fix: check trimmed text for heading marker so note bodies reach embeddings

src/lib/embedder.ts

@@ -1,4 +1,5 @@
 import { pipeline, type FeatureExtractionPipeline } from '@huggingface/transformers';
+import { firstBodyParagraph } from './text.js';
 
 const MODEL = 'Xenova/all-MiniLM-L6-v2';
 
@@ -32,7 +33,7 @@ export class Embedder {
     tags: string[],
     content: string,
   ): string {
-    const firstParagraph = content.split(/\n\n+/)[0] ?? '';
+    const firstParagraph = firstBodyParagraph(content).trim();
     const parts = [title];
     if (tags.length > 0) {
       parts.push(tags.join(', '));

src/lib/store.ts

@@ -1,6 +1,7 @@
 import Database from 'better-sqlite3';
 import * as sqliteVec from 'sqlite-vec';
 import type { ParsedNode, ParsedEdge, SearchResult } from './types.js';
+import { firstBodyParagraph } from './text.js';
 
 export class Store {
   db: Database.Database;
@@ -281,7 +282,7 @@ export class Store {
 }
 
 function firstParagraph(content: string, maxLen: number): string {
-  const para = content.split(/\n\n+/).find(p => p.trim().length > 0 && !p.startsWith('#'));
+  const para = firstBodyParagraph(content);
   if (!para) return '';
   const trimmed = para.trim();
   return trimmed.length > maxLen ? trimmed.slice(0, maxLen) + '...' : trimmed;

src/lib/text.ts

@@ -0,0 +1,14 @@
+/**
+ * Returns the first non-empty, non-heading paragraph in `content`.
+ *
+ * `gray-matter` leaves a leading newline after stripping the YAML
+ * frontmatter, so the first paragraph from a naive `split('\n\n')` is often
+ * `'\n# Title'` — which only looks like a heading once trimmed. Trimming
+ * before the `#` check ensures the title line is correctly skipped in
+ * favor of the actual body text.
+ */
+export function firstBodyParagraph(content: string): string {
+  return (
+    content.split(/\n\n+/).find(p => p.trim().length > 0 && !p.trim().startsWith('#')) ?? ''
+  );
+}

test/embedder.test.ts

@@ -39,6 +39,18 @@ describe('Embedder', () => {
     expect(text).toContain('theoretical framework');
     expect(text).not.toContain('More details here');
   });
+
+  it('includes the body paragraph, not the title heading, when content has a gray-matter-style leading newline', () => {
+    // gray-matter leaves a leading newline after stripping YAML frontmatter,
+    // so the raw content starts with '\n# Widget Theory'.
+    const text = Embedder.buildEmbeddingText(
+      'Widget Theory',
+      [],
+      '\n# Widget Theory\n\nA theoretical framework for understanding component interactions.\n\n## Section',
+    );
+    expect(text).toContain('A theoretical framework for understanding component interactions');
+    expect(text).not.toContain('# Widget Theory');
+  });
 });
 
 function cosineSimilarity(a: Float32Array, b: Float32Array): number {

test/store.test.ts

@@ -110,6 +110,23 @@ describe('Store', () => {
     expect(results[0].excerpt).toContain('framework');
   });
 
+  it('vector search excerpt skips the title heading when content has a gray-matter-style leading newline', () => {
+    store.upsertNode({
+      id: 'test.md',
+      title: 'Widget Theory',
+      // gray-matter leaves a leading newline after stripping frontmatter,
+      // so a naive split on '\n\n' yields '\n# Widget Theory' as paragraph 0.
+      content: '\n# Widget Theory\n\nA framework for understanding component interactions.\n\n## Section',
+      frontmatter: {},
+    });
+    const embedding = new Float32Array(384).fill(0.1);
+    store.upsertEmbedding('test.md', embedding);
+    const results = store.searchVector(embedding, 5);
+    expect(results.length).toBeGreaterThan(0);
+    expect(results[0].excerpt).toContain('A framework for understanding component interactions');
+    expect(results[0].excerpt).not.toContain('# Widget Theory');
+  });
+
   it('counts edges for a node', () => {
     store.upsertNode({ id: 'a.md', title: 'A', content: '', frontmatter: {} });
     store.upsertNode({ id: 'b.md', title: 'B', content: '', frontmatter: {} });

test/text.test.ts

@@ -0,0 +1,21 @@
+import { describe, it, expect } from 'vitest';
+import { firstBodyParagraph } from '../src/lib/text.js';
+
+describe('firstBodyParagraph', () => {
+  it('returns the first non-empty, non-heading paragraph', () => {
+    const content = '# Title\n\nFirst body paragraph.\n\nSecond paragraph.';
+    expect(firstBodyParagraph(content)).toBe('First body paragraph.');
+  });
+
+  it('skips a leading heading even when preceded by a stray newline', () => {
+    // gray-matter leaves a leading newline after stripping YAML frontmatter,
+    // so the first split segment is '\n# Title', not '# Title'.
+    const content = '\n# Title\n\nFirst body paragraph.\n\n## Section';
+    expect(firstBodyParagraph(content)).toBe('First body paragraph.');
+  });
+
+  it('returns empty string when content has no body paragraph', () => {
+    expect(firstBodyParagraph('\n# Title Only')).toBe('');
+    expect(firstBodyParagraph('')).toBe('');
+  });
+});