Recommendation: Target API level 34+ and compile with the latest SDK; set minSdk to at least 21 unless analytics show significant users below that. Use Kotlin 1.9+ with structured coroutines for background IO, adopt a single-activity architecture and Jetpack Compose for new UI modules to reduce view hierarchy and lower frame-time variance.

Performance goals: keep UI frame time under 16ms (60 fps), avoid main-thread work exceeding 2ms per interaction, and limit per-frame allocations to 1–2 objects. Aim for cold start ≤2s on mid-range devices and warm start ≤200ms. Throttle high-frequency input using Kotlin Flow or channels and debounce background tasks to prevent backpressure.

Security & stability: sign releases with Play App Signing, require TLS 1.3 for all endpoints, use Network Security Configuration with pinned roots for sensitive flows, apply scoped storage and one-time runtime permissions wherever feasible. Scan dependencies automatically, pin critical versions, and restrict exported components with explicit permission checks at entry points.

Build & distribution: enable R8 code shrinking and resource shrinking in release builds (minifyEnabled true, shrinkResources true), strip debug symbols and publish via AAB for most releases; expect binary size reductions of 20–40% with aggressive dead-code elimination. Run lint, Detekt and unit tests in CI and include instrumentation tests on physical devices with ≤2GB RAM to detect low-memory regressions.

Background work & battery: schedule deferrable jobs with WorkManager and use foreground services with persistent notifications for visible long-running tasks. Batch network syncs to reduce wakelocks; target average background network activity below 5KB/s per active account and keep periodic sync intervals to no more than once every 15 minutes unless user-initiated.

Profiling & observability: collect CPU, memory and energy traces for representative sessions and define performance budgets (startup, memory, jank). Fail CI when budgets are exceeded. Ship lightweight telemetry (sample rate 0.1–1%) for crashes and ANRs and upload symbol maps for obfuscated builds to speed triage.

App Architecture & Code Quality

Adopt a modular, feature-driven architecture with one explicit public API per module; enforce acyclic module dependency graph, target module compile time <30s and module size <400 classes to keep builds responsive.

Enforce strict layering: UI → Presentation (ViewModel/Presenter) → Use-cases/Interactors → Repository → Data Sources; prohibit direct data-source access from UI layers and require interface contracts for every cross-layer call.

Prefer compile-time dependency injection (Dagger or generated factories) over runtime-reflection frameworks; limit DI scope per feature, prefer constructor injection, avoid GlobalScope-like patterns for lifecycle-managed components.

Testing rules: 1) unit tests for business logic with ≥80% coverage on core modules; 2) integration tests for persistence and networking covering schema and contract migrations; 3) UI tests covering the top 10 user flows with flakiness <2%; keep PR test runs under 10 minutes.

Static analysis and formatting: run Detekt + Ktlint (Kotlin) or equivalent linters in CI; fail builds on new critical/major issues; maintain a baseline for legacy warnings and remove gradually; enforce single code style via pre-commit hooks.

Code review and workflow: require at least two approvers for feature merges, limit PRs to <300 LOC, attach automated checks (tests, linters, security scans), and reject merges if any check fails; use trunk-based development with feature flags for frequent, small merges.

Binary size and performance targets: enable R8/resource shrinking and ABI splits; set automated alerts for >5% binary growth per release; keep method count below ~50k pre-split to reduce cold-start overhead; aim for cold start <500ms on mid-tier hardware and main-thread blocking work <50ms per frame.

Database and storage: version all migrations, include automated migration tests for every schema change, run schema validation on startup in debug builds, and keep migration test coverage at 100% for breaking changes; snapshot fixtures for deterministic integration tests.

Observability: instrument screen load and network calls with traces and metrics; track 95th-percentile API latency and client-side rendering time; surface crash, ANR and memory-leak trends in release dashboards and set alerts for regression thresholds.

Documentation and decision tracking: store Architecture Decision Records (ADRs) in the repo for major choices, maintain per-module README with public API examples and compatibility guarantees, and schedule weekly dependency-update PRs with monthly manual reviews and security scans.

Adopt MVVM with ViewModel

Keep UI state inside the ViewModel using MutableStateFlow and expose it as an immutable StateFlow&lt;UIState&gt;. Use a single data class for UIState (val loading: Boolean, val items: List, val error: String?) and update with copy() to preserve immutability.

Persist small, critical UI values with SavedStateHandle. Store keys for form fields, selected IDs and paging cursors: savedStateHandle.set(“query”, query); restore with savedStateHandle.get&lt;String&gt;(“query”). Avoid using it for large binary blobs.

Scope coroutines to the ViewModel via viewModelScope and choose dispatchers explicitly. Use viewModelScope.launch(Dispatchers.IO) for network and disk, with withContext(Dispatchers.Default) for CPU work. Cancel or timeout long operations with withTimeout or structured concurrency patterns.

Deliver one-off events with a SharedFlow or a Channel instead of mutable LiveData hacks. Example pattern: private val _events = MutableSharedFlow&lt;UiEvent&gt;(replay = 0, extraBufferCapacity = 1); val events = _events. In case you loved this information and you wish to receive details relating to onexbet app kindly visit the site. asSharedFlow(); emit via _events.tryEmit(UiEvent.Navigate(…)). Consume using lifecycle-aware collectors.

Debounce and cancel obsolete requests with operators. For search use: queryState.debounce(300L).distinctUntilChanged().flatMapLatest repo.search(it) .collect _uiState.update it.copy(items = it) – 300 ms is a good default for user input throttling.

Keep ViewModel free of View and Context references. Never store Activity/Fragment/View instances. Inject repositories, data sources and application-level helpers via constructor injection or a ViewModelFactory. If an application context is required, provide it through an injected provider rather than a direct field.

Map domain models to UI models inside the ViewModel. Perform transformations in a dedicated function or mapper (domain -> UiModel) so views receive display-ready objects (formatted strings, localized numbers, display flags) and tests can assert mapping logic in isolation.

Write unit tests with coroutine test tooling and a controllable dispatcher. Use runTest and a TestDispatcher; set Dispatchers.setMain(testDispatcher) in setup; verify state flows by collecting StateFlow values or using Turbine for flow assertions. Mock repositories to return flows or suspend functions.

Initialize heavy work lazily and avoid long-running work in init. Use explicit load triggers from the UI (e.g., loadPage()) or use lazy flows combined with shareIn to start work only when there are collectors, reducing wasted CPU and memory.

Expose implementation details as interfaces and keep the ViewModel thin: orchestrate, don’t implement all business logic. Put validation, caching and network orchestration into repositories or use-case classes; the ViewModel should coordinate inputs, call use-cases and emit UIState and events.