@nomideusz/svelte-search 0.1.3 → 0.2.0

package/CHANGELOG.md

@@ -1,5 +1,15 @@
 # Changelog
 
+## 0.2.0 — 2026-07-06
+
+### Fixed
+- **Fuzzy quality gate is now per-word** — typo queries against multi-word entity names could never pass the Levenshtein gate (the query was compared to the whole field, so `"triranta"` vs `"Triratna Warszawa - Buddyzm i medytacja Mokotów"` scored ~0.17 and was rejected). The gate now uses the best of whole-field and per-word similarity; false-positive protection is unchanged (`"inowroclaw"` still does not match `"wroclaw"`).
+
+### Added
+- **`ftsColumnWeights` engine option** (SQLite) — per-column bm25 weights for FTS ranking, in FTS table column order. Without it all columns weigh equally, so a term repeated in a long description column outranks an exact name match before the result limit is applied. Weight name-like columns high and description-like columns low, e.g. `ftsColumnWeights: [10, 4, 4, 4, 3, 2, 0.5]`. Opt-in; behavior is unchanged when unset.
+- **`bestWordSimilarity(query, field, locale?)`** exported from the package — max of whole-field and per-word normalized Levenshtein similarity.
+- Community health files: code of conduct, contributing guide, security policy, issue and PR templates.
+
 ## 0.1.3 — 2026-06-29
 
 ### Security

package/dist/core/engine.d.ts

@@ -19,6 +19,13 @@ export interface SearchEngineConfig<TResult extends SearchResult = SearchResult>
     qualityThreshold?: number;
     /** Max FTS terms per query (default: 6) */
     maxFtsTerms?: number;
+    /**
+     * Per-column bm25 weights for FTS ranking, in FTS table column order
+     * (SQLite only). Without this, all columns weigh equally — a term repeated
+     * in a long description outranks an exact name match. Weight name-like
+     * columns high and description-like columns low.
+     */
+    ftsColumnWeights?: number[];
 }
 export declare function createSearchEngine<TResult extends SearchResult = SearchResult>(config: SearchEngineConfig<TResult>): {
     search: (params: SearchParams) => Promise<SearchResponse<TResult>>;

package/dist/core/engine.js

@@ -12,7 +12,7 @@
 // Dialect support:
 //   - 'sqlite' (default): FTS5 MATCH, custom trigram tables, rowid joins
 //   - 'postgres': tsvector/tsquery @@, pg_trgm similarity(), PK joins
-import { normalize, trigrams, trigramSimilarity, levenshteinSimilarity, hasGeoIntent, stripGeoIntent } from './normalize.js';
+import { normalize, trigrams, trigramSimilarity, levenshteinSimilarity, bestWordSimilarity, hasGeoIntent, stripGeoIntent } from './normalize.js';
 import { haversineKm, walkingMinutes, boundingBox } from './geo.js';
 // ── Query timeout helper ───────────────────────────────────
 function withTimeout(promise, ms, fallback) {
@@ -36,7 +36,7 @@ function ph(index, dialect) {
 }
 // ── Create search engine ───────────────────────────────────
 export function createSearchEngine(config) {
-    const { db, adapter, locale, dialect = 'sqlite', ftsTimeoutMs = 5000, fuzzyTimeoutMs = 3000, primaryRadiusKm = 15, nearbyRadiusKm = 30, maxNearby = 5, qualityThreshold = 0.75, maxFtsTerms = 6, } = config;
+    const { db, adapter, locale, dialect = 'sqlite', ftsTimeoutMs = 5000, fuzzyTimeoutMs = 3000, primaryRadiusKm = 15, nearbyRadiusKm = 30, maxNearby = 5, qualityThreshold = 0.75, maxFtsTerms = 6, ftsColumnWeights, } = config;
     const { tables, columns, trigramColumns } = adapter;
     // ── Main search ────────────────────────────────────────
     async function search(params) {
@@ -72,11 +72,13 @@ export function createSearchEngine(config) {
         // Merge, score, and rank
         const merged = deduplicateById([...ftsResults, ...fuzzyResults]);
         const scored = scoreResults(merged, normalized, lat, lng, geoIntent);
-        // Quality gate — fuzzy-only results need high Levenshtein similarity
+        // Quality gate — fuzzy-only results need high Levenshtein similarity.
+        // Per-word: a typo of one word in a multi-word name ("triranta" for
+        // "Triratna Warszawa ...") must qualify against the word, not the field.
         const qualified = scored.filter(r => {
             if (r._hasFts)
                 return true;
-            const best = Math.max(levenshteinSimilarity(normalized, r._nameN || '', locale), levenshteinSimilarity(normalized, r._locationN || '', locale), levenshteinSimilarity(normalized, r._categoriesN || '', locale));
+            const best = Math.max(bestWordSimilarity(normalized, r._nameN || '', locale), bestWordSimilarity(normalized, r._locationN || '', locale), bestWordSimilarity(normalized, r._categoriesN || '', locale));
             return best >= qualityThreshold;
         });
         qualified.sort((a, b) => b.score - a.score);
@@ -128,8 +130,12 @@ export function createSearchEngine(config) {
         return ftsSearchSqlite(ftsQuery, locationSlug, categorySlug, limit);
     }
     async function ftsSearchSqlite(ftsQuery, locationSlug, categorySlug, limit) {
+        // Weighted bm25 when configured — same semantics as rank (lower = better)
+        const rankExpr = ftsColumnWeights?.length
+            ? `bm25(${tables.fts}, ${ftsColumnWeights.join(', ')})`
+            : 'fts.rank';
         let sql = `
-      SELECT s.*, fts.rank AS _ftsRank
+      SELECT s.*, ${rankExpr} AS _ftsRank
       FROM ${tables.fts} fts
       JOIN ${tables.entities} s ON s.rowid = fts.rowid
       WHERE ${tables.fts} MATCH ?
@@ -144,7 +150,7 @@ export function createSearchEngine(config) {
             sql += ` AND s.${columns.categoriesNormalized} LIKE ?`;
             args.push(`%${catName}%`);
         }
-        sql += ' ORDER BY fts.rank LIMIT ?';
+        sql += ' ORDER BY _ftsRank LIMIT ?';
         args.push(limit);
         const result = await db.execute({ sql, args });
         return result.rows;

package/dist/core/normalize.d.ts

@@ -11,6 +11,13 @@ export declare function trigramSimilarity(a: string, b: string, locale?: SearchL
 export declare function levenshtein(a: string, b: string): number;
 /** Normalized Levenshtein similarity. 0..1, higher = more similar. */
 export declare function levenshteinSimilarity(a: string, b: string, locale?: SearchLocale): number;
+/**
+ * Best Levenshtein similarity of `query` against a multi-word field: the max
+ * of whole-field similarity and per-word similarity. A typo of one word in a
+ * long field ("triranta" vs "triratna warszawa buddyzm...") scores near 0
+ * against the whole field but 0.75 against the word it matches.
+ */
+export declare function bestWordSimilarity(query: string, field: string, locale?: SearchLocale): number;
 /** Detect postcode format: XX-XXX or XXXXX (Polish default, override via locale). */
 export declare function isPostcode(text: string): boolean;
 /** Detect "near me" intent using locale geo patterns. */

package/dist/core/normalize.js

@@ -75,6 +75,22 @@ export function levenshteinSimilarity(a, b, locale) {
     const max = Math.max(na.length, nb.length);
     return max === 0 ? 1 : 1 - levenshtein(na, nb) / max;
 }
+/**
+ * Best Levenshtein similarity of `query` against a multi-word field: the max
+ * of whole-field similarity and per-word similarity. A typo of one word in a
+ * long field ("triranta" vs "triratna warszawa buddyzm...") scores near 0
+ * against the whole field but 0.75 against the word it matches.
+ */
+export function bestWordSimilarity(query, field, locale) {
+    if (!field)
+        return 0;
+    let best = levenshteinSimilarity(query, field, locale);
+    for (const w of normalize(field, locale).split(' ')) {
+        if (w)
+            best = Math.max(best, levenshteinSimilarity(query, w, locale));
+    }
+    return best;
+}
 /** Detect postcode format: XX-XXX or XXXXX (Polish default, override via locale). */
 export function isPostcode(text) {
     return /^\d{2}-?\d{3}$/.test(text.trim());

package/dist/index.d.ts

@@ -1,6 +1,6 @@
 export type { SqlDialect, DatabaseClient, SchemaAdapter, SearchParams, SearchResult, SearchResponse, AutocompleteResult, SearchLocale, ResolverContext, ResolverLookups, ResolverAction, TrackSearchEvent, } from './core/types.js';
 export { haversineKm, walkingMinutes, boundingBox, formatDistance, formatWalkingTime, walkingRoute, } from './core/geo.js';
-export { normalize, stripDiacriticsGeneric, trigrams, trigramSimilarity, levenshtein, levenshteinSimilarity, isPostcode, hasGeoIntent, stripGeoIntent, stripStopWords, MIN_SEARCH_TOKEN_LENGTH, } from './core/normalize.js';
+export { normalize, stripDiacriticsGeneric, trigrams, trigramSimilarity, levenshtein, levenshteinSimilarity, bestWordSimilarity, isPostcode, hasGeoIntent, stripGeoIntent, stripStopWords, MIN_SEARCH_TOKEN_LENGTH, } from './core/normalize.js';
 export { createSearchEngine, type SearchEngineConfig } from './core/engine.js';
 export { createIndexer, createLookupsLoader, type IndexerConfig, type LoadLookupsConfig } from './core/indexer.js';
 export { parseQuery, findMatchingArea, findNearestLocationWithEntities, type ParsedQuery } from './core/resolver.js';

package/dist/index.js

@@ -3,7 +3,7 @@
 // Geo utilities
 export { haversineKm, walkingMinutes, boundingBox, formatDistance, formatWalkingTime, walkingRoute, } from './core/geo.js';
 // Normalization & similarity
-export { normalize, stripDiacriticsGeneric, trigrams, trigramSimilarity, levenshtein, levenshteinSimilarity, isPostcode, hasGeoIntent, stripGeoIntent, stripStopWords, MIN_SEARCH_TOKEN_LENGTH, } from './core/normalize.js';
+export { normalize, stripDiacriticsGeneric, trigrams, trigramSimilarity, levenshtein, levenshteinSimilarity, bestWordSimilarity, isPostcode, hasGeoIntent, stripGeoIntent, stripStopWords, MIN_SEARCH_TOKEN_LENGTH, } from './core/normalize.js';
 // Search engine
 export { createSearchEngine } from './core/engine.js';
 // Indexer

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nomideusz/svelte-search",
-  "version": "0.1.3",
+  "version": "0.2.0",
   "description": "Full-text search engine for Svelte 5 apps — FTS5, trigram fuzzy matching, geo proximity, autocomplete, Polish locale support.",
   "type": "module",
   "license": "MIT",
@@ -17,7 +17,12 @@
       "default": "./dist/locales/pl.js"
     }
   },
-  "files": ["dist", "CHANGELOG.md", "!dist/**/*.test.*", "!dist/**/*.spec.*"],
+  "files": [
+    "dist",
+    "CHANGELOG.md",
+    "!dist/**/*.test.*",
+    "!dist/**/*.spec.*"
+  ],
   "scripts": {
     "dev": "vite dev",
     "build": "vite build",
@@ -34,18 +39,29 @@
     "svelte": "^5.0.0"
   },
   "devDependencies": {
-    "@sveltejs/adapter-auto": "^7.0.0",
-    "@sveltejs/kit": "^2.60.1",
-    "@sveltejs/package": "^2.5.7",
+    "@sveltejs/adapter-auto": "^7.0.1",
+    "@sveltejs/kit": "^2.69.1",
+    "@sveltejs/package": "^2.5.8",
     "@sveltejs/vite-plugin-svelte": "^6.2.4",
-    "svelte": "^5.51.0",
-    "svelte-check": "^4.3.6",
+    "marked": "^17.0.6",
+    "svelte": "^5.56.4",
+    "svelte-check": "^4.7.1",
     "typescript": "^5.9.3",
-    "vite": "^7.3.5",
-    "vitest": "^4.1.0",
-    "marked": "^17.0.3"
+    "vite": "^7.3.6",
+    "vitest": "^4.1.9"
   },
-  "keywords": ["svelte", "search", "fts5", "trigram", "fuzzy", "geo", "autocomplete", "polish", "sqlite", "postgres"],
+  "keywords": [
+    "svelte",
+    "search",
+    "fts5",
+    "trigram",
+    "fuzzy",
+    "geo",
+    "autocomplete",
+    "polish",
+    "sqlite",
+    "postgres"
+  ],
   "repository": {
     "type": "git",
     "url": "https://github.com/nomideusz/svelte-search"