alvis/oO
1
0
Fork 0
Code Issues 34 Pull Requests Actions Packages Projects Releases Wiki Activity

feat(api): orchestrator cutover — replace bandit with multi-agent pipeline (ADR-0013 step 6)

Browse Source 
POST /recommend now calls ml/serving /recommend with pre-computed agent
snippets + task context instead of /generate + /score/egreedy/v2. Falls
back to a random signal candidate when ml/serving is unavailable.

Removes: remotePolicy, fetchLlmCandidates, sendRewardWithRetry,
candidateCache, pickPromptVersion. Feedback handler keeps inferReward +
tipFeedback writes for observability; reward delivery to the bandit is gone.
tipScores.policy is now 'orchestrator'; promptVersion is 'v4-orchestrator'.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
alvis
2026-05-04 10:37:15 +00:00
parent 7e958a779d
commit c65bedcf68
4 changed files with 117 additions and 402 deletions
Download Patch File Download Diff File Expand all files Collapse all files

45
services/api/src/routes/__tests__/recommender.unit.test.ts
View File

@@ -3,8 +3,7 @@
* These can import directly from the module without any mocking.
*/
import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
import { inferReward, dueAgeDays, pickPromptVersion } from '../recommender.js';
import { config } from '../../config.js';
import { inferReward, dueAgeDays } from '../recommender.js';
describe('inferReward', () => {
it('dismiss → -1', () => expect(inferReward('dismiss', null)).toBe(-1.0));
@@ -38,45 +37,3 @@ describe('dueAgeDays', () => {
expect(dueAgeDays({ date: yesterday })).toBeGreaterThan(0);
});
});
describe('pickPromptVersion', () => {
// Save + restore the original env-driven config field across tests.
let original: string;
beforeEach(() => { original = config.TIP_PROMPT_VERSION; });
afterEach(() => { (config as { TIP_PROMPT_VERSION: string }).TIP_PROMPT_VERSION = original; });
it('empty config → null (let ml/serving pick its default)', () => {
(config as { TIP_PROMPT_VERSION: string }).TIP_PROMPT_VERSION = '';
expect(pickPromptVersion()).toBeNull();
});
it('whitespace-only config → null', () => {
(config as { TIP_PROMPT_VERSION: string }).TIP_PROMPT_VERSION = ' ';
expect(pickPromptVersion()).toBeNull();
});
it('single value → that value', () => {
(config as { TIP_PROMPT_VERSION: string }).TIP_PROMPT_VERSION = 'v2-mentor';
expect(pickPromptVersion()).toBe('v2-mentor');
});
it('comma-separated → uniformly samples from the set', () => {
(config as { TIP_PROMPT_VERSION: string }).TIP_PROMPT_VERSION = 'v1,v2-mentor,v3-few-shot';
const seen = new Set<string>();
// With 100 trials, the chance of missing any of 3 buckets is (2/3)^100 ≈ 0 — test is reliable.
for (let i = 0; i < 100; i++) {
const picked = pickPromptVersion();
expect(picked).not.toBeNull();
seen.add(picked!);
}
expect(seen).toEqual(new Set(['v1', 'v2-mentor', 'v3-few-shot']));
});
it('trims whitespace around comma-separated entries', () => {
(config as { TIP_PROMPT_VERSION: string }).TIP_PROMPT_VERSION = ' v1 , v2-mentor ';
for (let i = 0; i < 20; i++) {
const picked = pickPromptVersion()!;
expect(['v1', 'v2-mentor']).toContain(picked);
}
});
});