage(segmentContent, fromIdx, pageIds, normalizedPages);
-	for (let i = fromIdx + 1; i <= toIdx; i++) {
-		const expectedBoundary = cumulativeOffsets[i] !== void 0 && cumulativeOffsets[fromIdx] !== void 0 ? Math.max(0, cumulativeOffsets[i] - cumulativeOffsets[fromIdx] - startOffsetInFromPage) : segmentContent.length;
-		const pos = findPageStartNearExpectedBoundary(segmentContent, fromIdx, i, expectedBoundary, pageIds, normalizedPages);
-		const prevBoundary = boundaryPositions[boundaryPositions.length - 1];
-		if (pos > 0 && pos > prevBoundary && Math.abs(pos - expectedBoundary) < 2e3) boundaryPositions.push(pos);
-		else {
-			const estimate = Math.max(prevBoundary + 1, expectedBoundary);
-			boundaryPositions.push(Math.min(estimate, segmentContent.length));
-		}
-	}
-	boundaryPositions.push(segmentContent.length);
-	return boundaryPositions;
-};
+const SIMPLE_TOKEN_REGEX = /\{\{(\w+)\}\}/g;
 /**
-* Binary search to find which page a position falls within.
-* Uses "largest i where boundaryPositions[i] <= position" semantics.
+* Checks if a query string contains template tokens.
 *
-* @param position - Character position in segmentContent
-* @param boundaryPositions - Precomputed boundary positions (from buildBoundaryPositions)
-* @param fromIdx - Base page index (boundaryPositions[0] corresponds to pageIds[fromIdx])
-* @returns Page index in pageIds array
+* Performs a quick test for `{{token}}` patterns without actually
+* expanding them. Useful for determining whether to apply token
+* expansion to a string.
+*
+* @param query - String to check for tokens
+* @returns `true` if the string contains at least one `{{token}}` pattern
 *
 * @example
-* // With boundaries [0, 20, 40, 60] and fromIdx=0:
-* findPageIndexForPosition(15, boundaries, 0) // → 0 (first page)
-* findPageIndexForPosition(25, boundaries, 0) // → 1 (second page)
-* findPageIndexForPosition(40, boundaries, 0) // → 2 (exactly on boundary = that page)
+* containsTokens('{{raqms}} {{dash}}') // → true
+* containsTokens('plain text')          // → false
+* containsTokens('[٠-٩]+ - ')           // → false (raw regex, no tokens)
 */
-const findPageIndexForPosition = (position, boundaryPositions, fromIdx) => {
-	if (boundaryPositions.length <= 1) return fromIdx;
-	let left = 0;
-	let right = boundaryPositions.length - 2;
-	while (left < right) {
-		const mid = Math.ceil((left + right) / 2);
-		if (boundaryPositions[mid] <= position) left = mid;
-		else right = mid - 1;
-	}
-	return fromIdx + left;
+const containsTokens = (query) => {
+	SIMPLE_TOKEN_REGEX.lastIndex = 0;
+	return SIMPLE_TOKEN_REGEX.test(query);
 };
-/**
-* Finds the end position of a breakpoint window inside `remainingContent`.
-*
-* The window end is defined as the start of the page AFTER `windowEndIdx` (i.e. `windowEndIdx + 1`),
-* found within the actual `remainingContent` string being split. This avoids relying on raw page offsets
-* that can diverge when structural rules strip markers (e.g. `lineStartsAfter`).
-*/
-const findBreakpointWindowEndPosition = (remainingContent, currentFromIdx, windowEndIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets) => {
-	if (windowEndIdx >= toIdx) return remainingContent.length;
-	const desiredNextIdx = windowEndIdx + 1;
-	const minNextIdx = currentFromIdx + 1;
-	const maxNextIdx = Math.min(desiredNextIdx, toIdx);
-	const startOffsetInCurrentPage = estimateStartOffsetInCurrentPage(remainingContent, currentFromIdx, pageIds, normalizedPages);
-	for (let nextIdx = maxNextIdx; nextIdx >= minNextIdx; nextIdx--) {
-		const expectedBoundary = cumulativeOffsets[nextIdx] !== void 0 && cumulativeOffsets[currentFromIdx] !== void 0 ? Math.max(0, cumulativeOffsets[nextIdx] - cumulativeOffsets[currentFromIdx] - startOffsetInCurrentPage) : remainingContent.length;
-		const pos = findPageStartNearExpectedBoundary(remainingContent, currentFromIdx, nextIdx, expectedBoundary, pageIds, normalizedPages);
-		if (pos > 0) return pos;
+const splitTemplateIntoSegments = (query) => {
+	const segments = [];
+	let lastIndex = 0;
+	TOKEN_WITH_CAPTURE_REGEX.lastIndex = 0;
+	let match;
+	while ((match = TOKEN_WITH_CAPTURE_REGEX.exec(query)) !== null) {
+		if (match.index > lastIndex) segments.push({
+			type: "text",
+			value: query.slice(lastIndex, match.index)
+		});
+		segments.push({
+			type: "token",
+			value: match[0]
+		});
+		lastIndex = match.index + match[0].length;
 	}
-	return remainingContent.length;
+	if (lastIndex < query.length) segments.push({
+		type: "text",
+		value: query.slice(lastIndex)
+	});
+	return segments;
 };
-/**
-* Finds exclusion-based break position using raw cumulative offsets.
-*
-* This is used to ensure pages excluded by breakpoints are never merged into the same output segment.
-* Returns a break position relative to the start of `remainingContent` (i.e. the currentFromIdx start).
-*/
-const findExclusionBreakPosition = (currentFromIdx, windowEndIdx, toIdx, pageIds, expandedBreakpoints, cumulativeOffsets) => {
-	const startingPageId = pageIds[currentFromIdx];
-	if (expandedBreakpoints.some((bp) => bp.excludeSet.has(startingPageId)) && currentFromIdx < toIdx) return cumulativeOffsets[currentFromIdx + 1] - cumulativeOffsets[currentFromIdx];
-	for (let pageIdx = currentFromIdx + 1; pageIdx <= windowEndIdx; pageIdx++) {
-		const pageId = pageIds[pageIdx];
-		if (expandedBreakpoints.some((bp) => bp.excludeSet.has(pageId))) return cumulativeOffsets[pageIdx] - cumulativeOffsets[currentFromIdx];
-	}
-	return -1;
+const maybeApplyFuzzyToText = (text, fuzzyTransform) => {
+	if (fuzzyTransform && /[\u0600-\u06FF]/u.test(text)) return fuzzyTransform(text);
+	return text;
 };
-/**
-* Checks if any page in a range is excluded by the given exclude set.
+const maybeApplyFuzzyToTokenPattern = (tokenPattern, fuzzyTransform) => {
+	if (!fuzzyTransform) return tokenPattern;
+	return tokenPattern.split("|").map((part) => /[\u0600-\u06FF]/u.test(part) ? fuzzyTransform(part) : part).join("|");
+};
+const parseTokenLiteral = (literal) => {
+	TOKEN_WITH_CAPTURE_REGEX.lastIndex = 0;
+	const tokenMatch = TOKEN_WITH_CAPTURE_REGEX.exec(literal);
+	if (!tokenMatch) return null;
+	const [, tokenName, captureName] = tokenMatch;
+	return {
+		captureName,
+		tokenName
+	};
+};
+const createCaptureRegistry = (capturePrefix) => {
+	const captureNames = [];
+	const captureNameCounts = /* @__PURE__ */ new Map();
+	const register = (baseName) => {
+		const count = captureNameCounts.get(baseName) ?? 0;
+		captureNameCounts.set(baseName, count + 1);
+		const uniqueName = count === 0 ? baseName : `${baseName}_${count + 1}`;
+		const prefixedName = capturePrefix ? `${capturePrefix}${uniqueName}` : uniqueName;
+		captureNames.push(prefixedName);
+		return prefixedName;
+	};
+	return {
+		captureNames,
+		register
+	};
+};
+const expandTokenLiteral = (literal, opts) => {
+	const parsed = parseTokenLiteral(literal);
+	if (!parsed) return literal;
+	const { tokenName, captureName } = parsed;
+	if (!tokenName && captureName) return `(?<${opts.registerCapture(captureName)}>.+)`;
+	let tokenPattern = TOKEN_PATTERNS[tokenName];
+	if (!tokenPattern) return literal;
+	tokenPattern = maybeApplyFuzzyToTokenPattern(tokenPattern, opts.fuzzyTransform);
+	if (captureName) return `(?<${opts.registerCapture(captureName)}>${tokenPattern})`;
+	return tokenPattern;
+};
+/**
+* Expands template tokens with support for named captures.
 *
-* @param excludeSet - Set of excluded page IDs
-* @param pageIds - Array of page IDs
-* @param fromIdx - Start index (inclusive)
-* @param toIdx - End index (inclusive)
-* @returns True if any page in range is excluded
+* This is the primary token expansion function that handles all token syntax:
+* - `{{token}}` → Expands to the token's pattern (no capture group)
+* - `{{token:name}}` → Expands to `(?<name>pattern)` (named capture)
+* - `{{:name}}` → Expands to `(?<name>.+)` (capture anything)
+*
+* Unknown tokens are left as-is in the output, allowing for partial templates.
+*
+* @param query - The template string containing tokens
+* @param fuzzyTransform - Optional function to transform Arabic text for fuzzy matching.
+*                         Applied to both token patterns and plain Arabic text between tokens.
+*                         Typically `makeDiacriticInsensitive` from the fuzzy module.
+* @returns Object with expanded pattern, capture names, and capture flag
+*
+* @example
+* // Simple token expansion
+* expandTokensWithCaptures('{{raqms}} {{dash}}')
+* // → { pattern: '[\\u0660-\\u0669]+ [-–—ـ]', captureNames: [], hasCaptures: false }
+*
+* @example
+* // Named capture
+* expandTokensWithCaptures('{{raqms:num}} {{dash}}')
+* // → { pattern: '(?<num>[\\u0660-\\u0669]+) [-–—ـ]', captureNames: ['num'], hasCaptures: true }
+*
+* @example
+* // Capture-only token
+* expandTokensWithCaptures('{{raqms:num}} {{dash}} {{:content}}')
+* // → { pattern: '(?<num>[٠-٩]+) [-–—ـ] (?<content>.+)', captureNames: ['num', 'content'], hasCaptures: true }
+*
+* @example
+* // With fuzzy transform
+* expandTokensWithCaptures('{{bab}}', makeDiacriticInsensitive)
+* // → { pattern: 'بَ?ا?بٌ?', captureNames: [], hasCaptures: false }
 */
-const hasExcludedPageInRange = (excludeSet, pageIds, fromIdx, toIdx) => {
-	if (excludeSet.size === 0) return false;
-	for (let pageIdx = fromIdx; pageIdx <= toIdx; pageIdx++) if (excludeSet.has(pageIds[pageIdx])) return true;
-	return false;
+const expandTokensWithCaptures = (query, fuzzyTransform, capturePrefix) => {
+	const segments = splitTemplateIntoSegments(query);
+	const registry = createCaptureRegistry(capturePrefix);
+	const processedParts = segments.map((segment) => {
+		if (segment.type === "text") return maybeApplyFuzzyToText(segment.value, fuzzyTransform);
+		return expandTokenLiteral(segment.value, {
+			capturePrefix,
+			fuzzyTransform,
+			registerCapture: registry.register
+		});
+	});
+	return {
+		captureNames: registry.captureNames,
+		hasCaptures: registry.captureNames.length > 0,
+		pattern: processedParts.join("")
+	};
 };
 /**
-* Finds the position of the next page content within remaining content.
-* Returns -1 if not found.
+* Expands template tokens in a query string to their regex equivalents.
 *
-* @param remainingContent - Content to search in
-* @param nextPageData - Normalized data for the next page
-* @returns Position of next page content, or -1 if not found
+* This is the simple version without capture support. It returns only the
+* expanded pattern string, not capture metadata.
+*
+* Unknown tokens are left as-is, allowing for partial templates.
+*
+* @param query - Template string containing `{{token}}` placeholders
+* @returns Expanded regex pattern string
+*
+* @example
+* expandTokens('، {{raqms}}')     // → '، [\\u0660-\\u0669]+'
+* expandTokens('{{raqm}}*')       // → '[\\u0660-\\u0669]*'
+* expandTokens('{{dash}}{{raqm}}') // → '[-–—ـ][\\u0660-\\u0669]'
+* expandTokens('{{unknown}}')     // → '{{unknown}}' (left as-is)
+*
+* @see expandTokensWithCaptures for full capture group support
 */
-const findNextPagePosition = (remainingContent, nextPageData) => {
-	const searchPrefix = nextPageData.content.trim().slice(0, Math.min(30, nextPageData.length));
-	if (searchPrefix.length === 0) return -1;
-	const pos = remainingContent.indexOf(searchPrefix);
-	return pos > 0 ? pos : -1;
-};
+const expandTokens = (query) => expandTokensWithCaptures(query).pattern;
 /**
-* Finds matches within a window and returns the selected position based on preference.
+* Converts a template string to a compiled RegExp.
 *
-* @param windowContent - Content to search
-* @param regex - Regex to match
-* @param prefer - 'longer' for last match, 'shorter' for first match
-* @returns Break position after the selected match, or -1 if no matches
+* Expands all tokens and attempts to compile the result as a RegExp
+* with Unicode flag. Returns `null` if the resulting pattern is invalid.
+*
+* @remarks
+* This function dynamically compiles regular expressions from template strings.
+* If templates may come from untrusted sources, be aware of potential ReDoS
+* (Regular Expression Denial of Service) risks due to catastrophic backtracking.
+* Consider validating pattern complexity or applying execution timeouts when
+* running user-submitted patterns.
+*
+* @param template - Template string containing `{{token}}` placeholders
+* @returns Compiled RegExp with 'u' flag, or `null` if invalid
+*
+* @example
+* templateToRegex('، {{raqms}}')  // → /، [٠-٩]+/u
+* templateToRegex('{{raqms}}+')   // → /[٠-٩]++/u (might be invalid in some engines)
+* templateToRegex('(((')          // → null (invalid regex)
 */
-const findPatternBreakPosition = (windowContent, regex, prefer) => {
-	let first;
-	let last;
-	for (const m of windowContent.matchAll(regex)) {
-		const match = {
-			index: m.index,
-			length: m[0].length
-		};
-		if (!first) first = match;
-		last = match;
+const templateToRegex = (template) => {
+	const expanded = expandTokens(template);
+	try {
+		return new RegExp(expanded, "u");
+	} catch {
+		return null;
 	}
-	if (!first) return -1;
-	const selected = prefer === "longer" ? last : first;
-	return selected.index + selected.length;
 };
 /**
-* Handles page boundary breakpoint (empty pattern).
-* Returns break position or -1 if no valid position found.
+* Lists all available token names defined in `TOKEN_PATTERNS`.
+*
+* Useful for documentation, validation, or building user interfaces
+* that show available tokens.
+*
+* @returns Array of token names (e.g., `['bab', 'basmala', 'bullet', ...]`)
+*
+* @example
+* getAvailableTokens()
+* // → ['bab', 'basmala', 'bullet', 'dash', 'harf', 'kitab', 'naql', 'raqm', 'raqms']
 */
-const handlePageBoundaryBreak = (remainingContent, windowEndIdx, windowEndPosition, toIdx, pageIds, normalizedPages) => {
-	const nextPageIdx = windowEndIdx + 1;
-	if (nextPageIdx <= toIdx) {
-		const nextPageData = normalizedPages.get(pageIds[nextPageIdx]);
-		if (nextPageData) {
-			const pos = findNextPagePosition(remainingContent, nextPageData);
-			if (pos > 0) return Math.min(pos, windowEndPosition, remainingContent.length);
-		}
-	}
-	return Math.min(windowEndPosition, remainingContent.length);
-};
+const getAvailableTokens = () => Object.keys(TOKEN_PATTERNS);
 /**
-* Tries to find a break position within the current window using breakpoint patterns.
-* Returns the break position or -1 if no suitable break was found.
+* Gets the regex pattern for a specific token name.
 *
-* @param remainingContent - Content remaining to be segmented
-* @param currentFromIdx - Current starting page index
-* @param toIdx - Ending page index
-* @param windowEndIdx - Maximum window end index
-* @param ctx - Breakpoint context with page data and patterns
-* @returns Break position in the content, or -1 if no break found
+* 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
+*
+* @example
+* getTokenPattern('raqms')   // → '[\\u0660-\\u0669]+'
+* getTokenPattern('dash')    // → '[-–—ـ]'
+* getTokenPattern('unknown') // → undefined
 */
-const findBreakPosition = (remainingContent, currentFromIdx, toIdx, windowEndIdx, windowEndPosition, ctx) => {
-	const { pageIds, normalizedPages, expandedBreakpoints, prefer } = ctx;
-	for (const { rule, regex, excludeSet, skipWhenRegex } of expandedBreakpoints) {
-		if (!isInBreakpointRange(pageIds[currentFromIdx], rule)) continue;
-		if (hasExcludedPageInRange(excludeSet, pageIds, currentFromIdx, windowEndIdx)) continue;
-		if (skipWhenRegex?.test(remainingContent)) continue;
-		if (regex === null) return handlePageBoundaryBreak(remainingContent, windowEndIdx, windowEndPosition, toIdx, pageIds, normalizedPages);
-		const breakPos = findPatternBreakPosition(remainingContent.slice(0, Math.min(windowEndPosition, remainingContent.length)), regex, prefer);
-		if (breakPos > 0) return breakPos;
-	}
-	return -1;
+const getTokenPattern = (tokenName) => TOKEN_PATTERNS[tokenName];
+/**
+* Regex to detect fuzzy-default tokens in a pattern string.
+* Matches {{token}} or {{token:name}} syntax.
+*/
+const FUZZY_TOKEN_REGEX = new RegExp(`\\{\\{(?:${[
+	"bab",
+	"basmalah",
+	"fasl",
+	"kitab",
+	"naql"
+].join("|")})(?::\\w+)?\\}\\}`, "g");
+/**
+* Checks if a pattern (or array of patterns) contains tokens that should
+* default to fuzzy matching.
+*
+* Fuzzy-default tokens are: bab, basmalah, fasl, kitab, naql
+*
+* @param patterns - Single pattern string or array of pattern strings
+* @returns `true` if any pattern contains a fuzzy-default token
+*
+* @example
+* shouldDefaultToFuzzy('{{bab}} الإيمان')     // true
+* shouldDefaultToFuzzy('{{raqms}} {{dash}}')  // false
+* shouldDefaultToFuzzy(['{{kitab}}', '{{raqms}}']) // true
+*/
+const shouldDefaultToFuzzy = (patterns) => {
+	return (Array.isArray(patterns) ? patterns : [patterns]).some((p) => {
+		FUZZY_TOKEN_REGEX.lastIndex = 0;
+		return FUZZY_TOKEN_REGEX.test(p);
+	});
 };
 
 //#endregion
-//#region src/segmentation/breakpoint-processor.ts
+//#region src/segmentation/pattern-validator.ts
 /**
-* Breakpoint post-processing engine extracted from segmenter.ts.
+* Pattern validation utilities for detecting common mistakes in rule patterns.
 *
-* This module is intentionally split into small helpers to reduce cognitive complexity
-* and allow unit testing of tricky edge cases (window sizing, next-page advancement, etc.).
+* These utilities help catch typos and issues early, before rules are used
+* for segmentation.
 */
-const buildPageIdToIndexMap = (pageIds) => new Map(pageIds.map((id, i) => [id, i]));
-const buildNormalizedPagesMap = (pages, normalizedContent) => {
-	const normalizedPages = /* @__PURE__ */ new Map();
-	for (let i = 0; i < pages.length; i++) {
-		const content = normalizedContent[i];
-		normalizedPages.set(pages[i].id, {
-			content,
-			index: i,
-			length: content.length
-		});
-	}
-	return normalizedPages;
+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");
 };
-const buildCumulativeOffsets = (pageIds, normalizedPages) => {
-	const cumulativeOffsets = [0];
-	let totalOffset = 0;
-	for (let i = 0; i < pageIds.length; i++) {
-		const pageData = normalizedPages.get(pageIds[i]);
-		totalOffset += pageData ? pageData.length : 0;
-		if (i < pageIds.length - 1) totalOffset += 1;
-		cumulativeOffsets.push(totalOffset);
-	}
-	return cumulativeOffsets;
-};
-const hasAnyExclusionsInRange = (expandedBreakpoints, pageIds, fromIdx, toIdx) => expandedBreakpoints.some((bp) => hasExcludedPageInRange(bp.excludeSet, pageIds, fromIdx, toIdx));
-const computeWindowEndIdx = (currentFromIdx, toIdx, pageIds, maxPages) => {
-	const maxWindowPageId = pageIds[currentFromIdx] + maxPages;
-	let windowEndIdx = currentFromIdx;
-	for (let i = currentFromIdx; i <= toIdx; i++) if (pageIds[i] <= maxWindowPageId) windowEndIdx = i;
-	else break;
-	return windowEndIdx;
-};
-const computeRemainingSpan = (currentFromIdx, toIdx, pageIds) => pageIds[toIdx] - pageIds[currentFromIdx];
-const createFinalSegment = (remainingContent, currentFromIdx, toIdx, pageIds, meta, includeMeta) => createSegment(remainingContent, pageIds[currentFromIdx], currentFromIdx !== toIdx ? pageIds[toIdx] : void 0, includeMeta ? meta : void 0);
 /**
-* Computes the actual start and end page indices for a piece using
-* precomputed boundary positions and binary search.
-*
-* @param pieceStartPos - Start position of the piece in the full segment content
-* @param pieceEndPos - End position (exclusive) of the piece
-* @param boundaryPositions - Precomputed boundary positions from buildBoundaryPositions
-* @param baseFromIdx - Base page index (boundaryPositions[0] corresponds to pageIds[baseFromIdx])
-* @param toIdx - Maximum page index
-* @returns Object with actualStartIdx and actualEndIdx
+* Validates a single pattern for common issues.
 */
-const computePiecePages = (pieceStartPos, pieceEndPos, boundaryPositions, baseFromIdx, toIdx) => {
-	const actualStartIdx = findPageIndexForPosition(pieceStartPos, boundaryPositions, baseFromIdx);
-	const endPos = Math.max(pieceStartPos, pieceEndPos - 1);
-	return {
-		actualEndIdx: Math.min(findPageIndexForPosition(endPos, boundaryPositions, baseFromIdx), toIdx),
-		actualStartIdx
+const validatePattern = (pattern, seenPatterns) => {
+	if (seenPatterns.has(pattern)) return {
+		message: `Duplicate pattern: "${pattern}"`,
+		type: "duplicate"
 	};
-};
-const computeNextFromIdx = (remainingContent, actualEndIdx, toIdx, pageIds, normalizedPages) => {
-	let nextFromIdx = actualEndIdx;
-	if (remainingContent && actualEndIdx + 1 <= toIdx) {
-		const nextPageData = normalizedPages.get(pageIds[actualEndIdx + 1]);
-		if (nextPageData) {
-			const nextPrefix = nextPageData.content.slice(0, Math.min(30, nextPageData.length));
-			const remainingPrefix = remainingContent.trimStart().slice(0, Math.min(30, remainingContent.length));
-			if (nextPrefix && (remainingContent.startsWith(nextPrefix) || nextPageData.content.startsWith(remainingPrefix))) nextFromIdx = actualEndIdx + 1;
-		}
+	seenPatterns.add(pattern);
+	const tokensInBraces = [...pattern.matchAll(TOKEN_INSIDE_BRACES)];
+	for (const match of tokensInBraces) {
+		const tokenName = match[1];
+		if (!KNOWN_TOKENS.has(tokenName)) return {
+			message: `Unknown token: {{${tokenName}}}. Available tokens: ${[...KNOWN_TOKENS].slice(0, 5).join(", ")}...`,
+			suggestion: `Check spelling or use a known token`,
+			type: "unknown_token"
+		};
 	}
-	return nextFromIdx;
-};
-const createPieceSegment = (pieceContent, actualStartIdx, actualEndIdx, pageIds, meta, includeMeta) => createSegment(pieceContent, pageIds[actualStartIdx], actualEndIdx > actualStartIdx ? pageIds[actualEndIdx] : void 0, includeMeta ? meta : void 0);
-/**
-* Finds the break offset within a window, trying exclusions first, then patterns.
-*
-* @returns Break offset relative to remainingContent, or windowEndPosition as fallback
-*/
-const findBreakOffsetForWindow = (remainingContent, currentFromIdx, windowEndIdx, toIdx, windowEndPosition, pageIds, expandedBreakpoints, cumulativeOffsets, normalizedPages, prefer) => {
-	if (hasAnyExclusionsInRange(expandedBreakpoints, pageIds, currentFromIdx, windowEndIdx)) {
-		const exclusionBreak = findExclusionBreakPosition(currentFromIdx, windowEndIdx, toIdx, pageIds, expandedBreakpoints, cumulativeOffsets);
-		if (exclusionBreak > 0) return exclusionBreak;
+	const bareTokenRegex = buildBareTokenRegex();
+	const bareMatches = [...pattern.matchAll(bareTokenRegex)];
+	for (const match of bareMatches) {
+		const tokenName = match[1];
+		const fullMatch = match[0];
+		const matchIndex = match.index;
+		const before = pattern.slice(Math.max(0, matchIndex - 2), matchIndex);
+		const after = pattern.slice(matchIndex + fullMatch.length, matchIndex + fullMatch.length + 2);
+		if (before !== "{{" && after !== "}}") return {
+			message: `Token "${tokenName}" appears to be missing {{}}. Did you mean "{{${fullMatch}}}"?`,
+			suggestion: `{{${fullMatch}}}`,
+			type: "missing_braces"
+		};
 	}
-	const patternBreak = findBreakPosition(remainingContent, currentFromIdx, toIdx, windowEndIdx, windowEndPosition, {
-		expandedBreakpoints,
-		normalizedPages,
-		pageIds,
-		prefer
-	});
-	return patternBreak > 0 ? patternBreak : windowEndPosition;
 };
 /**
-* Advances cursor position past any leading whitespace.
+* Validates an array of patterns, returning parallel array of issues.
 */
-const skipWhitespace = (content, startPos) => {
-	let pos = startPos;
-	while (pos < content.length && /\s/.test(content[pos])) pos++;
-	return pos;
+const validatePatternArray = (patterns) => {
+	const seenPatterns = /* @__PURE__ */ new Set();
+	const issues = patterns.map((p) => validatePattern(p, seenPatterns));
+	if (issues.every((i) => i === void 0)) return;
+	return issues;
 };
 /**
-* Processes an oversized segment by iterating through the content and
-* breaking it into smaller pieces that fit within maxPages constraints.
+* Validates split rules for common pattern issues.
 *
-* Uses precomputed boundary positions for O(log n) page attribution lookups.
-*/
-const processOversizedSegment = (segment, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, expandedBreakpoints, maxPages, prefer, logger) => {
-	const result = [];
-	const fullContent = segment.content;
-	let cursorPos = 0;
-	let currentFromIdx = fromIdx;
-	let isFirstPiece = true;
-	const boundaryPositions = buildBoundaryPositions(fullContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets);
-	logger?.debug?.("[breakpoints] boundaryPositions built", {
-		boundaryPositions,
-		fromIdx,
-		fullContentLength: fullContent.length,
-		toIdx
-	});
-	const maxIterations = 1e4;
-	for (let i = 0; i < maxIterations && cursorPos < fullContent.length && currentFromIdx <= toIdx; i++) {
-		const remainingContent = fullContent.slice(cursorPos);
-		if (!remainingContent.trim()) break;
-		const remainingSpan = computeRemainingSpan(currentFromIdx, toIdx, pageIds);
-		const remainingHasExclusions = hasAnyExclusionsInRange(expandedBreakpoints, pageIds, currentFromIdx, toIdx);
-		if (remainingSpan <= maxPages && !remainingHasExclusions) {
-			const finalSeg = createFinalSegment(remainingContent, currentFromIdx, toIdx, pageIds, segment.meta, isFirstPiece);
-			if (finalSeg) result.push(finalSeg);
-			break;
+* 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) => {
+	return rules.map((rule) => {
+		const result = {};
+		let hasIssues = false;
+		if ("lineStartsWith" in rule && rule.lineStartsWith) {
+			const issues = validatePatternArray(rule.lineStartsWith);
+			if (issues) {
+				result.lineStartsWith = issues;
+				hasIssues = true;
+			}
 		}
-		const windowEndIdx = computeWindowEndIdx(currentFromIdx, toIdx, pageIds, maxPages);
-		const windowEndPosition = findBreakpointWindowEndPosition(remainingContent, currentFromIdx, windowEndIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets);
-		logger?.debug?.(`[breakpoints] iteration=${i}`, {
-			currentFromIdx,
-			cursorPos,
-			windowEndIdx
-		});
-		const breakOffset = findBreakOffsetForWindow(remainingContent, currentFromIdx, windowEndIdx, toIdx, windowEndPosition, pageIds, expandedBreakpoints, cumulativeOffsets, normalizedPages, prefer);
-		const breakPos = cursorPos + breakOffset;
-		const pieceContent = fullContent.slice(cursorPos, breakPos).trim();
-		const { actualEndIdx, actualStartIdx } = computePiecePages(cursorPos, breakPos, boundaryPositions, fromIdx, toIdx);
-		logger?.trace?.("[breakpoints] piece", {
-			actualEndIdx,
-			actualStartIdx,
-			pieceLength: pieceContent.length
-		});
-		if (pieceContent) {
-			const pieceSeg = createPieceSegment(pieceContent, actualStartIdx, actualEndIdx, pageIds, segment.meta, isFirstPiece);
-			if (pieceSeg) result.push(pieceSeg);
+		if ("lineStartsAfter" in rule && rule.lineStartsAfter) {
+			const issues = validatePatternArray(rule.lineStartsAfter);
+			if (issues) {
+				result.lineStartsAfter = issues;
+				hasIssues = true;
+			}
 		}
-		cursorPos = skipWhitespace(fullContent, breakPos);
-		currentFromIdx = computeNextFromIdx(fullContent.slice(cursorPos), actualEndIdx, toIdx, pageIds, normalizedPages);
-		isFirstPiece = false;
+		if ("lineEndsWith" in rule && rule.lineEndsWith) {
+			const issues = validatePatternArray(rule.lineEndsWith);
+			if (issues) {
+				result.lineEndsWith = issues;
+				hasIssues = true;
+			}
+		}
+		if ("template" in rule && rule.template) {
+			const seenPatterns = /* @__PURE__ */ new Set();
+			const issue = validatePattern(rule.template, seenPatterns);
+			if (issue) {
+				result.template = issue;
+				hasIssues = true;
+			}
+		}
+		return hasIssues ? result : void 0;
+	});
+};
+
+//#endregion
+//#region src/segmentation/replace.ts
+const DEFAULT_REPLACE_FLAGS = "gu";
+const normalizeReplaceFlags = (flags) => {
+	if (!flags) return DEFAULT_REPLACE_FLAGS;
+	const allowed = new Set([
+		"g",
+		"i",
+		"m",
+		"s",
+		"u",
+		"y"
+	]);
+	const set = /* @__PURE__ */ new Set();
+	for (const ch of flags) {
+		if (!allowed.has(ch)) throw new Error(`Invalid replace regex flag: "${ch}" (allowed: gimsyu)`);
+		set.add(ch);
 	}
-	logger?.debug?.("[breakpoints] done", { resultCount: result.length });
-	return result;
+	set.add("g");
+	set.add("u");
+	return [
+		"g",
+		"i",
+		"m",
+		"s",
+		"y",
+		"u"
+	].filter((c) => set.has(c)).join("");
+};
+const compileReplaceRules = (rules) => {
+	const compiled = [];
+	for (const r of rules) {
+		if (r.pageIds && r.pageIds.length === 0) continue;
+		const flags = normalizeReplaceFlags(r.flags);
+		const re = new RegExp(r.regex, flags);
+		compiled.push({
+			pageIdSet: r.pageIds ? new Set(r.pageIds) : void 0,
+			re,
+			replacement: r.replacement
+		});
+	}
+	return compiled;
 };
 /**
-* Applies breakpoints to oversized segments.
+* Applies ordered regex replacements to page content (per page).
 *
-* Note: This is an internal engine used by `segmentPages()`.
+* - Replacement rules are applied in array order.
+* - Each rule is applied globally (flag `g` enforced) with unicode mode (flag `u` enforced).
+* - `pageIds` can scope a rule to specific pages. `pageIds: []` skips the rule entirely.
+*
+* This function is intentionally **pure**:
+* it returns a new pages array only when changes are needed, otherwise it returns the original pages.
 */
-const applyBreakpoints = (segments, pages, normalizedContent, maxPages, breakpoints, prefer, patternProcessor, logger, pageJoiner = "space") => {
-	const pageIds = pages.map((p) => p.id);
-	const pageIdToIndex = buildPageIdToIndexMap(pageIds);
-	const normalizedPages = buildNormalizedPagesMap(pages, normalizedContent);
-	const cumulativeOffsets = buildCumulativeOffsets(pageIds, normalizedPages);
-	const expandedBreakpoints = expandBreakpoints(breakpoints, patternProcessor);
-	const result = [];
-	logger?.info?.("Starting breakpoint processing", {
-		maxPages,
-		segmentCount: segments.length
-	});
-	logger?.debug?.("[breakpoints] inputSegments", {
-		segmentCount: segments.length,
-		segments: segments.map((s) => ({
-			contentLength: s.content.length,
-			from: s.from,
-			to: s.to
-		}))
-	});
-	for (const segment of segments) {
-		const fromIdx = pageIdToIndex.get(segment.from) ?? -1;
-		const toIdx = segment.to !== void 0 ? pageIdToIndex.get(segment.to) ?? fromIdx : fromIdx;
-		const segmentSpan = (segment.to ?? segment.from) - segment.from;
-		const hasExclusions = hasAnyExclusionsInRange(expandedBreakpoints, pageIds, fromIdx, toIdx);
-		if (segmentSpan <= maxPages && !hasExclusions) {
-			result.push(segment);
-			continue;
+const applyReplacements = (pages, rules) => {
+	if (!rules || rules.length === 0 || pages.length === 0) return pages;
+	const compiled = compileReplaceRules(rules);
+	if (compiled.length === 0) return pages;
+	return pages.map((p) => {
+		let content = p.content;
+		for (const rule of compiled) {
+			if (rule.pageIdSet && !rule.pageIdSet.has(p.id)) continue;
+			content = content.replace(rule.re, rule.replacement);
 		}
-		const broken = processOversizedSegment(segment, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, expandedBreakpoints, maxPages, prefer, logger);
-		result.push(...broken.map((s) => {
-			const segFromIdx = pageIdToIndex.get(s.from) ?? -1;
-			const segToIdx = s.to !== void 0 ? pageIdToIndex.get(s.to) ?? segFromIdx : segFromIdx;
-			if (segFromIdx >= 0 && segToIdx > segFromIdx) return {
-				...s,
-				content: applyPageJoinerBetweenPages(s.content, segFromIdx, segToIdx, pageIds, normalizedPages, pageJoiner)
-			};
-			return s;
-		}));
-	}
-	logger?.info?.("Breakpoint processing completed", { resultCount: result.length });
-	return result;
+		if (content === p.content) return p;
+		return {
+			...p,
+			content
+		};
+	});
 };
 
 //#endregion
-//#region src/segmentation/match-utils.ts
-/**
-* Utility functions for regex matching and result processing.
-*
-* These functions were extracted from `segmenter.ts` to reduce complexity
-* and enable independent testing. They handle match filtering, capture
-* extraction, and occurrence-based selection.
-*
-* @module match-utils
-*/
-/**
-* Extracts named capture groups from a regex match.
-*
-* Only includes groups that are in the `captureNames` list and have
-* defined values. This filters out positional captures and ensures
-* only explicitly requested named captures are returned.
-*
-* @param groups - The `match.groups` object from `RegExp.exec()`
-* @param captureNames - List of capture names to extract (from `{{token:name}}` syntax)
-* @returns Object with capture name → value pairs, or `undefined` if none found
+//#region src/segmentation/breakpoint-utils.ts
+const WINDOW_PREFIX_LENGTHS = [
+	80,
+	60,
+	40,
+	30,
+	20,
+	15
+];
+const JOINER_PREFIX_LENGTHS = [
+	80,
+	60,
+	40,
+	30,
+	20,
+	15,
+	12,
+	10,
+	8,
+	6
+];
+/**
+* Normalizes a breakpoint to the object form.
+* Strings are converted to { pattern: str } with no constraints.
 *
-* @example
-* const match = /(?<num>[٠-٩]+) -/.exec('٦٦٩٦ - text');
-* extractNamedCaptures(match.groups, ['num'])
-* // → { num: '٦٦٩٦' }
+* @param bp - Breakpoint as string or object
+* @returns Normalized BreakpointRule object
 *
 * @example
-* // No matching captures
-* extractNamedCaptures({}, ['num'])
-* // → undefined
+* normalizeBreakpoint('\\n\\n')
+* // → { pattern: '\\n\\n' }
 *
-* @example
-* // Undefined groups
-* extractNamedCaptures(undefined, ['num'])
-* // → undefined
+* normalizeBreakpoint({ pattern: '\\n', min: 10 })
+* // → { pattern: '\\n', min: 10 }
 */
-const extractNamedCaptures = (groups, captureNames) => {
-	if (!groups || captureNames.length === 0) return;
-	const namedCaptures = {};
-	for (const name of captureNames) if (groups[name] !== void 0) namedCaptures[name] = groups[name];
-	return Object.keys(namedCaptures).length > 0 ? namedCaptures : void 0;
-};
+const normalizeBreakpoint = (bp) => typeof bp === "string" ? { pattern: bp } : bp;
 /**
-* Gets the last defined positional capture group from a match array.
-*
-* Used for `lineStartsAfter` patterns where the content capture (`.*`)
-* is always at the end of the pattern. Named captures may shift the
-* positional indices, so we iterate backward to find the actual content.
+* Checks if a page ID is in an excluded list (single pages or ranges).
 *
-* @param match - RegExp exec result array
-* @returns The last defined capture group value, or `undefined` if none
+* @param pageId - Page ID to check
+* @param excludeList - List of page IDs or [from, to] ranges to exclude
+* @returns True if page is excluded
 *
 * @example
-* // Pattern: ^(?:(?<num>[٠-٩]+) - )(.*)
-* // Match array: ['٦٦٩٦ - content', '٦٦٩٦', 'content']
-* getLastPositionalCapture(match)
-* // → 'content'
+* isPageExcluded(5, [1, 5, 10])
+* // → true
 *
-* @example
-* // No captures
-* getLastPositionalCapture(['full match'])
-* // → undefined
+* isPageExcluded(5, [[3, 7]])
+* // → true
+*
+* isPageExcluded(5, [[10, 20]])
+* // → false
 */
-const getLastPositionalCapture = (match) => {
-	if (match.length <= 1) return;
-	for (let i = match.length - 1; i >= 1; i--) if (match[i] !== void 0) return match[i];
+const isPageExcluded = (pageId, excludeList) => {
+	if (!excludeList || excludeList.length === 0) return false;
+	for (const item of excludeList) if (typeof item === "number") {
+		if (pageId === item) return true;
+	} else {
+		const [from, to] = item;
+		if (pageId >= from && pageId <= to) return true;
+	}
+	return false;
 };
 /**
-* Filters matches to only include those within page ID constraints.
-*
-* Applies the `min`, `max`, and `exclude` constraints from a rule to filter out
-* matches that occur on pages outside the allowed range or explicitly excluded.
+* Checks if a page ID is within a breakpoint's min/max range and not excluded.
 *
-* @param matches - Array of match results to filter
-* @param rule - Rule containing `min`, `max`, and/or `exclude` page constraints
-* @param getId - Function that returns the page ID for a given offset
-* @returns Filtered array containing only matches within constraints
+* @param pageId - Page ID to check
+* @param rule - Breakpoint rule with optional min/max/exclude constraints
+* @returns True if page is within valid range
 *
 * @example
-* const matches = [
-*   { start: 0, end: 10 },   // Page 1
-*   { start: 100, end: 110 }, // Page 5
-*   { start: 200, end: 210 }, // Page 10
-* ];
-* filterByConstraints(matches, { min: 3, max: 8 }, getId)
-* // → [{ start: 100, end: 110 }] (only page 5 match)
+* isInBreakpointRange(50, { pattern: '\\n', min: 10, max: 100 })
+* // → true
+*
+* isInBreakpointRange(5, { pattern: '\\n', min: 10 })
+* // → false (below min)
 */
-const filterByConstraints = (matches, rule, getId) => {
-	return matches.filter((m) => {
-		const id = getId(m.start);
-		if (rule.min !== void 0 && id < rule.min) return false;
-		if (rule.max !== void 0 && id > rule.max) return false;
-		if (isPageExcluded(id, rule.exclude)) return false;
-		return true;
-	});
+const isInBreakpointRange = (pageId, rule) => {
+	if (rule.min !== void 0 && pageId < rule.min) return false;
+	if (rule.max !== void 0 && pageId > rule.max) return false;
+	return !isPageExcluded(pageId, rule.exclude);
 };
 /**
-* Checks if any rule in the list allows the given page ID.
-*
-* A rule allows an ID if it falls within the rule's `min`/`max` constraints.
-* Rules without constraints allow all page IDs.
+* Builds an exclude set from a PageRange array for O(1) lookups.
 *
-* This is used to determine whether to create a segment for content
-* that appears before any split points (the "first segment").
+* @param excludeList - List of page IDs or [from, to] ranges
+* @returns Set of all excluded page IDs
 *
-* @param rules - Array of rules with optional `min` and `max` constraints
-* @param pageId - Page ID to check
-* @returns `true` if at least one rule allows the page ID
+* @remarks
+* This expands ranges into explicit page IDs for fast membership checks. For typical
+* book-scale inputs (thousands of pages), this is small and keeps downstream logic
+* simple and fast. If you expect extremely large ranges (e.g., millions of pages),
+* consider avoiding broad excludes or introducing a range-based membership structure.
 *
 * @example
-* const rules = [
-*   { min: 5, max: 10 },  // Allows pages 5-10
-*   { min: 20 },          // Allows pages 20+
-* ];
+* buildExcludeSet([1, 5, [10, 12]])
+* // → Set { 1, 5, 10, 11, 12 }
+*/
+const buildExcludeSet = (excludeList) => {
+	const excludeSet = /* @__PURE__ */ new Set();
+	for (const item of excludeList || []) if (typeof item === "number") excludeSet.add(item);
+	else for (let i = item[0]; i <= item[1]; i++) excludeSet.add(i);
+	return excludeSet;
+};
+/**
+* Creates a segment with optional to and meta fields.
+* Returns null if content is empty after trimming.
 *
-* anyRuleAllowsId(rules, 7)   // → true (first rule allows)
-* anyRuleAllowsId(rules, 3)   // → false (no rule allows)
-* anyRuleAllowsId(rules, 25)  // → true (second rule allows)
+* @param content - Segment content
+* @param fromPageId - Starting page ID
+* @param toPageId - Optional ending page ID (omitted if same as from)
+* @param meta - Optional metadata to attach
+* @returns Segment object or null if empty
 *
 * @example
-* // Rules without constraints allow everything
-* anyRuleAllowsId([{}], 999) // → true
+* createSegment('Hello world', 1, 3, { chapter: 1 })
+* // → { content: 'Hello world', from: 1, to: 3, meta: { chapter: 1 } }
+*
+* createSegment('   ', 1, undefined, undefined)
+* // → null (empty content)
 */
-const anyRuleAllowsId = (rules, pageId) => {
-	return rules.some((r) => {
-		const minOk = r.min === void 0 || pageId >= r.min;
-		const maxOk = r.max === void 0 || pageId <= r.max;
-		return minOk && maxOk;
-	});
-};
-
-//#endregion
-//#region src/segmentation/replace.ts
-const DEFAULT_REPLACE_FLAGS = "gu";
-const normalizeReplaceFlags = (flags) => {
-	if (!flags) return DEFAULT_REPLACE_FLAGS;
-	const allowed = new Set([
-		"g",
-		"i",
-		"m",
-		"s",
-		"u",
-		"y"
-	]);
-	const set = /* @__PURE__ */ new Set();
-	for (const ch of flags) {
-		if (!allowed.has(ch)) throw new Error(`Invalid replace regex flag: "${ch}" (allowed: gimsyu)`);
-		set.add(ch);
-	}
-	set.add("g");
-	set.add("u");
-	return [
-		"g",
-		"i",
-		"m",
-		"s",
-		"y",
-		"u"
-	].filter((c) => set.has(c)).join("");
-};
-const compileReplaceRules = (rules) => {
-	const compiled = [];
-	for (const r of rules) {
-		if (r.pageIds && r.pageIds.length === 0) continue;
-		const flags = normalizeReplaceFlags(r.flags);
-		const re = new RegExp(r.regex, flags);
-		compiled.push({
-			pageIdSet: r.pageIds ? new Set(r.pageIds) : void 0,
-			re,
-			replacement: r.replacement
-		});
-	}
-	return compiled;
+const createSegment = (content, fromPageId, toPageId, meta) => {
+	const trimmed = content.trim();
+	if (!trimmed) return null;
+	const seg = {
+		content: trimmed,
+		from: fromPageId
+	};
+	if (toPageId !== void 0 && toPageId !== fromPageId) seg.to = toPageId;
+	if (meta) seg.meta = meta;
+	return seg;
 };
 /**
-* Applies ordered regex replacements to page content (per page).
+* Expands breakpoint patterns and pre-computes exclude sets.
 *
-* - Replacement rules are applied in array order.
-* - Each rule is applied globally (flag `g` enforced) with unicode mode (flag `u` enforced).
-* - `pageIds` can scope a rule to specific pages. `pageIds: []` skips the rule entirely.
+* @param breakpoints - Array of breakpoint patterns or rules
+* @param processPattern - Function to expand tokens in patterns
+* @returns Array of expanded breakpoints with compiled regexes
 *
-* This function is intentionally **pure**:
-* it returns a new pages array only when changes are needed, otherwise it returns the original pages.
+* @remarks
+* This function compiles regex patterns dynamically. This can be a ReDoS vector
+* if patterns come from untrusted sources. In typical usage, breakpoint rules
+* are application configuration, not user input.
 */
-const applyReplacements = (pages, rules) => {
-	if (!rules || rules.length === 0 || pages.length === 0) return pages;
-	const compiled = compileReplaceRules(rules);
-	if (compiled.length === 0) return pages;
-	return pages.map((p) => {
-		let content = p.content;
-		for (const rule of compiled) {
-			if (rule.pageIdSet && !rule.pageIdSet.has(p.id)) continue;
-			content = content.replace(rule.re, rule.replacement);
-		}
-		if (content === p.content) return p;
+const expandBreakpoints = (breakpoints, processPattern$1) => breakpoints.map((bp) => {
+	const rule = normalizeBreakpoint(bp);
+	const excludeSet = buildExcludeSet(rule.exclude);
+	const skipWhenRegex = rule.skipWhen !== void 0 ? (() => {
+		const expandedSkip = processPattern$1(rule.skipWhen);
+		try {
+			return new RegExp(expandedSkip, "mu");
+		} catch (error) {
+			const message = error instanceof Error ? error.message : String(error);
+			throw new Error(`Invalid breakpoint skipWhen regex: ${rule.skipWhen}\n  Cause: ${message}`);
+		}
+	})() : null;
+	if (rule.pattern === "") return {
+		excludeSet,
+		regex: null,
+		rule,
+		skipWhenRegex
+	};
+	const expanded = processPattern$1(rule.pattern);
+	try {
 		return {
-			...p,
-			content
+			excludeSet,
+			regex: new RegExp(expanded, "gmu"),
+			rule,
+			skipWhenRegex
 		};
-	});
-};
-
-//#endregion
-//#region src/segmentation/tokens.ts
+	} catch (error) {
+		const message = error instanceof Error ? error.message : String(error);
+		throw new Error(`Invalid breakpoint regex: ${rule.pattern}\n  Cause: ${message}`);
+	}
+});
 /**
-* Token-based template system for Arabic text pattern matching.
+* Applies a configured joiner at detected page boundaries within a multi-page content chunk.
 *
-* This module provides a human-readable way to define regex patterns using
-* `{{token}}` placeholders that expand to their regex equivalents. It supports
-* named capture groups for extracting matched values into metadata.
+* This is used for breakpoint-generated segments which don't have access to the original
+* `pageMap.pageBreaks` offsets. We detect page starts sequentially by searching for each page's
+* prefix after the previous boundary, then replace ONLY the single newline immediately before
+* that page start.
 *
-* @module tokens
+* This avoids converting real in-page newlines, while still normalizing page joins consistently.
+*/
+const applyPageJoinerBetweenPages = (content, fromIdx, toIdx, pageIds, normalizedPages, joiner) => {
+	if (joiner === "newline" || fromIdx >= toIdx || !content.includes("\n")) return content;
+	let updated = content;
+	let searchFrom = 0;
+	for (let pi = fromIdx + 1; pi <= toIdx; pi++) {
+		const pageData = normalizedPages.get(pageIds[pi]);
+		if (!pageData) continue;
+		const found = findPrefixPositionInContent(updated, pageData.content.trimStart(), searchFrom);
+		if (found > 0 && updated[found - 1] === "\n") updated = `${updated.slice(0, found - 1)} ${updated.slice(found)}`;
+		if (found > 0) searchFrom = found;
+	}
+	return updated;
+};
+/**
+* Finds the position of a page prefix in content, trying multiple prefix lengths.
+*/
+const findPrefixPositionInContent = (content, trimmedPageContent, searchFrom) => {
+	for (const len of JOINER_PREFIX_LENGTHS) {
+		const prefix = trimmedPageContent.slice(0, Math.min(len, trimmedPageContent.length)).trim();
+		if (!prefix) continue;
+		const pos = content.indexOf(prefix, searchFrom);
+		if (pos > 0) return pos;
+	}
+	return -1;
+};
+/**
+* Estimates how far into the current page `remainingContent` begins.
 *
-* @example
-* // Simple token expansion
-* expandTokens('{{raqms}} {{dash}}')
-* // → '[\\u0660-\\u0669]+ [-–—ـ]'
+* During breakpoint processing, `remainingContent` can begin mid-page after a previous split.
+* When that happens, raw cumulative page offsets (computed from full page starts) can overestimate
+* expected boundary positions. This helper computes an approximate starting offset by matching
+* a short prefix of `remainingContent` inside the current page content.
+*/
+const estimateStartOffsetInCurrentPage = (remainingContent, currentFromIdx, pageIds, normalizedPages) => {
+	const currentPageData = normalizedPages.get(pageIds[currentFromIdx]);
+	if (!currentPageData) return 0;
+	const remStart = remainingContent.trimStart().slice(0, Math.min(60, remainingContent.length));
+	const needle = remStart.slice(0, Math.min(30, remStart.length));
+	if (!needle) return 0;
+	const idx = currentPageData.content.indexOf(needle);
+	return idx > 0 ? idx : 0;
+};
+/**
+* Attempts to find the start position of a target page within remainingContent,
+* anchored near an expected boundary position to reduce collisions.
 *
-* @example
-* // Named capture groups
-* expandTokensWithCaptures('{{raqms:num}} {{dash}}')
-* // → { pattern: '(?<num>[\\u0660-\\u0669]+) [-–—ـ]', captureNames: ['num'], hasCaptures: true }
+* This is used to define breakpoint windows in terms of actual content being split, rather than
+* raw per-page offsets which can desync when structural rules strip markers.
 */
+const findPageStartNearExpectedBoundary = (remainingContent, _currentFromIdx, targetPageIdx, expectedBoundary, pageIds, normalizedPages) => {
+	const targetPageData = normalizedPages.get(pageIds[targetPageIdx]);
+	if (!targetPageData) return -1;
+	const approx = Math.min(Math.max(0, expectedBoundary), remainingContent.length);
+	const searchStart = Math.max(0, approx - 1e4);
+	const searchEnd = Math.min(remainingContent.length, approx + 2e3);
+	const targetTrimmed = targetPageData.content.trimStart();
+	for (const len of WINDOW_PREFIX_LENGTHS) {
+		const prefix = targetTrimmed.slice(0, Math.min(len, targetTrimmed.length)).trim();
+		if (!prefix) continue;
+		let pos = remainingContent.indexOf(prefix, searchStart);
+		while (pos !== -1 && pos <= searchEnd) {
+			if (pos > 0 && /\s/.test(remainingContent[pos - 1] ?? "")) return pos;
+			pos = remainingContent.indexOf(prefix, pos + 1);
+		}
+		const last = remainingContent.lastIndexOf(prefix, approx);
+		if (last > 0) return last;
+	}
+	return -1;
+};
 /**
-* Token definitions mapping human-readable token names to regex patterns.
+* Builds a boundary position map for pages within the given range.
 *
-* Tokens are used in template strings with double-brace syntax:
-* - `{{token}}` - Expands to the pattern (non-capturing in context)
-* - `{{token:name}}` - Expands to a named capture group `(?<name>pattern)`
-* - `{{:name}}` - Captures any content with the given name `(?<name>.+)`
+* This function computes page boundaries once per segment and enables
+* O(log n) page lookups via binary search with `findPageIndexForPosition`.
 *
-* @remarks
-* These patterns are designed for Arabic text matching. For diacritic-insensitive
-* matching of Arabic patterns, use the `fuzzy: true` option in split rules,
-* which applies `makeDiacriticInsensitive()` to the expanded patterns.
+* Boundaries are derived from segmentContent (post-structural-rules).
+* When the segment starts mid-page, an offset correction is applied to
+* keep boundary estimates aligned with the segment's actual content space.
 *
-* @example
-* // Using tokens in a split rule
-* { lineStartsWith: ['{{kitab}}', '{{bab}}'], split: 'at', fuzzy: true }
+* @param segmentContent - Full segment content (already processed by structural rules)
+* @param fromIdx - Starting page index
+* @param toIdx - Ending page index
+* @param pageIds - Array of all page IDs
+* @param normalizedPages - Map of page ID to normalized content
+* @param cumulativeOffsets - Cumulative character offsets (for estimates)
+* @returns Array where boundaryPositions[i] = start position of page (fromIdx + i),
+*          with a sentinel boundary at segmentContent.length as the last element
 *
 * @example
-* // Using tokens with named captures
-* { lineStartsAfter: ['{{raqms:hadithNum}} {{dash}} '], split: 'at' }
+* // For a 3-page segment:
+* buildBoundaryPositions(content, 0, 2, pageIds, normalizedPages, offsets)
+* // → [0, 23, 45, 67] where 67 is content.length (sentinel)
 */
+const buildBoundaryPositions = (segmentContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets) => {
+	const boundaryPositions = [0];
+	const startOffsetInFromPage = estimateStartOffsetInCurrentPage(segmentContent, fromIdx, pageIds, normalizedPages);
+	for (let i = fromIdx + 1; i <= toIdx; i++) {
+		const expectedBoundary = cumulativeOffsets[i] !== void 0 && cumulativeOffsets[fromIdx] !== void 0 ? Math.max(0, cumulativeOffsets[i] - cumulativeOffsets[fromIdx] - startOffsetInFromPage) : segmentContent.length;
+		const pos = findPageStartNearExpectedBoundary(segmentContent, fromIdx, i, expectedBoundary, pageIds, normalizedPages);
+		const prevBoundary = boundaryPositions[boundaryPositions.length - 1];
+		if (pos > 0 && pos > prevBoundary && Math.abs(pos - expectedBoundary) < 2e3) boundaryPositions.push(pos);
+		else {
+			const estimate = Math.max(prevBoundary + 1, expectedBoundary);
+			boundaryPositions.push(Math.min(estimate, segmentContent.length));
+		}
+	}
+	boundaryPositions.push(segmentContent.length);
+	return boundaryPositions;
+};
 /**
-* 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.
+* Binary search to find which page a position falls within.
+* Uses "largest i where boundaryPositions[i] <= position" semantics.
 *
-* @param pattern - Template pattern that may contain `()[]` and `{{tokens}}`
-* @returns Pattern with `()[]` escaped outside of `{{...}}` delimiters
+* @param position - Character position in segmentContent
+* @param boundaryPositions - Precomputed boundary positions (from buildBoundaryPositions)
+* @param fromIdx - Base page index (boundaryPositions[0] corresponds to pageIds[fromIdx])
+* @returns Page index in pageIds array
 *
 * @example
-* escapeTemplateBrackets('({{harf}}): ')
-* // → '\\({{harf}}\\): '
+* // With boundaries [0, 20, 40, 60] and fromIdx=0:
+* findPageIndexForPosition(15, boundaries, 0) // → 0 (first page)
+* findPageIndexForPosition(25, boundaries, 0) // → 1 (second page)
+* findPageIndexForPosition(40, boundaries, 0) // → 2 (exactly on boundary = that page)
+*/
+const findPageIndexForPosition = (position, boundaryPositions, fromIdx) => {
+	if (boundaryPositions.length <= 1) return fromIdx;
+	let left = 0;
+	let right = boundaryPositions.length - 2;
+	while (left < right) {
+		const mid = Math.ceil((left + right) / 2);
+		if (boundaryPositions[mid] <= position) left = mid;
+		else right = mid - 1;
+	}
+	return fromIdx + left;
+};
+/**
+* Finds the end position of a breakpoint window inside `remainingContent`.
 *
-* @example
-* escapeTemplateBrackets('[{{raqm}}] ')
-* // → '\\[{{raqm}}\\] '
+* The window end is defined as the start of the page AFTER `windowEndIdx` (i.e. `windowEndIdx + 1`),
+* found within the actual `remainingContent` string being split. This avoids relying on raw page offsets
+* that can diverge when structural rules strip markers (e.g. `lineStartsAfter`).
+*/
+const findBreakpointWindowEndPosition = (remainingContent, currentFromIdx, windowEndIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets) => {
+	if (windowEndIdx >= toIdx) return remainingContent.length;
+	const desiredNextIdx = windowEndIdx + 1;
+	const minNextIdx = currentFromIdx + 1;
+	const maxNextIdx = Math.min(desiredNextIdx, toIdx);
+	const startOffsetInCurrentPage = estimateStartOffsetInCurrentPage(remainingContent, currentFromIdx, pageIds, normalizedPages);
+	for (let nextIdx = maxNextIdx; nextIdx >= minNextIdx; nextIdx--) {
+		const expectedBoundary = cumulativeOffsets[nextIdx] !== void 0 && cumulativeOffsets[currentFromIdx] !== void 0 ? Math.max(0, cumulativeOffsets[nextIdx] - cumulativeOffsets[currentFromIdx] - startOffsetInCurrentPage) : remainingContent.length;
+		const pos = findPageStartNearExpectedBoundary(remainingContent, currentFromIdx, nextIdx, expectedBoundary, pageIds, normalizedPages);
+		if (pos > 0) return pos;
+	}
+	return remainingContent.length;
+};
+/**
+* Finds exclusion-based break position using raw cumulative offsets.
 *
-* @example
-* escapeTemplateBrackets('{{harf}}')
-* // → '{{harf}}' (unchanged - no brackets outside tokens)
+* This is used to ensure pages excluded by breakpoints are never merged into the same output segment.
+* Returns a break position relative to the start of `remainingContent` (i.e. the currentFromIdx start).
 */
-const escapeTemplateBrackets = (pattern) => {
-	return pattern.replace(/(\{\{[^}]*\}\})|([()[\]])/g, (_match, token, bracket) => {
-		if (token) return token;
-		return `\\${bracket}`;
-	});
-};
-const RUMUZ_ATOM = `(?:${[
-	"خت",
-	"خغ",
-	"بخ",
-	"عخ",
-	"مق",
-	"مت",
-	"عس",
-	"سي",
-	"سن",
-	"كن",
-	"مد",
-	"قد",
-	"خد",
-	"فد",
-	"دل",
-	"كد",
-	"غد",
-	"صد",
-	"دت",
-	"دس",
-	"تم",
-	"فق",
-	"دق",
-	"[خرزيمنصسدفلتقع]",
-	"(?<![\\u0660-\\u0669])٤(?![\\u0660-\\u0669])"
-].join("|")})`;
-const RUMUZ_BLOCK = `${RUMUZ_ATOM}(?:\\s+${RUMUZ_ATOM})*`;
-const BASE_TOKENS = {
-	bab: "باب",
-	basmalah: ["بسم الله", "﷽"].join("|"),
-	bullet: "[•*°]",
-	dash: "[-–—ـ]",
-	fasl: ["مسألة", "فصل"].join("|"),
-	harf: "[أ-ي]",
-	harfs: "[أ-ي](?:\\s+[أ-ي])*",
-	kitab: "كتاب",
-	naql: [
-		"حدثني",
-		"وأخبرنا",
-		"حدثنا",
-		"سمعت",
-		"أنبأنا",
-		"وحدثنا",
-		"أخبرنا",
-		"وحدثني",
-		"وحدثنيه"
-	].join("|"),
-	raqm: "[\\u0660-\\u0669]",
-	raqms: "[\\u0660-\\u0669]+",
-	rumuz: RUMUZ_BLOCK,
-	tarqim: "[.!?؟؛]"
+const findExclusionBreakPosition = (currentFromIdx, windowEndIdx, toIdx, pageIds, expandedBreakpoints, cumulativeOffsets) => {
+	const startingPageId = pageIds[currentFromIdx];
+	if (expandedBreakpoints.some((bp) => bp.excludeSet.has(startingPageId)) && currentFromIdx < toIdx) return cumulativeOffsets[currentFromIdx + 1] - cumulativeOffsets[currentFromIdx];
+	for (let pageIdx = currentFromIdx + 1; pageIdx <= windowEndIdx; pageIdx++) {
+		const pageId = pageIds[pageIdx];
+		if (expandedBreakpoints.some((bp) => bp.excludeSet.has(pageId))) return cumulativeOffsets[pageIdx] - cumulativeOffsets[currentFromIdx];
+	}
+	return -1;
 };
 /**
-* Composite token definitions using template syntax.
-*
-* These tokens reference base tokens using `{{token}}` syntax and are
-* automatically expanded to their final regex patterns at module load time.
-*
-* This provides better abstraction - if base tokens change, composites
-* automatically update on the next build.
+* Checks if any page in a range is excluded by the given exclude set.
 *
-* @internal
+* @param excludeSet - Set of excluded page IDs
+* @param pageIds - Array of page IDs
+* @param fromIdx - Start index (inclusive)
+* @param toIdx - End index (inclusive)
+* @returns True if any page in range is excluded
 */
-const COMPOSITE_TOKENS = { numbered: "{{raqms}} {{dash}} " };
+const hasExcludedPageInRange = (excludeSet, pageIds, fromIdx, toIdx) => {
+	if (excludeSet.size === 0) return false;
+	for (let pageIdx = fromIdx; pageIdx <= toIdx; pageIdx++) if (excludeSet.has(pageIds[pageIdx])) return true;
+	return false;
+};
 /**
-* Expands any *composite* tokens (like `{{numbered}}`) into their underlying template form
-* (like `{{raqms}} {{dash}} `).
-*
-* This is useful when you want to take a signature produced by `analyzeCommonLineStarts()`
-* and turn it into an editable template where you can add named captures, e.g.:
+* Finds the position of the next page content within remaining content.
+* Returns -1 if not found.
 *
-* - `{{numbered}}` → `{{raqms}} {{dash}} `
-* - then: `{{raqms:num}} {{dash}} ` to capture the number
+* @param remainingContent - Content to search in
+* @param nextPageData - Normalized data for the next page
+* @returns Position of next page content, or -1 if not found
+*/
+const findNextPagePosition = (remainingContent, nextPageData) => {
+	const searchPrefix = nextPageData.content.trim().slice(0, Math.min(30, nextPageData.length));
+	if (searchPrefix.length === 0) return -1;
+	const pos = remainingContent.indexOf(searchPrefix);
+	return pos > 0 ? pos : -1;
+};
+/**
+* Finds matches within a window and returns the selected position based on preference.
 *
-* Notes:
-* - This only expands the plain `{{token}}` form (not `{{token:name}}`).
-* - Expansion is repeated a few times to support nested composites.
+* @param windowContent - Content to search
+* @param regex - Regex to match
+* @param prefer - 'longer' for last match, 'shorter' for first match
+* @returns Break position after the selected match, or -1 if no matches
 */
-const expandCompositeTokensInTemplate = (template) => {
-	let out = template;
-	for (let i = 0; i < 10; i++) {
-		const next = out.replace(/\{\{(\w+)\}\}/g, (m, tokenName) => {
-			return COMPOSITE_TOKENS[tokenName] ?? m;
-		});
-		if (next === out) break;
-		out = next;
+const findPatternBreakPosition = (windowContent, regex, prefer) => {
+	let first;
+	let last;
+	for (const m of windowContent.matchAll(regex)) {
+		const match = {
+			index: m.index,
+			length: m[0].length
+		};
+		if (!first) first = match;
+		last = match;
 	}
-	return out;
+	if (!first) return -1;
+	const selected = prefer === "longer" ? last : first;
+	return selected.index + selected.length;
 };
 /**
-* Expands base tokens in a template string.
-* Used internally to pre-expand composite tokens.
-*
-* @param template - Template string with `{{token}}` placeholders
-* @returns Expanded pattern with base tokens replaced
-* @internal
+* Handles page boundary breakpoint (empty pattern).
+* Returns break position or -1 if no valid position found.
 */
-const expandBaseTokens = (template) => {
-	return template.replace(/\{\{(\w+)\}\}/g, (_, tokenName) => {
-		return BASE_TOKENS[tokenName] ?? `{{${tokenName}}}`;
-	});
+const handlePageBoundaryBreak = (remainingContent, windowEndIdx, windowEndPosition, toIdx, pageIds, normalizedPages) => {
+	const nextPageIdx = windowEndIdx + 1;
+	if (nextPageIdx <= toIdx) {
+		const nextPageData = normalizedPages.get(pageIds[nextPageIdx]);
+		if (nextPageData) {
+			const pos = findNextPagePosition(remainingContent, nextPageData);
+			if (pos > 0) return Math.min(pos, windowEndPosition, remainingContent.length);
+		}
+	}
+	return Math.min(windowEndPosition, remainingContent.length);
 };
 /**
-* Token definitions mapping human-readable token names to regex patterns.
-*
-* Tokens are used in template strings with double-brace syntax:
-* - `{{token}}` - Expands to the pattern (non-capturing in context)
-* - `{{token:name}}` - Expands to a named capture group `(?<name>pattern)`
-* - `{{:name}}` - Captures any content with the given name `(?<name>.+)`
-*
-* @remarks
-* These patterns are designed for Arabic text matching. For diacritic-insensitive
-* matching of Arabic patterns, use the `fuzzy: true` option in split rules,
-* which applies `makeDiacriticInsensitive()` to the expanded patterns.
-*
-* @example
-* // Using tokens in a split rule
-* { lineStartsWith: ['{{kitab}}', '{{bab}}'], split: 'at', fuzzy: true }
-*
-* @example
-* // Using tokens with named captures
-* { lineStartsAfter: ['{{raqms:hadithNum}} {{dash}} '], split: 'at' }
+* Tries to find a break position within the current window using breakpoint patterns.
+* Returns the break position or -1 if no suitable break was found.
 *
-* @example
-* // Using the numbered convenience token
-* { lineStartsAfter: ['{{numbered}}'], split: 'at' }
+* @param remainingContent - Content remaining to be segmented
+* @param currentFromIdx - Current starting page index
+* @param toIdx - Ending page index
+* @param windowEndIdx - Maximum window end index
+* @param ctx - Breakpoint context with page data and patterns
+* @returns Break position in the content, or -1 if no break found
 */
-const TOKEN_PATTERNS = {
-	...BASE_TOKENS,
-	...Object.fromEntries(Object.entries(COMPOSITE_TOKENS).map(([k, v]) => [k, expandBaseTokens(v)]))
+const findBreakPosition = (remainingContent, currentFromIdx, toIdx, windowEndIdx, windowEndPosition, ctx) => {
+	const { pageIds, normalizedPages, expandedBreakpoints, prefer } = ctx;
+	for (const { rule, regex, excludeSet, skipWhenRegex } of expandedBreakpoints) {
+		if (!isInBreakpointRange(pageIds[currentFromIdx], rule)) continue;
+		if (hasExcludedPageInRange(excludeSet, pageIds, currentFromIdx, windowEndIdx)) continue;
+		if (skipWhenRegex?.test(remainingContent)) continue;
+		if (regex === null) return handlePageBoundaryBreak(remainingContent, windowEndIdx, windowEndPosition, toIdx, pageIds, normalizedPages);
+		const breakPos = findPatternBreakPosition(remainingContent.slice(0, Math.min(windowEndPosition, remainingContent.length)), regex, prefer);
+		if (breakPos > 0) return breakPos;
+	}
+	return -1;
 };
+
+//#endregion
+//#region src/segmentation/breakpoint-processor.ts
 /**
-* Regex pattern for matching tokens with optional named capture syntax.
-*
-* Matches:
-* - `{{token}}` - Simple token (group 1 = token name, group 2 = empty)
-* - `{{token:name}}` - Token with capture (group 1 = token, group 2 = name)
-* - `{{:name}}` - Capture-only (group 1 = empty, group 2 = name)
+* Breakpoint post-processing engine extracted from segmenter.ts.
 *
-* @internal
+* This module is intentionally split into small helpers to reduce cognitive complexity
+* and allow unit testing of tricky edge cases (window sizing, next-page advancement, etc.).
 */
-const TOKEN_WITH_CAPTURE_REGEX = /\{\{(\w*):?(\w*)\}\}/g;
+const buildPageIdToIndexMap = (pageIds) => new Map(pageIds.map((id, i) => [id, i]));
+const buildNormalizedPagesMap = (pages, normalizedContent) => {
+	const normalizedPages = /* @__PURE__ */ new Map();
+	for (let i = 0; i < pages.length; i++) {
+		const content = normalizedContent[i];
+		normalizedPages.set(pages[i].id, {
+			content,
+			index: i,
+			length: content.length
+		});
+	}
+	return normalizedPages;
+};
+const buildCumulativeOffsets = (pageIds, normalizedPages) => {
+	const cumulativeOffsets = [0];
+	let totalOffset = 0;
+	for (let i = 0; i < pageIds.length; i++) {
+		const pageData = normalizedPages.get(pageIds[i]);
+		totalOffset += pageData ? pageData.length : 0;
+		if (i < pageIds.length - 1) totalOffset += 1;
+		cumulativeOffsets.push(totalOffset);
+	}
+	return cumulativeOffsets;
+};
+const hasAnyExclusionsInRange = (expandedBreakpoints, pageIds, fromIdx, toIdx) => expandedBreakpoints.some((bp) => hasExcludedPageInRange(bp.excludeSet, pageIds, fromIdx, toIdx));
+const computeWindowEndIdx = (currentFromIdx, toIdx, pageIds, maxPages) => {
+	const maxWindowPageId = pageIds[currentFromIdx] + maxPages;
+	let windowEndIdx = currentFromIdx;
+	for (let i = currentFromIdx; i <= toIdx; i++) if (pageIds[i] <= maxWindowPageId) windowEndIdx = i;
+	else break;
+	return windowEndIdx;
+};
+const computeRemainingSpan = (currentFromIdx, toIdx, pageIds) => pageIds[toIdx] - pageIds[currentFromIdx];
+const createFinalSegment = (remainingContent, currentFromIdx, toIdx, pageIds, meta, includeMeta) => createSegment(remainingContent, pageIds[currentFromIdx], currentFromIdx !== toIdx ? pageIds[toIdx] : void 0, includeMeta ? meta : void 0);
 /**
-* Regex pattern for simple token matching (no capture syntax).
-*
-* Matches only `{{token}}` format where token is one or more word characters.
-* Used by `containsTokens()` for quick detection.
+* Computes the actual start and end page indices for a piece using
+* precomputed boundary positions and binary search.
 *
-* @internal
+* @param pieceStartPos - Start position of the piece in the full segment content
+* @param pieceEndPos - End position (exclusive) of the piece
+* @param boundaryPositions - Precomputed boundary positions from buildBoundaryPositions
+* @param baseFromIdx - Base page index (boundaryPositions[0] corresponds to pageIds[baseFromIdx])
+* @param toIdx - Maximum page index
+* @returns Object with actualStartIdx and actualEndIdx
 */
-const SIMPLE_TOKEN_REGEX = /\{\{(\w+)\}\}/g;
+const computePiecePages = (pieceStartPos, pieceEndPos, boundaryPositions, baseFromIdx, toIdx) => {
+	const actualStartIdx = findPageIndexForPosition(pieceStartPos, boundaryPositions, baseFromIdx);
+	const endPos = Math.max(pieceStartPos, pieceEndPos - 1);
+	return {
+		actualEndIdx: Math.min(findPageIndexForPosition(endPos, boundaryPositions, baseFromIdx), toIdx),
+		actualStartIdx
+	};
+};
+const computeNextFromIdx = (remainingContent, actualEndIdx, toIdx, pageIds, normalizedPages) => {
+	let nextFromIdx = actualEndIdx;
+	if (remainingContent && actualEndIdx + 1 <= toIdx) {
+		const nextPageData = normalizedPages.get(pageIds[actualEndIdx + 1]);
+		if (nextPageData) {
+			const nextPrefix = nextPageData.content.slice(0, Math.min(30, nextPageData.length));
+			const remainingPrefix = remainingContent.trimStart().slice(0, Math.min(30, remainingContent.length));
+			if (nextPrefix && (remainingContent.startsWith(nextPrefix) || nextPageData.content.startsWith(remainingPrefix))) nextFromIdx = actualEndIdx + 1;
+		}
+	}
+	return nextFromIdx;
+};
+const createPieceSegment = (pieceContent, actualStartIdx, actualEndIdx, pageIds, meta, includeMeta) => createSegment(pieceContent, pageIds[actualStartIdx], actualEndIdx > actualStartIdx ? pageIds[actualEndIdx] : void 0, includeMeta ? meta : void 0);
 /**
-* Checks if a query string contains template tokens.
-*
-* Performs a quick test for `{{token}}` patterns without actually
-* expanding them. Useful for determining whether to apply token
-* expansion to a string.
-*
-* @param query - String to check for tokens
-* @returns `true` if the string contains at least one `{{token}}` pattern
+* Finds the break offset within a window, trying exclusions first, then patterns.
 *
-* @example
-* containsTokens('{{raqms}} {{dash}}') // → true
-* containsTokens('plain text')          // → false
-* containsTokens('[٠-٩]+ - ')           // → false (raw regex, no tokens)
+* @returns Break offset relative to remainingContent, or windowEndPosition as fallback
 */
-const containsTokens = (query) => {
-	SIMPLE_TOKEN_REGEX.lastIndex = 0;
-	return SIMPLE_TOKEN_REGEX.test(query);
-};
-const splitTemplateIntoSegments = (query) => {
-	const segments = [];
-	let lastIndex = 0;
-	TOKEN_WITH_CAPTURE_REGEX.lastIndex = 0;
-	let match;
-	while ((match = TOKEN_WITH_CAPTURE_REGEX.exec(query)) !== null) {
-		if (match.index > lastIndex) segments.push({
-			type: "text",
-			value: query.slice(lastIndex, match.index)
-		});
-		segments.push({
-			type: "token",
-			value: match[0]
-		});
-		lastIndex = match.index + match[0].length;
+const findBreakOffsetForWindow = (remainingContent, currentFromIdx, windowEndIdx, toIdx, windowEndPosition, pageIds, expandedBreakpoints, cumulativeOffsets, normalizedPages, prefer) => {
+	if (hasAnyExclusionsInRange(expandedBreakpoints, pageIds, currentFromIdx, windowEndIdx)) {
+		const exclusionBreak = findExclusionBreakPosition(currentFromIdx, windowEndIdx, toIdx, pageIds, expandedBreakpoints, cumulativeOffsets);
+		if (exclusionBreak > 0) return exclusionBreak;
 	}
-	if (lastIndex < query.length) segments.push({
-		type: "text",
-		value: query.slice(lastIndex)
+	const patternBreak = findBreakPosition(remainingContent, currentFromIdx, toIdx, windowEndIdx, windowEndPosition, {
+		expandedBreakpoints,
+		normalizedPages,
+		pageIds,
+		prefer
 	});
-	return segments;
-};
-const maybeApplyFuzzyToText = (text, fuzzyTransform) => {
-	if (fuzzyTransform && /[\u0600-\u06FF]/u.test(text)) return fuzzyTransform(text);
-	return text;
-};
-const maybeApplyFuzzyToTokenPattern = (tokenPattern, fuzzyTransform) => {
-	if (!fuzzyTransform) return tokenPattern;
-	return tokenPattern.split("|").map((part) => /[\u0600-\u06FF]/u.test(part) ? fuzzyTransform(part) : part).join("|");
+	return patternBreak > 0 ? patternBreak : windowEndPosition;
 };
-const parseTokenLiteral = (literal) => {
-	TOKEN_WITH_CAPTURE_REGEX.lastIndex = 0;
-	const tokenMatch = TOKEN_WITH_CAPTURE_REGEX.exec(literal);
-	if (!tokenMatch) return null;
-	const [, tokenName, captureName] = tokenMatch;
-	return {
-		captureName,
-		tokenName
-	};
+/**
+* Advances cursor position past any leading whitespace.
+*/
+const skipWhitespace = (content, startPos) => {
+	let pos = startPos;
+	while (pos < content.length && /\s/.test(content[pos])) pos++;
+	return pos;
 };
-const createCaptureRegistry = (capturePrefix) => {
-	const captureNames = [];
-	const captureNameCounts = /* @__PURE__ */ new Map();
-	const register = (baseName) => {
-		const count = captureNameCounts.get(baseName) ?? 0;
-		captureNameCounts.set(baseName, count + 1);
-		const uniqueName = count === 0 ? baseName : `${baseName}_${count + 1}`;
-		const prefixedName = capturePrefix ? `${capturePrefix}${uniqueName}` : uniqueName;
-		captureNames.push(prefixedName);
-		return prefixedName;
-	};
-	return {
-		captureNames,
-		register
-	};
+/**
+* Processes an oversized segment by iterating through the content and
+* breaking it into smaller pieces that fit within maxPages constraints.
+*
+* Uses precomputed boundary positions for O(log n) page attribution lookups.
+*/
+const processOversizedSegment = (segment, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, expandedBreakpoints, maxPages, prefer, logger) => {
+	const result = [];
+	const fullContent = segment.content;
+	let cursorPos = 0;
+	let currentFromIdx = fromIdx;
+	let isFirstPiece = true;
+	const boundaryPositions = buildBoundaryPositions(fullContent, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets);
+	logger?.debug?.("[breakpoints] boundaryPositions built", {
+		boundaryPositions,
+		fromIdx,
+		fullContentLength: fullContent.length,
+		toIdx
+	});
+	const maxIterations = 1e4;
+	for (let i = 0; i < maxIterations && cursorPos < fullContent.length && currentFromIdx <= toIdx; i++) {
+		const remainingContent = fullContent.slice(cursorPos);
+		if (!remainingContent.trim()) break;
+		const remainingSpan = computeRemainingSpan(currentFromIdx, toIdx, pageIds);
+		const remainingHasExclusions = hasAnyExclusionsInRange(expandedBreakpoints, pageIds, currentFromIdx, toIdx);
+		if (remainingSpan <= maxPages && !remainingHasExclusions) {
+			const finalSeg = createFinalSegment(remainingContent, currentFromIdx, toIdx, pageIds, segment.meta, isFirstPiece);
+			if (finalSeg) result.push(finalSeg);
+			break;
+		}
+		const windowEndIdx = computeWindowEndIdx(currentFromIdx, toIdx, pageIds, maxPages);
+		const windowEndPosition = findBreakpointWindowEndPosition(remainingContent, currentFromIdx, windowEndIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets);
+		logger?.debug?.(`[breakpoints] iteration=${i}`, {
+			currentFromIdx,
+			cursorPos,
+			windowEndIdx
+		});
+		const breakOffset = findBreakOffsetForWindow(remainingContent, currentFromIdx, windowEndIdx, toIdx, windowEndPosition, pageIds, expandedBreakpoints, cumulativeOffsets, normalizedPages, prefer);
+		const breakPos = cursorPos + breakOffset;
+		const pieceContent = fullContent.slice(cursorPos, breakPos).trim();
+		const { actualEndIdx, actualStartIdx } = computePiecePages(cursorPos, breakPos, boundaryPositions, fromIdx, toIdx);
+		logger?.trace?.("[breakpoints] piece", {
+			actualEndIdx,
+			actualStartIdx,
+			pieceLength: pieceContent.length
+		});
+		if (pieceContent) {
+			const pieceSeg = createPieceSegment(pieceContent, actualStartIdx, actualEndIdx, pageIds, segment.meta, isFirstPiece);
+			if (pieceSeg) result.push(pieceSeg);
+		}
+		cursorPos = skipWhitespace(fullContent, breakPos);
+		currentFromIdx = computeNextFromIdx(fullContent.slice(cursorPos), actualEndIdx, toIdx, pageIds, normalizedPages);
+		isFirstPiece = false;
+	}
+	logger?.debug?.("[breakpoints] done", { resultCount: result.length });
+	return result;
 };
-const expandTokenLiteral = (literal, opts) => {
-	const parsed = parseTokenLiteral(literal);
-	if (!parsed) return literal;
-	const { tokenName, captureName } = parsed;
-	if (!tokenName && captureName) return `(?<${opts.registerCapture(captureName)}>.+)`;
-	let tokenPattern = TOKEN_PATTERNS[tokenName];
-	if (!tokenPattern) return literal;
-	tokenPattern = maybeApplyFuzzyToTokenPattern(tokenPattern, opts.fuzzyTransform);
-	if (captureName) return `(?<${opts.registerCapture(captureName)}>${tokenPattern})`;
-	return tokenPattern;
+/**
+* Applies breakpoints to oversized segments.
+*
+* Note: This is an internal engine used by `segmentPages()`.
+*/
+const applyBreakpoints = (segments, pages, normalizedContent, maxPages, breakpoints, prefer, patternProcessor, logger, pageJoiner = "space") => {
+	const pageIds = pages.map((p) => p.id);
+	const pageIdToIndex = buildPageIdToIndexMap(pageIds);
+	const normalizedPages = buildNormalizedPagesMap(pages, normalizedContent);
+	const cumulativeOffsets = buildCumulativeOffsets(pageIds, normalizedPages);
+	const expandedBreakpoints = expandBreakpoints(breakpoints, patternProcessor);
+	const result = [];
+	logger?.info?.("Starting breakpoint processing", {
+		maxPages,
+		segmentCount: segments.length
+	});
+	logger?.debug?.("[breakpoints] inputSegments", {
+		segmentCount: segments.length,
+		segments: segments.map((s) => ({
+			contentLength: s.content.length,
+			from: s.from,
+			to: s.to
+		}))
+	});
+	for (const segment of segments) {
+		const fromIdx = pageIdToIndex.get(segment.from) ?? -1;
+		const toIdx = segment.to !== void 0 ? pageIdToIndex.get(segment.to) ?? fromIdx : fromIdx;
+		const segmentSpan = (segment.to ?? segment.from) - segment.from;
+		const hasExclusions = hasAnyExclusionsInRange(expandedBreakpoints, pageIds, fromIdx, toIdx);
+		if (segmentSpan <= maxPages && !hasExclusions) {
+			result.push(segment);
+			continue;
+		}
+		const broken = processOversizedSegment(segment, fromIdx, toIdx, pageIds, normalizedPages, cumulativeOffsets, expandedBreakpoints, maxPages, prefer, logger);
+		result.push(...broken.map((s) => {
+			const segFromIdx = pageIdToIndex.get(s.from) ?? -1;
+			const segToIdx = s.to !== void 0 ? pageIdToIndex.get(s.to) ?? segFromIdx : segFromIdx;
+			if (segFromIdx >= 0 && segToIdx > segFromIdx) return {
+				...s,
+				content: applyPageJoinerBetweenPages(s.content, segFromIdx, segToIdx, pageIds, normalizedPages, pageJoiner)
+			};
+			return s;
+		}));
+	}
+	logger?.info?.("Breakpoint processing completed", { resultCount: result.length });
+	return result;
 };
+
+//#endregion
+//#region src/segmentation/match-utils.ts
 /**
-* Expands template tokens with support for named captures.
+* Utility functions for regex matching and result processing.
 *
-* This is the primary token expansion function that handles all token syntax:
-* - `{{token}}` → Expands to the token's pattern (no capture group)
-* - `{{token:name}}` → Expands to `(?<name>pattern)` (named capture)
-* - `{{:name}}` → Expands to `(?<name>.+)` (capture anything)
+* These functions were extracted from `segmenter.ts` to reduce complexity
+* and enable independent testing. They handle match filtering, capture
+* extraction, and occurrence-based selection.
 *
-* Unknown tokens are left as-is in the output, allowing for partial templates.
+* @module match-utils
+*/
+/**
+* Extracts named capture groups from a regex match.
 *
-* @param query - The template string containing tokens
-* @param fuzzyTransform - Optional function to transform Arabic text for fuzzy matching.
-*                         Applied to both token patterns and plain Arabic text between tokens.
-*                         Typically `makeDiacriticInsensitive` from the fuzzy module.
-* @returns Object with expanded pattern, capture names, and capture flag
+* Only includes groups that are in the `captureNames` list and have
+* defined values. This filters out positional captures and ensures
+* only explicitly requested named captures are returned.
 *
-* @example
-* // Simple token expansion
-* expandTokensWithCaptures('{{raqms}} {{dash}}')
-* // → { pattern: '[\\u0660-\\u0669]+ [-–—ـ]', captureNames: [], hasCaptures: false }
+* @param groups - The `match.groups` object from `RegExp.exec()`
+* @param captureNames - List of capture names to extract (from `{{token:name}}` syntax)
+* @returns Object with capture name → value pairs, or `undefined` if none found
 *
 * @example
-* // Named capture
-* expandTokensWithCaptures('{{raqms:num}} {{dash}}')
-* // → { pattern: '(?<num>[\\u0660-\\u0669]+) [-–—ـ]', captureNames: ['num'], hasCaptures: true }
+* const match = /(?<num>[٠-٩]+) -/.exec('٦٦٩٦ - text');
+* extractNamedCaptures(match.groups, ['num'])
+* // → { num: '٦٦٩٦' }
 *
 * @example
-* // Capture-only token
-* expandTokensWithCaptures('{{raqms:num}} {{dash}} {{:content}}')
-* // → { pattern: '(?<num>[٠-٩]+) [-–—ـ] (?<content>.+)', captureNames: ['num', 'content'], hasCaptures: true }
+* // No matching captures
+* extractNamedCaptures({}, ['num'])
+* // → undefined
 *
 * @example
-* // With fuzzy transform
-* expandTokensWithCaptures('{{bab}}', makeDiacriticInsensitive)
-* // → { pattern: 'بَ?ا?بٌ?', captureNames: [], hasCaptures: false }
+* // Undefined groups
+* extractNamedCaptures(undefined, ['num'])
+* // → undefined
 */
-const expandTokensWithCaptures = (query, fuzzyTransform, capturePrefix) => {
-	const segments = splitTemplateIntoSegments(query);
-	const registry = createCaptureRegistry(capturePrefix);
-	const processedParts = segments.map((segment) => {
-		if (segment.type === "text") return maybeApplyFuzzyToText(segment.value, fuzzyTransform);
-		return expandTokenLiteral(segment.value, {
-			capturePrefix,
-			fuzzyTransform,
-			registerCapture: registry.register
-		});
-	});
-	return {
-		captureNames: registry.captureNames,
-		hasCaptures: registry.captureNames.length > 0,
-		pattern: processedParts.join("")
-	};
+const extractNamedCaptures = (groups, captureNames) => {
+	if (!groups || captureNames.length === 0) return;
+	const namedCaptures = {};
+	for (const name of captureNames) if (groups[name] !== void 0) namedCaptures[name] = groups[name];
+	return Object.keys(namedCaptures).length > 0 ? namedCaptures : void 0;
 };
 /**
-* Expands template tokens in a query string to their regex equivalents.
-*
-* This is the simple version without capture support. It returns only the
-* expanded pattern string, not capture metadata.
+* Gets the last defined positional capture group from a match array.
 *
-* Unknown tokens are left as-is, allowing for partial templates.
+* Used for `lineStartsAfter` patterns where the content capture (`.*`)
+* is always at the end of the pattern. Named captures may shift the
+* positional indices, so we iterate backward to find the actual content.
 *
-* @param query - Template string containing `{{token}}` placeholders
-* @returns Expanded regex pattern string
+* @param match - RegExp exec result array
+* @returns The last defined capture group value, or `undefined` if none
 *
 * @example
-* expandTokens('، {{raqms}}')     // → '، [\\u0660-\\u0669]+'
-* expandTokens('{{raqm}}*')       // → '[\\u0660-\\u0669]*'
-* expandTokens('{{dash}}{{raqm}}') // → '[-–—ـ][\\u0660-\\u0669]'
-* expandTokens('{{unknown}}')     // → '{{unknown}}' (left as-is)
+* // Pattern: ^(?:(?<num>[٠-٩]+) - )(.*)
+* // Match array: ['٦٦٩٦ - content', '٦٦٩٦', 'content']
+* getLastPositionalCapture(match)
+* // → 'content'
 *
-* @see expandTokensWithCaptures for full capture group support
+* @example
+* // No captures
+* getLastPositionalCapture(['full match'])
+* // → undefined
 */
-const expandTokens = (query) => expandTokensWithCaptures(query).pattern;
+const getLastPositionalCapture = (match) => {
+	if (match.length <= 1) return;
+	for (let i = match.length - 1; i >= 1; i--) if (match[i] !== void 0) return match[i];
+};
 /**
-* Converts a template string to a compiled RegExp.
-*
-* Expands all tokens and attempts to compile the result as a RegExp
-* with Unicode flag. Returns `null` if the resulting pattern is invalid.
+* Filters matches to only include those within page ID constraints.
 *
-* @remarks
-* This function dynamically compiles regular expressions from template strings.
-* If templates may come from untrusted sources, be aware of potential ReDoS
-* (Regular Expression Denial of Service) risks due to catastrophic backtracking.
-* Consider validating pattern complexity or applying execution timeouts when
-* running user-submitted patterns.
+* Applies the `min`, `max`, and `exclude` constraints from a rule to filter out
+* matches that occur on pages outside the allowed range or explicitly excluded.
 *
-* @param template - Template string containing `{{token}}` placeholders
-* @returns Compiled RegExp with 'u' flag, or `null` if invalid
+* @param matches - Array of match results to filter
+* @param rule - Rule containing `min`, `max`, and/or `exclude` page constraints
+* @param getId - Function that returns the page ID for a given offset
+* @returns Filtered array containing only matches within constraints
 *
 * @example
-* templateToRegex('، {{raqms}}')  // → /، [٠-٩]+/u
-* templateToRegex('{{raqms}}+')   // → /[٠-٩]++/u (might be invalid in some engines)
-* templateToRegex('(((')          // → null (invalid regex)
+* const matches = [
+*   { start: 0, end: 10 },   // Page 1
+*   { start: 100, end: 110 }, // Page 5
+*   { start: 200, end: 210 }, // Page 10
+* ];
+* filterByConstraints(matches, { min: 3, max: 8 }, getId)
+* // → [{ start: 100, end: 110 }] (only page 5 match)
 */
-const templateToRegex = (template) => {
-	const expanded = expandTokens(template);
-	try {
-		return new RegExp(expanded, "u");
-	} catch {
-		return null;
-	}
+const filterByConstraints = (matches, rule, getId) => {
+	return matches.filter((m) => {
+		const id = getId(m.start);
+		if (rule.min !== void 0 && id < rule.min) return false;
+		if (rule.max !== void 0 && id > rule.max) return false;
+		if (isPageExcluded(id, rule.exclude)) return false;
+		return true;
+	});
 };
 /**
-* Lists all available token names defined in `TOKEN_PATTERNS`.
+* Checks if any rule in the list allows the given page ID.
 *
-* Useful for documentation, validation, or building user interfaces
-* that show available tokens.
+* A rule allows an ID if it falls within the rule's `min`/`max` constraints.
+* Rules without constraints allow all page IDs.
 *
-* @returns Array of token names (e.g., `['bab', 'basmala', 'bullet', ...]`)
+* This is used to determine whether to create a segment for content
+* that appears before any split points (the "first segment").
 *
-* @example
-* getAvailableTokens()
-* // → ['bab', 'basmala', 'bullet', 'dash', 'harf', 'kitab', 'naql', 'raqm', 'raqms']
-*/
-const getAvailableTokens = () => Object.keys(TOKEN_PATTERNS);
-/**
-* Gets the regex pattern for a specific token name.
+* @param rules - Array of rules with optional `min` and `max` constraints
+* @param pageId - Page ID to check
+* @returns `true` if at least one rule allows the page ID
 *
-* Returns the raw pattern string as defined in `TOKEN_PATTERNS`,
-* without any expansion or capture group wrapping.
+* @example
+* const rules = [
+*   { min: 5, max: 10 },  // Allows pages 5-10
+*   { min: 20 },          // Allows pages 20+
+* ];
 *
-* @param tokenName - The token name to look up (e.g., 'raqms', 'dash')
-* @returns The regex pattern string, or `undefined` if token doesn't exist
+* anyRuleAllowsId(rules, 7)   // → true (first rule allows)
+* anyRuleAllowsId(rules, 3)   // → false (no rule allows)
+* anyRuleAllowsId(rules, 25)  // → true (second rule allows)
 *
 * @example
-* getTokenPattern('raqms')   // → '[\\u0660-\\u0669]+'
-* getTokenPattern('dash')    // → '[-–—ـ]'
-* getTokenPattern('unknown') // → undefined
+* // Rules without constraints allow everything
+* anyRuleAllowsId([{}], 999) // → true
 */
-const getTokenPattern = (tokenName) => TOKEN_PATTERNS[tokenName];
+const anyRuleAllowsId = (rules, pageId) => {
+	return rules.some((r) => {
+		const minOk = r.min === void 0 || pageId >= r.min;
+		const maxOk = r.max === void 0 || pageId <= r.max;
+		return minOk && maxOk;
+	});
+};
 
 //#endregion
 //#region src/segmentation/rule-regex.ts
@@ -1548,7 +1692,12 @@ const determineUsesCapture = (regexSource, _captureNames) => hasCapturingGroup(r
 */
 const buildRuleRegex = (rule, capturePrefix) => {
 	const s = { ...rule };
-	const fuzzy = rule.fuzzy ?? false;
+	const allPatterns = [
+		...s.lineStartsWith ?? [],
+		...s.lineStartsAfter ?? [],
+		...s.lineEndsWith ?? []
+	];
+	const fuzzy = rule.fuzzy ?? shouldDefaultToFuzzy(allPatterns);
 	let allCaptureNames = [];
 	if (s.lineStartsAfter?.length) {
 		const { regex, captureNames } = buildLineStartsAfterRegexSource(s.lineStartsAfter, fuzzy, capturePrefix);
@@ -1836,6 +1985,7 @@ const normalizeLineEndings = (content) => {
 *
 * @module segmenter
 */
+const MAX_REGEX_ITERATIONS = 1e5;
 /**
 * Builds a concatenated content string and page mapping from input pages.
 *
@@ -1943,9 +2093,18 @@ const ensureFallbackSegment = (segments, pages, normalizedContent, pageJoiner) =
 	if (lastPage.id !== firstPage.id) initialSeg.to = lastPage.id;
 	return [initialSeg];
 };
-const collectSplitPointsFromRules = (rules, matchContent, pageMap) => {
+const collectSplitPointsFromRules = (rules, matchContent, pageMap, logger) => {
+	logger?.debug?.("[segmenter] collecting split points from rules", {
+		contentLength: matchContent.length,
+		ruleCount: rules.length
+	});
 	const passesPageStartGuard = createPageStartGuardChecker(matchContent, pageMap);
 	const { combinableRules, fastFuzzyRules, standaloneRules } = partitionRulesForMatching(rules);
+	logger?.debug?.("[segmenter] rules partitioned", {
+		combinableCount: combinableRules.length,
+		fastFuzzyCount: fastFuzzyRules.length,
+		standaloneCount: standaloneRules.length
+	});
 	const splitPointsByRule = collectFastFuzzySplitPoints(matchContent, pageMap, fastFuzzyRules, passesPageStartGuard);
 	if (combinableRules.length > 0) {
 		const ruleRegexes = combinableRules.map(({ rule, prefix }) => {
@@ -1958,9 +2117,22 @@ const collectSplitPointsFromRules = (rules, matchContent, pageMap) => {
 		});
 		const combinedSource = ruleRegexes.map((r) => r.source).join("|");
 		const combinedRegex = new RegExp(combinedSource, "gm");
+		logger?.debug?.("[segmenter] combined regex built", {
+			combinableRuleCount: combinableRules.length,
+			combinedSourceLength: combinedSource.length
+		});
 		combinedRegex.lastIndex = 0;
 		let m = combinedRegex.exec(matchContent);
+		let iterationCount = 0;
 		while (m !== null) {
+			iterationCount++;
+			if (iterationCount > MAX_REGEX_ITERATIONS) throw new Error(`[segmenter] Possible infinite loop detected: regex matching exceeded ${MAX_REGEX_ITERATIONS} iterations. Last match at position ${m.index} (length ${m[0].length}). Check for patterns that may match empty strings or cause catastrophic backtracking.`);
+			if (iterationCount % 1e4 === 0) logger?.warn?.("[segmenter] high iteration count in regex loop", {
+				iterationCount,
+				lastIndex: combinedRegex.lastIndex,
+				matchLength: m[0].length,
+				matchPosition: m.index
+			});
 			const matchedRuleIndex = combinableRules.findIndex(({ prefix }) => m?.groups?.[prefix] !== void 0);
 			if (matchedRuleIndex !== -1) {
 				const { rule, prefix, index: originalIndex } = combinableRules[matchedRuleIndex];
@@ -1981,8 +2153,7 @@ const collectSplitPointsFromRules = (rules, matchContent, pageMap) => {
 				const start = m.index;
 				const end = m.index + m[0].length;
 				const pageId = pageMap.getId(start);
-				if ((rule.min === void 0 || pageId >= rule.min) && (rule.max === void 0 || pageId <= rule.max) && !isPageExcluded(pageId, rule.exclude)) {
-					if (!passesPageStartGuard(rule, originalIndex, start)) continue;
+				if ((rule.min === void 0 || pageId >= rule.min) && (rule.max === void 0 || pageId <= rule.max) && !isPageExcluded(pageId, rule.exclude) && passesPageStartGuard(rule, originalIndex, start)) {
 					const sp = {
 						capturedContent: void 0,
 						contentStartOffset,
@@ -2168,7 +2339,7 @@ const segmentPages = (pages, options) => {
 		pageIds: pageMap.pageIds,
 		totalContentLength: matchContent.length
 	});
-	const splitPoints = collectSplitPointsFromRules(rules, matchContent, pageMap);
+	const splitPoints = collectSplitPointsFromRules(rules, matchContent, pageMap, logger);
 	const unique = dedupeSplitPoints(splitPoints);
 	logger?.debug?.("[segmenter] split points collected", {
 		rawSplitPoints: splitPoints.length,
@@ -2660,5 +2831,5 @@ const analyzeTextForRule = (text) => {
 };
 
 //#endregion
-export { TOKEN_PATTERNS, analyzeCommonLineStarts, analyzeTextForRule, applyReplacements, containsTokens, detectTokenPatterns, escapeRegex, escapeTemplateBrackets, expandCompositeTokensInTemplate, expandTokens, expandTokensWithCaptures, generateTemplateFromText, getAvailableTokens, getTokenPattern, makeDiacriticInsensitive, segmentPages, suggestPatternConfig, templateToRegex };
+export { TOKEN_PATTERNS, analyzeCommonLineStarts, analyzeTextForRule, applyReplacements, containsTokens, detectTokenPatterns, escapeRegex, escapeTemplateBrackets, expandCompositeTokensInTemplate, expandTokens, expandTokensWithCaptures, generateTemplateFromText, getAvailableTokens, getTokenPattern, makeDiacriticInsensitive, segmentPages, suggestPatternConfig, templateToRegex, validateRules };
 //# sourceMappingURL=index.mjs.map