feat(profile): user-profile feature registry + builder (phase A)

Centralizes user-level features (completion_rate_30d, dismiss_rate_30d,
mean_dwell_ms_30d, preferred_hour, tip_volume_30d) in a TS registry that
owns both definition and SQL aggregation, since the data lives in the
TS-owned SQLite tables (tip_views/tip_feedback). Lazy TTL refresh keeps
recommend latency bounded; values persist in user_profile_features (KV).

ml/serving accepts profile_features on /score + /generate but does not
yet consume them — extending the bandit feature vector changes D and
resets every user's learned state, so that's a deliberate phase-B step.

Includes ml/features/profile_schema.py as a contract mirror with a sync
test that diffs name sets against registry.ts.

ADR-0011 records the data-locality reasoning (registry in TS, not Python
as the issue originally suggested).

Phase B (deferred): event-driven incremental updates, bandit consumption
with state migration, admin per-user profile page, staleness alerts.

Refs #81.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

services/api/src/routes/recommender.ts

@@ -10,6 +10,7 @@ import type { TipCandidate, Signal } from '@oo/shared-types';
 import { todoistSource, dueAgeDays } from '../signals/todoist.js';
 export { dueAgeDays };
 import { SignalAggregator } from '../signals/aggregator.js';
+import { getProfile, type Profile } from '../profile/builder.js';
 
 const router: ExpressRouter = Router();
 
@@ -82,6 +83,7 @@ function signalToCandidate(signal: Signal): TipCandidate {
 async function remotePolicy(
   userId: string,
   tasks: TipCandidate[],
+  profile: Profile,
 ): Promise<{ tipId: string; score: number; policy: string } | null> {
   const hour = new Date().getHours();
   const dayOfWeek = new Date().getDay();
@@ -96,6 +98,7 @@ async function remotePolicy(
       features: t.features,
     })),
     context: { hour_of_day: hour, day_of_week: dayOfWeek },
+    profile_features: profile,
   };
 
   // Active policy: egreedy-v1 (selected over linucb-v1 after offline sim — ADR-0007)
@@ -141,6 +144,7 @@ async function fetchLlmCandidates(
   hour: number,
   dayOfWeek: number,
   promptVersion: string | null,
+  profile: Profile,
 ): Promise<LlmGenerateResult> {
   try {
     const tasks = signals.slice(0, 10).map((s) => ({
@@ -156,6 +160,7 @@ async function fetchLlmCandidates(
         user_id: userId,
         context: { tasks, hour_of_day: hour, day_of_week: dayOfWeek },
         n: 3,
+        profile_features: profile,
         ...(promptVersion ? { prompt_version: promptVersion } : {}),
       }),
       signal: AbortSignal.timeout(15_000),
@@ -208,6 +213,8 @@ router.post('/recommend', requireAuth, async (req: AuthenticatedRequest, res: Re
 
   // Stage 1: assemble candidates — aggregated signals + LLM-generated advice (parallel)
   const signals = await aggregator.fetchAll(req.userId!);
+  // Refresh + load the user-level profile feature dict (lazy TTL refresh).
+  const profile = await getProfile(req.userId!);
 
   const signalCandidates = signals.map(signalToCandidate);
   const requestedPromptVersion = pickPromptVersion();
@@ -217,6 +224,7 @@ router.post('/recommend', requireAuth, async (req: AuthenticatedRequest, res: Re
     hour,
     dayOfWeek,
     requestedPromptVersion,
+    profile,
   );
 
   const allCandidates: TipCandidate[] = [...signalCandidates, ...llmResult.candidates];
@@ -231,7 +239,7 @@ router.post('/recommend', requireAuth, async (req: AuthenticatedRequest, res: Re
   const t0 = Date.now();
 
   // Stage 2: score — egreedy bandit with random fallback
-  const scored = await remotePolicy(req.userId!, allCandidates);
+  const scored = await remotePolicy(req.userId!, allCandidates, profile);
   const latencyMs = Date.now() - t0;
   const tip = scored
     ? (allCandidates.find((t) => t.id === scored.tipId) ?? randomPolicy(allCandidates))