loto/TODO.md

Next-session TODO

Hand-off list as of 2026-10-12 (reconciled against the code after the 2026-09-01 fix cycle). Covers both web/ (SvelteKit PWA, deployed to GitHub Pages at the /loto/ base) and android/ (native Kotlin + Jetpack Compose app — the Capacitor wrapper was retired 2026-08-31). Shipped plan folders are swept after release; residual and new items live here directly.

Highest leverage (start here)

Android device QA

The acceptance spec for the native rewrite's release QA (plans/260831-1131-native-android-rewrite/phase-10-qa-and-release.md). Every item is device/emulator-only.

• Tap a cell → vibrates; system "remove animations" → silent.
• Auto-call 3+ min untouched → screen stays lit. Stop → sleeps normally.
• Background during auto-call, return → still lit.
• Finish a round (remaining 0) → lock releases.
• Back with bingo modal open → modal closes, app stays.
• Back with settings open → sheet closes.
• Back at root → Vietnamese exit dialog; "Ở lại" keeps state.
• System font 200% → grid intact (Compose sp scaling); Settings → Cỡ chữ bảng resizes; survives reload.
• Gesture nav + status bar clear of the settings gear and footer.
• Volume rocker during a call → media slider.
• Launcher shows Lô tô with the brand icon; themed-icon mode tints it.
• Cold start in dark mode → dark splash, no white flash.
• Offline: airplane mode, kill app, relaunch — fully functional.
• Audio: draw a number → Vietnamese voice plays; both voices switch correctly (hoai-my ↔ nam-minh); rapid draws cut each other off; chờ + number plays gaplessly when the setting is on. v0.2.1 moved the clips to the assets root (paths lost the audio/ prefix); CI built and tested it, but playback is unconfirmed on a device.
• Audio focus: playback layers over background music without pausing it (decided: no focus request — see ExoVoicePlayer); incoming call behaves normally.
• Sideload onto real devices (API 24, 31, 36) and BlueStacks.
• Store listing assets: feature graphic 1024×500, 2–8 phone screenshots, short (≤80) and full (≤4000) descriptions.

versionCode footgun (resolved 2026-08-31). android-release.yml now fails fast when the tag's versionCode in android/app/build.gradle.kts is not greater than the previous release tag's. Bumping stays manual; forgetting is no longer silent.

Play production access (status unrecorded)

12 testers were collected on 2026-08-19, so the 14-day continuous window could have closed around 2026-09-02 if opt-ins never dropped below 12. Whether the production-access application was filed or granted is not recorded here — check Play Console and record the outcome. The application asks for tester feedback and the features they exercised; Play has been rejecting thin engagement since April 2026.

• Record the production-access outcome (applied / granted / window reset).

Delivery is wired: android-release.yml uploads to tracks: alpha, the closed track, as of 2026-08-20. Tagged releases now reach the testers directly. v0.1.1 and earlier went to the internal track — promote them by hand if any is wanted in the closed test. See docs/play-store-publishing.md → "Closed testing → production access".

• Track id confirmed. v0.1.2 (2026-08-20) logged Validating tracks: 'alpha' and committed the edit, so alpha exists on com.miti99.loto and the AAB reached it.

PWA install verification (manual, post-deploy)

Priority after the SvelteKit 3 upgrade (2026-10-12). Before the upgrade, the Pages build's sw.js precached / and bound navigations to /. Both point at the origin root, which returns 404 on Pages, so the service worker most likely never installed there. The new build (plain vite-plugin-pwa, replacing @vite-pwa/sveltekit, which breaks under Kit 3) uses /loto/. On Pages, confirm that the service worker installs, that the app works offline, and that the update banner appears after the next deploy.

Needs production deploy on GitHub Pages + physical Android Chrome + iOS Safari. No code; verification only.

Lighthouse — GitHub Pages (/loto/ base)

• Open https://tiennm99.github.io/loto/ in incognito Chrome.
• DevTools → Lighthouse → PWA + Perf + Best Practices + a11y.
• PWA score = 100. No mixed-content warnings.
• Confirm SW URL /loto/sw.js, manifest /loto/manifest.webmanifest, icons /loto/icons/... resolve.

Android Chrome (physical or emulator)

• "Add to Home Screen" → install → launch.
• Splash uses theme color (#1565c0 light / #0a0f1f dark, see app.html:9-10). Status bar matches.
• Standalone display (no Chrome chrome).
• Airplane mode → reload from home → app shell + default voice work.
• Maskable icon: long-press app icon, ensure mask doesn't crop the centered glyph. Verify in DevTools "Show maskable preview" too. The maskable art (source-maskable.svg) is now full-bleed with the glyph inside the 80% safe zone, matching the Android adaptive icon.

iOS Safari

• Share → Add to Home Screen.
• Icon uses apple-touch-icon (/icons/icon-192.png) — round-ish glyph, no white bars.
• Launch standalone, fonts legible under translucent status bar.
• Airplane mode → app shell + default voice play.

Common gotchas

• Manifest paths break under /loto/ base → check vite.config.js PWA manifest: false + app.html uses %sveltekit.assets%.
• iOS install shows wrong icon → ensure icons/icon-192.png is 192×192 actual size.

Supply chain

• npm audit web/ — 0 vulnerabilities as of 2026-10-12 (vitest moved to ^4.1.11). android/ has no npm dependencies.
• Dependabot covers npm (web/), gradle (android/) and github-actions (.github/dependabot.yml).

Parked (Android)

• Share sheet for the called-numbers card (native Intent.ACTION_SEND).
• Background audio (play numbers while the screen is off) — needs a foreground service.
• iOS target — would now mean a separate SwiftUI app or reviving a cross-platform layer; substantially bigger than in the wrapper era.

Decisions record (Android)

• Fresh native Kotlin/Compose rewrite (2026-08-31) replaced the Capacitor wrapper, reversing the 2026-05-10 retirement decision (user decision; trade-off accepted: web features now need manual Kotlin ports). Fresh code — nothing resurrected from the old port. Wrapper history preserved in git, like the old port before it.
• No INTERNET permission — makes "fully offline" a hard guarantee. The native app strips Media3's merged ACCESS_NETWORK_STATE too. Would have to come back if any remote feature is introduced.
• No audio focus (2026-08-31): voice clips are ~1s speech cues; the player never claims focus so background music keeps playing, but pauses on becoming-noisy (headphones unplugged). See ExoVoicePlayer.
• Historical: Capacitor 8 wrapper, not TWA (2026-08) — chosen for the first-launch offline guarantee. Native Kotlin/Compose port retired (2026-05-10) in favour of the wrapper, for maintenance parity; that port lives at commits fe52232 / 9a35686.

UX polish (carried over)

• MasterEmptyState ↔ PlayerBoard ghost-grid duplication. Two near-identical decorative components — extract a shared <GhostBoardPreview rows={N} /> only if a third use appears. (Rule-of-three not met yet.)

Tech debt

• M6 recomposition perf — needs a profiled baseline on a device first. Reviewed 2026-10-12: Kotlin 2.2 enables strong skipping, but the 99 MasterBoard cells are inlined in for loops inside one composable, so every draw re-runs all of them. If profiling shows jank, the cheap fix is a MasterCell(num, order, isLast) composable so only the two changed cells recompose per draw.
• Voice list growth. If we add voices > 2 (esp. > 10), revisit the precache strategy — currently we precache only the default voice. The 7d runtime cache covers the rest, but cold-start cost on alternate voices grows linearly.

New features (parking lot)

• Multiple cards per player.
• Long-press hero to undo last call (master).
• Spacebar to draw next number (master power-user shortcut).
• Internationalization (English UI). Vi default; en strings in a flat map. Maybe pick paraglide-js or roll our own minimal map.

Open questions (carried from the Android pass)

• boardTextScale rungs are 0.9 / 1 / 1.15 / 1.3. Enough range for users who already run system font at 200%?
• The exit confirm fires at root regardless of round state. Worth gating on "is a round live" so an idle app just exits?