export type SolutionStatus = 'experiment' | 'candidate' | 'verified' | 'mature' | 'retired';
export type SolutionType = 'pattern' | 'solution' | 'decision' | 'troubleshoot' | 'anti-pattern' | 'convention';
export interface SolutionEvidence {
    injected: number;
    reflected: number;
    negative: number;
    sessions: number;
    reExtracted: number;
}
export interface SolutionFrontmatter {
    name: string;
    version: number;
    status: SolutionStatus;
    confidence: number;
    type: SolutionType;
    scope: 'me' | 'team' | 'project' | 'universal';
    tags: string[];
    identifiers: string[];
    evidence: SolutionEvidence;
    created: string;
    updated: string;
    supersedes: string | null;
    extractedBy: 'auto' | 'manual';
}
export interface SolutionV3 {
    frontmatter: SolutionFrontmatter;
    context: string;
    content: string;
    filePath?: string;
}
export interface SolutionIndexEntry {
    name: string;
    status: SolutionStatus;
    confidence: number;
    type: SolutionType;
    scope: 'me' | 'team' | 'project' | 'universal';
    tags: string[];
    /**
     * Pre-expanded tag set, computed at index build time via the term normalizer.
     * Contains `tags` plus every related match term from the canonical families
     * they belong to.
     *
     * T2 scope (current): populated as forward-looking metadata for T3's
     * ranking-decision log (which records raw + normalized query/solution
     * terms for offline explainability) and T4's BM25 term-frequency stats.
     * **NOT consumed by the current matcher** — `rankCandidates` and
     * `calculateRelevance` still use raw `tags` for intersection to preserve
     * the Round 3 baseline (bidirectional expansion would inflate recall 5-10×
     * and invalidate fixture metrics). A future PR that uses `normalizedTags`
     * in scoring must update `ROUND3_BASELINE` in the same commit.
     *
     * Not persisted — recomputed on every index build. Safe to regenerate whenever
     * `DEFAULT_MATCH_TERMS` changes.
     */
    normalizedTags: string[];
    identifiers: string[];
    filePath: string;
}
export declare const DEFAULT_EVIDENCE: SolutionEvidence;
export declare function slugify(text: string): string;
/** Runtime type guard for SolutionFrontmatter */
export declare function validateFrontmatter(fm: unknown): fm is SolutionFrontmatter;
/**
 * Return a list of validation errors for a parsed frontmatter object.
 *
 * Empty array = valid. Non-empty = each entry describes one missing/wrong
 * field. Callers that only need a boolean should use `validateFrontmatter`.
 * Slow path (quarantine logging) uses this to produce actionable diagnostics.
 */
export declare function diagnoseFrontmatter(fm: unknown): string[];
/** Parse YAML frontmatter from solution file content */
export declare function parseFrontmatterOnly(content: string): SolutionFrontmatter | null;
/** Parse a full V3 solution file into its components */
export declare function parseSolutionV3(content: string): SolutionV3 | null;
/** Serialize a SolutionV3 to a markdown string with YAML frontmatter */
export declare function serializeSolutionV3(solution: SolutionV3): string;
/** Check if content is in V1 format (# Title + > Type: pattern) */
export declare function isV1Format(content: string): boolean;
/** 한국어 일반 조사/어미 — strip 대상 (긴 것부터 매칭)
 *
 * term-matcher에서 재사용 가능하도록 export — 매칭 시점과 추출 시점의 stripping
 * 규칙을 단일 source of truth로 유지해 한국어 stem 비교 정합성 보장.
 *
 * 주의: 이 리스트는 **추출 시점에도 적용**되므로 1글자 suffix를 추가할 때
 * `집중`→`집`, `시도`→`시` 같은 한자어 명사가 깨지지 않도록 극도로 보수적으로
 * 유지한다. 동사 활용형(`리팩토링중`, `배포시`)처럼 매칭 전용 suffix가 필요하면
 * term-matcher의 `KO_VERBAL_SUFFIXES`에 따로 둔다.
 */
export declare const KO_SUFFIXES: string[];
export declare function stripKoSuffix(word: string): string;
/**
 * Extract tags from text.
 * Korean 2-char words preserved (e.g. "에러", "배포"), stopwords filtered.
 * English words require 3+ chars, stopwords filtered.
 * Tags capped at MAX_TAGS, ranked by frequency.
 *
 * NOTE on hyphens: this function strips `-` to a space (`api-key` query token
 * becomes `api` and `key` separately). Solution-side compound tags are
 * recovered downstream by `expandCompoundTags`, and query-side bigram
 * recovery is done by `expandQueryBigrams`. Both ship as part of R4-T1
 * (compound-tag tokenizer fix) — see `docs/plans/2026-04-08-t4-bm25-skip-adr.md`
 * "Round 4 candidates" section for the rationale. Changing this regex
 * directly was considered but rejected: it would silently shift the index
 * representation of every existing solution, requiring an index rebuild and
 * a fresh `ROUND3_BASELINE` measurement on every downstream PR.
 */
export declare function extractTags(text: string): string[];
/**
 * Expand a solution tag list with hyphen-split alternatives.
 *
 * Each input tag is preserved verbatim, and any tag containing `-` also
 * contributes its parts (length ≥ 3 each) as additional tags. The output
 * is deduplicated.
 *
 * Examples:
 *   - `['api-key', 'security']` → `['api-key', 'api', 'key', 'security']`
 *   - `['code-review', 'quality']` → `['code-review', 'code', 'review', 'quality']`
 *   - `['n+1', 'database']` → `['n+1', 'database']` (no hyphen, n+1 unchanged)
 *   - `['red-green-refactor']` → `['red-green-refactor', 'red', 'green', 'refactor']`
 *   - `['typescript']` → `['typescript']` (no hyphen, no expansion)
 *
 * Korean compound tags (`API에러`, `테스트주도개발`) are preserved verbatim
 * because they contain no `-`. The expansion is intentionally English-
 * compound-aware only — Korean compound recovery is not in scope for R4-T1
 * (the existing `term-normalizer` family expansion handles Korean ↔ English
 * cross-mapping).
 *
 * The output ordering is insertion order: original tags first, then split
 * parts in left-to-right order. Stable across runs (Set + Array dedup).
 */
export declare function expandCompoundTags(tags: readonly string[]): string[];
/**
 * Expand a query tag list with adjacent-token bigram alternatives.
 *
 * For each adjacent (a, b) pair where both tokens are length ≥ 3, the
 * function adds:
 *   - `a-b`   (hyphen-joined form, e.g. `api-key`)
 *   - `ab`    (concatenated form, e.g. `apikey`)
 *   - `a-b'`  (singular stem of b, only if b ends in `s` and length > 3)
 *   - `ab'`   (concatenated singular stem)
 *
 * Examples:
 *   - `['api', 'keys']` → `['api', 'keys', 'api-key', 'apikey', 'api-keys', 'apikeys']`
 *   - `['code', 'review']` → `['code', 'review', 'code-review', 'codereview']`
 *   - `['red', 'green', 'refactor']` → `[..., 'red-green', 'redgreen', 'green-refactor', 'greenrefactor']`
 *
 * Plural→singular stem is intentionally minimal: only `s`-suffix removal,
 * no `es`/`ies` handling. The cost-benefit is asymmetric — `apis → api`
 * is the highest-value case and is handled correctly; `classes → classe`
 * is wrong but doesn't matter because no solution tag is `classe`.
 *
 * Why both `-` and concatenated forms: solution tag conventions vary
 * across packs (`api-key` vs `apikey`), and this expansion is cheap.
 * The downstream intersection check is O(M) per solution where M = expanded
 * query tag count, so even doubling the query tag count is well within
 * the matcher's hot-path budget for the corpus sizes Forgen targets
 * (N ≤ 200 solutions).
 *
 * Korean tokens (`/[가-힣]/`) are passed through verbatim: bigram
 * concatenation of Korean compound words is meaningless because the
 * boundary is lexical, not whitespace-driven (`디버깅` is one word, not
 * two adjacent tokens). Only ASCII-letter pairs participate.
 */
export declare function expandQueryBigrams(tags: readonly string[]): string[];
export declare function expandQueryKoreanStems(tags: readonly string[]): string[];
/** Migrate a V1-format solution file to V3 format */
export declare function migrateV1toV3(content: string, filePath: string): string;