Improve mobile data retrieval and state management

- Add race protection and packId normalization to swipe screen
- Add loading/error states to matchStore with couple_id filter
- Sync per-match unread counts in realtime
- Use useFocusEffect for data freshness on Home screen
- Wire RefreshControl to actual loading state on Matches

🤖 Generated with [Claude Code](https://claude.com/claude-code)

apps/mobile/DATA_RETRIEVAL_TASKS.md

@@ -0,0 +1,120 @@
+# Mobile data retrieval review: task list
+
+This task list targets the **Home**, **Swipe**, and **Matches** flows in `apps/mobile`, focusing on correctness, performance, and realtime consistency.
+
+## Overview (current state)
+
+- Data fetching is split between Zustand stores (`useMatchStore`, `usePacksStore`) and screen-local Supabase RPC calls (notably `SwipeScreen`).
+- Realtime updates are centralized in `apps/mobile/app/(app)/_layout.tsx` (matches/messages/profile/pack-settings subscriptions).
+- Main risks today: missing query filters & error/loading states, redundant fetches, state races on param changes, and unread-count drift between realtime and the matches list.
+
+## Scope
+
+- Screens
+  - Home: `apps/mobile/app/(app)/index.tsx`
+  - Swipe: `apps/mobile/app/(app)/swipe.tsx`
+  - Matches: `apps/mobile/app/(app)/matches.tsx`
+- Stores
+  - Matches: `apps/mobile/src/store/matchStore.ts`
+  - Packs: `apps/mobile/src/store/packsStore.ts`
+  - Messages: `apps/mobile/src/store/messageStore.ts`
+- Realtime orchestration
+  - `apps/mobile/app/(app)/_layout.tsx`
+
+## Priority 0 — correctness (must fix)
+
+- [ ] **Home: fix packs count source-of-truth**
+  - Current: `enabledPacksCount = packs.length` (counts available packs).
+  - Decide + implement:
+    - If label means *enabled*: use `enabledPackIds.length`.
+    - If label means *available*: rename the variable + label.
+  - Acceptance: packs number matches user expectations and store state.
+
+- [ ] **Swipe: normalize `packId` and reset state on param changes**
+  - Normalize `packId` from `useLocalSearchParams()` (`string | string[]`) before sending to RPC.
+  - Reset `currentIndex`, `isLoading`, and any derived UI state when `packId` changes.
+  - Acceptance: switching packs never shows stale index/loading/UI.
+
+- [ ] **Swipe: add race protection for overlapping async fetches**
+  - Use a request-id guard (or abort pattern) so older `fetchQuestions()` results cannot overwrite newer results.
+  - Acceptance: rapid navigation/focus changes never clobber newer state.
+
+- [ ] **Swipe: prevent stale “blocked” state on answer-gap errors**
+  - Ensure `checkAnswerGap()` clears/updates `isBlocked` + `gapInfo` deterministically even when RPC errors.
+  - Acceptance: blocked UI never “sticks” due to transient failures.
+
+## Priority 1 — UX and data freshness
+
+- [ ] **Home: ensure data freshness on revisit**
+  - Add `useFocusEffect` or a simple staleness TTL in stores.
+  - Acceptance: counts (matches/new/packs) refresh when returning to Home.
+
+- [ ] **Home: align “Recent Matches” semantics**
+  - `useMatchStore.fetchMatches()` sorts unread first; Home slices top 3.
+  - Decide:
+    - True recent by `created_at`, or
+    - “Top matches” by unread-first.
+  - Acceptance: section name and behavior match.
+
+- [ ] **Matches screen: wire up pull-to-refresh correctly**
+  - Add `isLoading`/`isRefreshing` + `error` to `useMatchStore`.
+  - Use those flags for `RefreshControl.refreshing`.
+  - Acceptance: pull-to-refresh shows spinner and recovers from errors.
+
+## Priority 2 — performance and query efficiency
+
+- [ ] **MatchStore: add explicit filtering and guard empty `in(...)` lists**
+  - Filter matches by `couple_id` (in addition to RLS) for clarity/perf.
+  - If no matches, avoid calling `.in('question_id', [])`.
+  - Acceptance: fewer noisy requests; no empty-list edge behavior.
+
+- [ ] **MatchStore: reduce unread-count work**
+  - Current approach fetches one row per unread message and counts on client.
+  - Options:
+    - Add RPC/view that returns per-match unread counts.
+    - Or store a denormalized `unread_count` per match (if acceptable).
+  - Acceptance: fetching matches scales with large message histories.
+
+- [ ] **Swipe: remove heavy dependency on `fetchPacks()`**
+  - Swipe often only needs `enabledPackIds`.
+  - Add lightweight `ensureEnabledPacksLoaded()` or call `fetchEnabledPacks()` directly.
+  - Acceptance: less data fetched when entering Swipe.
+
+- [ ] **Swipe: replace random comparator sort**
+  - Replace `Array.sort()` comparator using `Math.random()` with a stable score/shuffle pass.
+  - Acceptance: consistent ordering logic; avoids comparator contract violations.
+
+## Priority 3 — realtime correctness and consistency
+
+- [ ] **Realtime messages: tighten subscription filter**
+  - Current `messages` subscription listens without a `filter`.
+  - Improve by filtering to the user’s couple scope (schema permitting) or alternative channel strategy.
+  - Acceptance: avoids noise; reduces risk of unintended cross-couple events.
+
+- [ ] **Sync per-match unread counts with realtime events**
+  - Currently global unread count updates live; per-match counts update only on `fetchMatches()`.
+  - Options:
+    - Update `useMatchStore` unreadCount per match when realtime message arrives.
+    - Or trigger a targeted refresh when new partner message arrives.
+  - Acceptance: Matches list ordering/badges update as messages arrive.
+
+## Optional: structural improvements (bigger refactors)
+
+- [ ] Consolidate fetching into a query layer (e.g., TanStack Query) with caching, invalidation, retries, and focus refetch.
+- [ ] Create a single “matches list” RPC that returns:
+  - match + question
+  - who-answered-first info (for two-part display)
+  - per-match unread counts
+  - new match flag
+- [ ] Add lightweight integration tests around:
+  - `packId` parsing
+  - request race guards
+  - unread count updates
+
+## Suggested implementation order
+
+1. P0 correctness (Home packs count, Swipe param/reset/races, gap-state)
+2. Matches loading/refresh UX
+3. Query efficiency (couple filters, unread-count optimization)
+4. Realtime tightening + unread sync
+5. Optional refactors