feat(scoring): nightly grant-embedding workflow (Stage 2, step 1)

embedGrants (cron 04:15, after the ingest crons): open grants with a
synopsis and no vector → buildGrantEmbeddingText (pure, tested:
title+funder+program areas+synopsis, 8K cap) → gemini-embedding-001 @
1536 dims RETRIEVAL_DOCUMENT (org profiles will embed as
RETRIEVAL_QUERY on the other side) → grants.synopsis_embedding.
Chunked embed→store (100/chunk) so failures resume from the last
stored chunk; 500/run spend cap.

Core: serverListGrantsNeedingEmbedding (open+unembedded, closest
deadline first), serverSetGrantEmbeddings. Worker gains
@novelpad/outreach-ai dep; wired into main.ts and run-once (incl. the
missed run-once deps injection).

Live-verified: 199/199 open grants embedded in 16s; semantic probe
('after-school STEM education for youth') ranks NCI Youth Enjoy
Science R25 first at 0.639 cosine via the hnsw index.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

packages/outreach-ai/src/grant-embedding-text.test.ts

@@ -0,0 +1,39 @@
+import { describe, expect, it } from 'vitest';
+
+import { buildGrantEmbeddingText } from './grant-embedding-text.js';
+
+describe('buildGrantEmbeddingText', () => {
+  it('composes title, funder, program areas, and synopsis', () => {
+    const text = buildGrantEmbeddingText({
+      funder: 'National Science Foundation',
+      title: 'STEM Education Grants',
+      synopsis: 'Funding for after-school STEM programs.',
+      programAreas: ['B25', 'O50'],
+    });
+    expect(text).toBe(
+      'STEM Education Grants — National Science Foundation.\n' +
+        'Program areas: B25, O50.\n' +
+        'Funding for after-school STEM programs.',
+    );
+  });
+
+  it('omits empty program areas and blank synopsis', () => {
+    const text = buildGrantEmbeddingText({
+      funder: 'F',
+      title: 'T',
+      synopsis: '   ',
+      programAreas: [],
+    });
+    expect(text).toBe('T — F.');
+  });
+
+  it('truncates very long synopses', () => {
+    const text = buildGrantEmbeddingText({
+      funder: 'F',
+      title: 'T',
+      synopsis: 'x'.repeat(20_000),
+      programAreas: null,
+    });
+    expect(text.length).toBe(8_000);
+  });
+});

packages/outreach-ai/src/grant-embedding-text.ts

@@ -0,0 +1,35 @@
+/**
+ * Composes the text that represents a grant in embedding space. Pure —
+ * shared by the nightly embed job and any ad-hoc re-embedding so a grant
+ * is always embedded from identically-composed text.
+ *
+ * Kept intentionally simple: funder + title + program areas + synopsis,
+ * truncated to stay well inside the embedding model's context. Mission-fit
+ * retrieval compares org-profile text against this, so the composition
+ * should read like a description, not a key-value dump.
+ */
+
+export interface GrantEmbeddingInput {
+  readonly funder: string;
+  readonly title: string;
+  readonly synopsis: string | null;
+  readonly programAreas: readonly string[] | null;
+}
+
+/** ~8K chars ≈ well under gemini-embedding-001's 2048-token input limit
+ * for typical English prose after the API's own truncation; the tail of a
+ * long federal synopsis is boilerplate anyway. */
+const MAX_TEXT_LENGTH = 8_000;
+
+export function buildGrantEmbeddingText(grant: GrantEmbeddingInput): string {
+  const parts: string[] = [`${grant.title} — ${grant.funder}.`];
+
+  if (grant.programAreas != null && grant.programAreas.length > 0) {
+    parts.push(`Program areas: ${grant.programAreas.join(', ')}.`);
+  }
+  if (grant.synopsis != null && grant.synopsis.trim() !== '') {
+    parts.push(grant.synopsis.trim());
+  }
+
+  return parts.join('\n').slice(0, MAX_TEXT_LENGTH);
+}

packages/outreach-ai/src/index.ts

@@ -5,3 +5,4 @@ export * from './agents/org-profiler/schema.js';
 export * from './agents/org-profiler/run.js';
 export * from './agents/mission-fit-judge/schema.js';
 export * from './agents/mission-fit-judge/run.js';
+export * from './grant-embedding-text.js';