refactor(search): reduce topK limit for search queries and adjust related components

- Updated the maximum value of topK from 20 to 15 in the search request schema and related functions.
- Adjusted default topK values in various components to align with the new limit, ensuring consistent behavior across the application.
- Modified tests to reflect the updated topK values, maintaining the integrity of search functionality.

Author: kentcdodds

services/search-worker/src/index.ts

@@ -10,7 +10,7 @@ import { type Env } from './env'
 
 const searchRequestSchema = z.object({
 	query: z.string(),
-	topK: z.number().int().min(1).max(20).optional(),
+	topK: z.number().int().min(1).max(15).optional(),
 })
 
 const syncRequestSchema = z.object({

services/search-worker/src/search-service.test.ts

@@ -211,7 +211,7 @@ test('search drops weak tail below relative confidence of top hit', async () =>
 
 	const { results, lowRankingResults, noCloseMatches } = await service.search({
 		query: 'test',
-		topK: 20,
+		topK: 15,
 	})
 
 	expect(noCloseMatches).toBe(false)

services/search-worker/src/search-service.ts

@@ -60,11 +60,14 @@ type SearchDependencies = {
 	syncArtifacts: (args: { force?: boolean }) => Promise<{ syncedAt: string }>
 }
 
+const SEARCH_TOP_K_MAX = 15
+const SEARCH_TOP_K_DEFAULT = 8
+
 function clampTopK(topK: number | undefined) {
 	if (typeof topK === 'number' && Number.isFinite(topK)) {
-		return Math.max(1, Math.min(20, Math.floor(topK)))
+		return Math.max(1, Math.min(SEARCH_TOP_K_MAX, Math.floor(topK)))
 	}
-	return 15
+	return SEARCH_TOP_K_DEFAULT
 }
 
 function isLexicalOnlySearch(env: Env) {

services/site/app/routes/search.tsx

@@ -66,7 +66,7 @@ export async function loader({ request }: Route.LoaderArgs) {
 
 	const searchPayload = searchKCD({
 		query: normalizedQ,
-		topK: 20,
+		topK: 8,
 		request,
 	}).catch((e) => {
 		console.error(e)

services/site/app/utils/search.server.ts

@@ -76,9 +76,9 @@ function makeSearchCacheKey({
 
 function clampTopK(topK: number | undefined) {
 	if (typeof topK === 'number' && Number.isFinite(topK)) {
-		return Math.max(1, Math.min(20, Math.floor(topK)))
+		return Math.max(1, Math.min(15, Math.floor(topK)))
 	}
-	return 15
+	return 8
 }
 
 function normalizeCachedSearchPayload(
@@ -109,7 +109,7 @@ async function enrichResults(results: Array<SearchResult>) {
 
 export async function searchKCD({
 	query,
-	topK = 15,
+	topK = 8,
 	request,
 	timings,
 }: {

services/site/content/blog/building-semantic-search-on-my-content.mdx

@@ -188,8 +188,8 @@ I overfetch chunk-level matches (`safeTopK * 5`, capped at 20), because several
 
 ```ts
 // `K` means "how many nearest neighbors/results to return".
-const safeTopK = Math.max(1, Math.min(20, Math.floor(topK)))
-const rawTopK = Math.min(20, safeTopK * 5)
+const safeTopK = Math.max(1, Math.min(15, Math.floor(topK)))
+const rawTopK = Math.min(15, safeTopK * 5)
 
 const [queryVector] = await getEmbeddings({ texts: [query] })