Skip to content

StreamVault is a TV-first free IPTV player for Android TV built with Kotlin, Jetpack Compose, Room, Hilt, and Media3.

docs: simplify release notes and remove compose docs

David Nashash67446a8ce4a0
diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index 852a7348..1d2b3fe3 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -6,6 +6,9 @@ All notable product changes are recorded in this document.
### Changed
+- Improved Android TV navigation, focus handling, and Back-button behavior across Live TV, player controls, provider setup, and Settings.
+- Improved Live TV control reliability, including timeshift seeking and rewind interactions.
+- Improved reliability when configuring providers, importing settings, and preserving playback preferences.
- Improved Stalker EPG synchronization by increasing channel batch sizes from 500 to 5,000, reducing request overhead for large channel lineups while retaining bounded processing.
- Reduced storage usage by limiting the shared HTTP cache to 16 MiB and pruning past EPG entries after 12 hours when a provider does not require a longer catch-up window.
- Added VOD category pinning for Movies and Series, with pin-first browsing, Android TV pin controls, and pinned category content shelves on the Home dashboard.
diff --git a/docs/COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md b/docs/COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md
deleted file mode 100644
index a38ae01b..00000000
--- a/docs/COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md
+++ /dev/null
@@ -1,1510 +0,0 @@
-# Compose Reduction and UI Architecture Modernization Plan
-
-Status: In progress
-Owner: Unassigned
-Created: 2026-08-15
-Scope: Android application UI, navigation, build structure, and UI-facing module boundaries
-
-## Current execution state
-
-Phase 0 is complete for the available development environment. The build, artifact, memory, Compose compiler, Macrobenchmark, and live-playback baseline is recorded in [`COMPOSE_REDUCTION_BASELINE.md`](COMPOSE_REDUCTION_BASELINE.md), and the Compose compiler diagnostic outputs are generated under `app/build/reports/compose-compiler/`. Physical-device measurements and a stable provider stream are explicitly listed as follow-up limitations, not treated as successful release-quality measurements.
-
-The Phase 1 source decomposition is now code-complete: `PlayerControlsOverlayHost`, `PlayerLifecycleHost`, the platform-free `PlayerBackNavigationPolicy`, the placement-preserving `PlayerModalHosts`, and the event-time-tested `PlayerInputPolicy` are extracted from the player root; ProviderSetup is split into provider-specific forms, dialogs, import helpers, and components; and AppNavigation is split into startup/external-navigation policy, adapters, contracts, and graph registration. These changes preserve the existing callback behavior and keep packages and Gradle modules unchanged. The remaining manual and baseline instrumentation gates are recorded as incomplete in the execution plan.
-
-The full `:app:testDebugUnitTest` task now passes after the test fixture was updated to provide the already-required `m3uClassificationRepository` mock. No production behavior was changed; existing coroutine opt-in warnings remain.
-
-Phase 7 dependency/API implementation work is complete for provider setup, settings/preferences, and playback. `:feature:provider` and `:feature:settings` no longer depend on `:data` or import data-layer types; `:feature:playback` now consumes domain-owned player preferences and playback-resolution contracts, has no `:data` dependency, and remains free of direct Media3 dependencies or source references. Manual smoke and broader baseline follow-up gates remain tracked separately.
-
-The ProviderSetup slice now also places the source selector panel, provider-specific form content, and advanced-options section in dedicated files; the root continues to own draft state, launchers, effects, and ViewModel calls. The second no-behavior-change decomposition moved the remaining provider branches into explicit Xtream, Stalker, M3U, and Jellyfin form composables, and moved completion, compatibility-selector, validation-error, and password-transformation leaves into dedicated same-package files.
-
-The AppNavigation split is now complete for Phase 1: startup resolution, route/request contracts, navigation adapters, external-navigation dispatch, and graph registration are in dedicated same-package files; `AppNavigation.kt` retains state collection and orchestration only. No typed-route or module boundary was introduced.
-
-The review follow-up also closed player input freshness regressions: input and Back decisions now build their snapshots at event time, including the ViewModel-backed numeric-channel buffer, with regression tests covering digit-entry updates followed by immediate confirmation or Back. Focused player, provider, and navigation unit tests pass. Long-duration live playback and manual TV smoke checks remain required gates. The later API 36 emulator pass is recorded in the Phase 5 playback report; it covers seeded startup, preview, fullscreen, controls, and Back, while the full stability/manual matrix remains open.
-
-This branch was synchronized with the current local `develop` in merge commit `c2a2658f`, and synchronized again after `develop` advanced in merge commit `6e85d5c7` from local `develop` `5294fbd5`. Both merges were conflict-free. The second merge brought the backup/restore ledger, database migration, TV-input synchronization, and related settings/data changes. No production compatibility changes were needed for the Phase 1 extraction; the only adaptation was removing a duplicate `m3uClassificationRepository` field and constructor argument from `HomeViewModelTest` created by overlapping test-fixture changes. The merged tree passes `:app:testDebugUnitTest`, the focused backup/sync data tests, and `:app:assembleDebug`. The full `:data:testDebugUnitTest` suite runs 1,060 tests but retains the `ProviderExecutionArchitectureTest` failure at `data/src/test/java/com/streamvault/data/provider/ProviderExecutionArchitectureTest.kt:276` (expected dependency budget 40, actual 42).
-
-## 1. Executive summary
-
-StreamVault should reduce Compose by making it an intentional UI technology choice instead of the default location for all application behavior. Compose should remain the primary technology for dynamic, focus-heavy TV experiences such as Live TV, EPG, playback overlays, media shelves, and MultiView. Form-heavy and mostly static surfaces such as provider setup, settings, backup/import, and simple onboarding should be evaluated for classic Views after their state and navigation boundaries are cleaned up.
-
-The largest gains will not come from mechanically converting composables to XML. They will come from:
-
-1. Moving UI code out of the monolithic `:app` compilation unit into a small number of feature modules.
-2. Separating route/state ownership from pure rendering.
-3. Preventing frequently changing state from invalidating entire screens.
-4. Removing data-layer and Media3 implementation details from presentation code.
-5. Keeping shared UI APIs small and stable.
-6. Measuring build and runtime behavior before and after every migration phase.
-
-The recommended end state is a hybrid UI architecture with five to seven feature modules, a stable `:core:ui` module, feature-owned navigation registration, and explicit UI technology decisions per feature.
-
-## 2. Current-state assessment
-
-The following measurements were taken from the repository on 2026-08-15. They are architectural indicators, not performance benchmarks.
-
-| Area | Current observation | Architectural effect |
-|---|---:|---|
-| `app/src/main` | Approximately 296 Kotlin files and 88,885 lines | Most application-facing changes converge on one Android module |
-| Compose footprint | Approximately 458 composable functions | Compose compilation and presentation behavior are concentrated in `:app` |
-| UI screens | Approximately 69,741 lines and 360 composables below `ui/screens` | Features are organized by package but not isolated at the build boundary |
-| `ProviderSetupScreen.kt` | 3,651 lines and 32 composables | Provider forms, Android file operations, pairing, draft state, and dialogs share one file |
-| `PlayerScreen.kt` | 1,561 lines and approximately 57 `collectAsState*` calls | Independent state streams can invalidate a very broad composition scope |
-| `AppNavigation.kt` | 1,108 lines | Route declaration, startup resolution, navigation policy, and feature wiring are centralized |
-| Player overlay package | Approximately 6,284 lines and 47 composables | Playback UI is partially decomposed by file but still coordinated through a large root screen |
-| Settings package | Approximately 19,318 lines and 127 composables | Settings has become a large sub-application inside `:app` |
-| Shared card ticker | Each `ChannelCard` collects the same 30-second ticker | Every visible card owns a subscription and participates in periodic recomposition |
-
-The graph report also identifies the following highly connected hubs:
-
-- `SettingsViewModel`: 172 edges.
-- `Media3PlayerEngine`: 111 edges.
-- `TvClickableSurface()`: 91 edges.
-
-High connectivity is not automatically a defect. It does indicate that changes to these APIs can affect many features and that their responsibilities should be deliberately constrained.
-
-### 2.1 Build-time causes
-
-The likely build-time contributors are:
-
-- Nearly all Compose UI lives in `:app`, so incremental isolation stops at the file level rather than the feature-module level.
-- `:app` owns navigation, feature screens, Hilt wiring, services, Cast integration, TV input, update handling, and UI.
-- Large source files create expensive invalidation and analysis units for Kotlin and the Compose compiler.
-- `:app` directly declares Compose Material 3, TV Material, Media3, Room, Retrofit, OkHttp, WorkManager, Cast, and other dependencies.
-- Hilt and KSP also run in the same module that contains the largest Kotlin/Compose source set.
-- Shared UI code currently imports app navigation, `MainActivity`, domain models, and concrete application concerns, which makes extraction harder.
-
-The existing Gradle profile showing an up-to-date `assembleDebug` in roughly four seconds is not a useful compilation baseline because `compileDebugKotlin` was up-to-date. Phase 0 must establish clean, warm, and realistic incremental measurements.
-
-### 2.2 Runtime causes
-
-The likely runtime contributors are:
-
-- Broad state observation at screen roots, particularly in `PlayerScreen`.
-- Large UI state objects that contain many lists and maps with uncertain Compose stability.
-- Frequent player state changes mixed with infrequent metadata and dialog state.
-- Repeated state collectors inside list items.
-- Complex TV focus, pointer, key, animation, image, and overlay behavior sharing the same composition scopes.
-- Feature screens that combine rendering, file I/O launchers, navigation policy, lifecycle handling, local draft state, and ViewModel events.
-- Potentially unnecessary use of both Material 3 and TV Material components in the same presentation layer.
-
-These are hypotheses until Phase 0 records traces and benchmark results.
-
-## 3. Goals
-
-### 3.1 Primary goals
-
-- Reduce incremental build time for changes isolated to one UI feature.
-- Reduce broad and unnecessary recomposition on low-powered TV devices.
-- Make state ownership and navigation ownership obvious from the directory structure.
-- Allow selected features to use classic Views without forcing the entire application to use one UI toolkit.
-- Keep Live TV and player behavior unchanged during architectural migration.
-- Reduce direct dependencies from presentation code to `data` implementation types and concrete Media3 classes.
-- Make feature code independently testable.
-- Preserve TV remote focus, accessibility, RTL, phone, tablet, and TV behavior.
-
-### 3.2 Secondary goals
-
-- Reduce large source files to reviewable units.
-- Reduce `:app` to the Android composition root.
-- Standardize design-system and interaction primitives.
-- Make Compose compiler stability and recomposition reports part of performance work.
-- Prepare for baseline profiles and macrobenchmarks.
-
-### 3.3 Goal priority and timing
-
-| Goal | When | How we know it is working |
-|---|---|---|
-| Preserve current UI and playback behavior during structural changes | Every phase | Existing tests pass; manual smoke tests pass; player validation shows no new errors, stuckness, or focus regressions |
-| Make state, rendering, and platform responsibilities easier to locate | Phase 1 | Large screens are split into focused files with explicit route/screen/state boundaries |
-| Reduce unnecessary recomposition in confirmed hotspots | Phase 2 | Compose reports and runtime traces show fewer invalidated scopes or lower frame cost |
-| Improve incremental build isolation | Phases 3–6 | Feature-only edits recompile fewer tasks and meet the baseline-relative build target |
-| Use Compose or Views intentionally per feature | Phases 5–6 | Each candidate has a measured maintenance/runtime reason, not a toolkit preference |
-| Reduce `:app` coupling and direct implementation dependencies | Phases 3–7 | Feature modules depend on contracts and `:app` becomes primarily the composition root |
-
-The first active goal is structural safety: complete the player decomposition while preserving behavior. Performance targets are later goals until runtime and repeated build measurements confirm the hotspots.
-
-### 3.4 Non-goals
-
-- Rewriting the entire UI in XML.
-- Changing the product design during the architecture migration.
-- Replacing Navigation Compose immediately.
-- Replacing Hilt, Room, Media3, or the domain/data architecture.
-- Creating one Gradle module per screen.
-- Annotating mutable or uncertain models as `@Stable` or `@Immutable` merely to silence Compose diagnostics.
-- Combining architecture work with playback recovery policy changes.
-
-## 4. Architecture principles
-
-1. Feature ownership comes before toolkit ownership. A feature owns its screens, state, actions, components, tests, and resources.
-2. Routes collect state; screens render state. Android integration and ViewModels stay at the route boundary.
-3. Frequently changing state belongs near the leaf that renders it.
-4. Shared UI must be generic. `:core:ui` must not import app routes, `MainActivity`, DAOs, provider implementations, or concrete Media3 engines.
-5. Features depend on contracts, not implementations. Presentation should prefer domain/use-case interfaces and player API abstractions.
-6. Module boundaries should reflect meaningful team and rebuild boundaries, not individual files.
-7. Each phase must preserve behavior and have a rollback point.
-8. Runtime changes are validated in release-like builds on representative low-powered devices.
-9. Player architecture changes require full live playback validation, not launch-only validation.
-
-## 5. Target module architecture
-
-```text
-:app
- MainActivity
- StreamVaultApp
- root Hilt composition
- root navigation host
- platform services
- deep-link and external-intent dispatch
-
-:domain
- business models
- repository contracts
- use cases
- platform-independent policies
-
-:data
- Room
- network providers
- repository implementations
- sync
- DataStore and encrypted preferences implementations
-
-:player
- player API
- Media3 implementation
- playback policies
- timeshift
- surface/view binding
-
-:core:ui
- theme and design tokens
- generic TV focus and input primitives
- generic cards, dialogs, loading, and empty states
- image-loading wrappers
- adaptive screen primitives
-
-:core:navigation
- typed destination contracts
- common navigation request models
- no screen implementations
-
-:feature:live
- Live TV browsing
- channel/category UI
- EPG and guide
- preview UI
-
-:feature:catalog
- dashboard media shelves
- movies
- series
- VOD
- favorites
- search
-
-:feature:playback
- Player route and screen
- player controls and overlays
- player dialogs
- MultiView presentation
-
-:feature:provider
- provider setup and editing
- import and pairing presentation
-
-:feature:settings
- settings
- parental controls
- backup/restore presentation
-
-:feature:system
- welcome
- downloads
- plugin management UI
-```
-
-The final module count can be adjusted after measuring Gradle configuration and clean-build effects. `:feature:system` can remain in `:app` initially if its size does not justify extraction.
-
-### 5.1 Dependency direction
-
-```text
- +----------------+
- | :app |
- +--------+-------+
- |
- +-------------------+-------------------+
- | | |
- +------v------+ +-----v------+ +-----v------+
- | :feature:* | | :core:ui | | :core:nav |
- +------+------+ +-----+------+ +-----+------+
- | | |
- +-------------------+-------------------+
- |
- +--------v-------+
- | :domain |
- +--------+-------+
- ^
- +-------------+-------------+
- | |
- +------+-+ +----+----+
- | :data | | :player |
- +--------+ +---------+
-```
-
-Practical dependency rules:
-
-- `:app` may depend on every feature because it is the composition root.
-- Feature modules may depend on `:core:ui`, `:core:navigation`, `:domain`, and explicitly approved API modules.
-- Feature modules should not depend on another feature's implementation.
-- Feature-to-feature navigation uses destination contracts and callbacks.
-- `:data` and `:player` must not depend on feature modules or `:core:ui`.
-- `:core:ui` must not depend on `:app`, `:data`, or feature modules.
-- A temporary feature-to-`:data` dependency may be allowed during extraction, but it must be tracked and removed before that feature's migration is considered complete.
-
-### 5.2 Why feature modules improve builds
-
-Moving code into modules means a player UI edit recompiles `:feature:playback` and its downstream consumer rather than the complete UI source set. This primarily improves incremental builds. It may make clean builds slightly slower because there are more Gradle tasks, so clean-build time is a guardrail rather than the main success metric.
-
-The shared modules must remain small and stable. A frequently edited `:core:ui` module would invalidate every feature and erase much of the incremental-build benefit.
-
-## 6. UI technology strategy
-
-Compose should be selected where its declarative state, TV focus APIs, animation, and dynamic layout provide a clear benefit.
-
-| Feature or surface | Initial technology decision | Reason |
-|---|---|---|
-| Live TV browser | Keep Compose | Dynamic lists, focus, preview, overlays, adaptive modes |
-| EPG | Keep Compose | Custom grid behavior, time-based rendering, focus, dynamic loading |
-| Player overlays | Keep Compose | Layered visibility, state-driven controls, TV input and focus |
-| Player video surface | Keep View/Media3 surface behind an interop boundary | Video rendering should remain outside Compose drawing |
-| MultiView | Keep Compose initially | Dynamic slot layout and state-driven controls |
-| Dashboard/media shelves | Keep Compose | Lazy media rows and adaptive content |
-| Movies/series/VOD | Keep Compose initially | Shared media components and dynamic catalog lists |
-| Provider setup | Evaluate classic Views after state extraction | Large form, many text fields, launchers, imports, and provider-specific sections |
-| Settings | Evaluate classic Views after feature extraction | Form/list-heavy, mostly static, high source volume |
-| Backup/import | Strong candidate for Views or DialogFragment | Platform document contracts and form-like flow |
-| Welcome/onboarding | Either toolkit | Small surface; choose based on maintenance cost |
-| Plugin management | Keep current implementation until measured | No evidence yet that toolkit choice is the bottleneck |
-
-### 6.1 Decision gate for converting a feature to Views
-
-A feature should move from Compose to Views only when all of the following are true:
-
-- Its Route/Screen and state boundaries are already clean.
-- A benchmark or trace shows meaningful composition, layout, allocation, or interaction cost.
-- The feature is mostly static or form-based.
-- TV focus and accessibility behavior can be preserved or improved.
-- The conversion removes Compose from the feature module or substantially reduces a hot composition tree.
-- The additional interop/navigation complexity is acceptable.
-
-Using `AndroidView` around an entire XML screen is not considered a complete Compose reduction. It may reduce the internal composition tree, but it retains a Compose host and introduces an interop boundary. A Views-only feature should preferably use a Fragment or Activity entry point with ViewBinding and an explicit result/navigation contract.
-
-## 7. Standard feature hierarchy
-
-Each feature should use a predictable presentation structure:
-
-```text
-feature/<name>/src/main/java/com/streamvault/feature/<name>/
- navigation/
- <Feature>Destination.kt
- <Feature>NavGraph.kt
-
- presentation/
- <Feature>Route.kt
- <Feature>Screen.kt
- <Feature>UiState.kt
- <Feature>Action.kt
- <Feature>Effect.kt
- <Feature>ViewModel.kt
-
- components/
- dialogs/
- model/
-
- domain/ # optional feature-local orchestration
- resources/
-```
-
-### 7.1 Route responsibility
-
-The route may:
-
-- Obtain a ViewModel.
-- Collect lifecycle-aware state.
-- Translate navigation callbacks.
-- Observe one-shot effects.
-- Register Activity Result contracts.
-- Access Android-only services when they cannot be behind an interface.
-
-The route should not contain substantial layout code.
-
-### 7.2 Screen responsibility
-
-The screen should:
-
-- Accept immutable state and event callbacks.
-- Render the feature.
-- Be previewable and testable without Hilt.
-- Avoid repositories, DAOs, concrete player engines, and direct navigation-controller calls.
-- Avoid launching business work directly.
-
-### 7.3 State frequency boundaries
-
-State should be grouped by update frequency and consumer:
-
-```text
-FeatureRouteState infrequent route/identity/loading changes
-FeatureContentState content data and selection changes
-FeatureOverlayState dialogs, panels, transient interaction state
-FeatureRealtimeState position, progress, clock, diagnostics
-```
-
-Do not combine a one-second playback position with large channel lists and infrequently changing metadata in a single root state object.
-
-### 7.4 Event pattern
-
-Prefer a small event surface over dozens of unrelated lambdas when a feature has many actions:
-
-```kotlin
-sealed interface PlayerAction {
- data object TogglePlayback : PlayerAction
- data object OpenChannels : PlayerAction
- data class SelectChannel(val channelId: Long) : PlayerAction
- data class SeekTo(val positionMs: Long) : PlayerAction
-}
-
-@Composable
-fun PlayerScreen(
- state: PlayerScreenState,
- onAction: (PlayerAction) -> Unit,
- modifier: Modifier = Modifier,
-)
-```
-
-Do not force every small component to understand the feature-wide action type. Translate feature actions into narrow callbacks at section boundaries when that produces a clearer component API.
-
-## 8. Detailed refactoring plans
-
-### 8.1 Root application and navigation
-
-Current concern: `AppNavigation` owns startup routing, external requests, destination resolution, route argument parsing, and every feature's callback wiring.
-
-Target structure:
-
-```text
-app/navigation/
- AppNavHost.kt
- AppNavigator.kt
- AppStartupCoordinator.kt
- ExternalNavigationCoordinator.kt
- TopLevelNavigationPolicy.kt
-
-feature/*/navigation/
- LiveNavGraph.kt
- CatalogNavGraph.kt
- PlaybackNavGraph.kt
- ProviderNavGraph.kt
- SettingsNavGraph.kt
-```
-
-`AppNavHost` should read like a registry:
-
-```kotlin
-NavHost(...) {
- welcomeGraph(...)
- providerGraph(...)
- liveGraph(...)
- catalogGraph(...)
- playbackGraph(...)
- settingsGraph(...)
-}
-```
-
-Actions:
-
-- Move pure route construction and parsing into typed destination contracts.
-- Move startup landing resolution out of the composable into a coordinator or ViewModel.
-- Keep the `NavController` in `:app`.
-- Pass navigation callbacks into feature graph registration.
-- Avoid passing `MainActivity` into shared UI or feature screens.
-- Expose explicit platform capabilities, such as PiP, through small interfaces.
-
-### 8.2 Core UI and AppShell
-
-Current concern: `AppShell` contains reusable visual structures but also imports app routes, `MainActivity`, and app-specific navigation behavior.
-
-Split it into:
-
-```text
-:core:ui
- AppScreenScaffold
- AppScreenHeader
- NavigationRailVisuals
- TopBarVisuals
- Focus primitives
- Generic destination item model
-
-:app or :core:navigation
- destination list resolution
- current route mapping
- navigation callbacks
- activity/platform actions
-```
-
-The generic destination model should contain only rendering information and a stable ID. It should not contain a `NavController` or app repository.
-
-Design-system rules:
-
-- Feature code should use StreamVault wrappers for common surfaces, text, buttons, dialogs, focus indication, and spacing.
-- Raw Material 3 and TV Material usage should be limited to `:core:ui` unless a feature needs a specialized primitive.
-- Do not make a single universal component with dozens of optional parameters.
-- Keep focus behavior explicit and testable.
-
-### 8.3 Player
-
-Current concern: `PlayerScreen` observes many flows at the root and owns lifecycle, PiP, window flags, key handling, overlay focus, preparation, dialogs, player controls, and navigation.
-
-Target composition:
-
-```text
-PlayerRoute
- PlayerLifecycleHost
- PlayerScreen
- PlayerVideoSurface
- PlayerInputHost
- PlayerOverlayHost
- PlayerTransportRoute
- PlayerLiveOverlayRoute
- PlayerDialogHost
- PlayerNoticeHost
- PlayerDiagnosticsRoute
-```
-
-Recommended state groups:
-
-- `PlayerSessionState`: identity, content type, title, current media/channel/episode.
-- `PlayerPlaybackState`: ready/buffering/error/playing, video format, engine identity.
-- `PlayerControlsState`: visibility, mute, speed, aspect ratio, tracks, timers.
-- `PlayerLiveState`: channel/category/EPG overlays and numeric input.
-- `PlayerDialogState`: one sealed representation of the active modal instead of many booleans.
-- `PlayerRealtimeState`: position, duration, seek preview, and diagnostics.
-
-Specific actions:
-
-- Keep `currentPosition` and `duration` collection inside transport controls.
-- Collect diagnostics only while diagnostics are visible; the current conditional collection is a pattern to preserve.
-- Replace many dialog booleans with a sealed `PlayerModal` where modal states are mutually exclusive.
-- Move root key handling into a dedicated, testable input policy that returns actions.
-- Move PiP and window-flag behavior behind a platform interface owned by the route.
-- Keep player preparation keyed by a stable preparation identity.
-- Stop casting `PlayerEngine` to `Media3PlayerEngine` from presentation code; expose required capabilities through the player API.
-- Keep player surface creation and binding in `:player`.
-- Preserve all recovery, timeshift, and live transport policies during UI refactoring.
-
-Player acceptance checks are defined in Section 12 and include the project's long-duration live playback validation.
-
-### 8.4 Provider setup
-
-Current concern: `ProviderSetupScreen` owns more than 20 local draft fields, file import and cleanup, QR generation, Android launchers, protocol-specific forms, advanced Stalker options, pairing, sync progress, and dialogs.
-
-Target files:
-
-```text
-provider/presentation/
- ProviderSetupRoute.kt
- ProviderSetupScreen.kt
- ProviderSetupUiState.kt
- ProviderSetupAction.kt
- ProviderSetupEffect.kt
- ProviderSetupViewModel.kt
-
- forms/
- M3uProviderForm.kt
- XtreamProviderForm.kt
- StalkerProviderForm.kt
- JellyfinProviderForm.kt
- ProviderAdvancedOptions.kt
-
- import/
- ProviderImportCoordinator.kt
- ProviderImportDialogs.kt
- ProviderFileImport.kt
-
- pairing/
- ProviderPairingCard.kt
- ProviderPairingUiModel.kt
-
- components/
- ProviderTextField.kt
- ProviderSourceSelector.kt
- ProviderPolicyOption.kt
-```
-
-Specific actions:
-
-- Move the provider draft into the ViewModel, backed by `SavedStateHandle` where process restoration matters.
-- Keep only transient visual state in `rememberSaveable`, such as the currently expanded section.
-- Move file copy/cleanup and QR generation out of the composable file.
-- Use provider-specific draft models rather than one flat list of unrelated strings.
-- Convert data-layer Stalker configuration types to feature/domain UI models at the route/ViewModel boundary.
-- Keep Activity Result launchers in `ProviderSetupRoute` and communicate results through actions.
-- Split create, edit, import, pairing, and sync-result behavior into explicit states.
-- After decomposition, benchmark a Compose implementation against a ViewBinding implementation before choosing a final toolkit.
-
-### 8.5 Settings
-
-Current concern: settings presentation and `SettingsViewModel` are major graph hubs, and several settings files directly use DAOs, DataStore implementation types, and operational data entities.
-
-Target structure:
-
-```text
-settings/presentation/
- SettingsRoute.kt
- SettingsScreen.kt
- SettingsSection.kt
- SettingsAction.kt
- SettingsEffect.kt
-
- sections/
- GeneralSettings.kt
- PlaybackSettings.kt
- LiveTvSettings.kt
- GuideSettings.kt
- RecordingSettings.kt
- ProviderSettings.kt
- BackupSettings.kt
- AboutSettings.kt
-
- dialogs/
- backup/
- parental/
-```
-
-Specific actions:
-
-- Split the ViewModel by responsibility or introduce focused coordinators/use cases.
-- Replace direct DAO dependencies with domain-level query/command contracts.
-- Replace concrete `PreferencesRepository` use in presentation-facing classes with an application settings contract.
-- Keep backup file launchers and platform directory interactions at the route/platform boundary.
-- Define a stable row model for list-based settings if a classic `RecyclerView` implementation is evaluated.
-- Keep each settings section independently testable.
-
-### 8.6 Live TV and EPG
-
-Current concern: Live TV and EPG are naturally complex and should not be rewritten merely to reduce Compose count.
-
-Specific actions:
-
-- Keep Compose.
-- Split dialog ownership from the main screen state.
-- Keep preview player state isolated from channel/category list state.
-- Ensure all lazy items have stable keys and useful `contentType` values.
-- Avoid rebuilding provider lookup maps and filtered collections in composition.
-- Expose only visible-window data where possible.
-- Ensure time ticker changes invalidate only current-program/progress elements.
-- Keep focus restoration and remote input policies in dedicated helpers with tests.
-- Benchmark channel navigation, category switching, EPG horizontal movement, and preview start/stop.
-
-### 8.7 Catalog, favorites, and search
-
-Specific actions:
-
-- Consolidate shared movie/series/VOD browsing behavior behind feature-local presentation models.
-- Keep shared media cards in `:core:ui` only when they are genuinely feature-neutral.
-- Keep feature-specific card decorations and actions inside `:feature:catalog`.
-- Move filtering, ranking, grouping, and sort construction out of composables.
-- Ensure list instances are reused when content has not changed.
-- Use paging/windowed loading where lists are too large to materialize at once.
-- Standardize media item keys across combined-provider variants.
-
-## 9. Compose runtime rules
-
-These rules should be documented and enforced during the migration:
-
-- Collect state at the lowest practical owner.
-- Use `collectAsStateWithLifecycle` for UI-observed flows.
-- Do not create one collector per list item for shared state.
-- Use stable keys for every dynamic lazy item.
-- Add `contentType` to mixed lazy lists and grids.
-- Perform sorting, grouping, and large map creation outside composition.
-- Use `remember` for expensive pure derivations whose keys are correct.
-- Use `derivedStateOf` only when it reduces invalidations; do not use it as decoration.
-- Keep high-frequency reads in drawing/layout lambdas or leaf composables when possible.
-- Prefer immutable feature UI models.
-- Use immutable collections when they materially improve stability and allocation behavior.
-- Apply `@Immutable` only when every exposed property is deeply immutable by contract.
-- Avoid passing mutable collections into composables.
-- Avoid broad `CompositionLocal` state for feature data.
-- Avoid unnecessary animations on reduced-motion or constrained-device paths.
-- Do not load full-resolution artwork when the rendered size is small.
-- Keep image requests stable and sized to the display target.
-- Treat focus state as local visual state unless the feature needs to restore it across navigation.
-
-## 10. Build performance plan
-
-### 10.1 Establish scenarios
-
-Measure at least these scenarios:
-
-1. Clean `assembleDebug`.
-2. Warm no-change `assembleDebug`.
-3. Incremental change to a player composable.
-4. Incremental change to a settings composable.
-5. Incremental change to `:core:ui`.
-6. Incremental change to a domain model used by several features.
-7. Unit-test compilation for a single feature.
-8. Release or beta compilation for representative CI behavior.
-
-Use Gradle Profiler for repeatable incremental scenarios when practical. Record median and p95 across multiple runs rather than one result.
-
-### 10.2 Compose compiler reports
-
-Add reports for Compose modules during the diagnostic phase:
-
-```kotlin
-composeCompiler {
- reportsDestination = layout.buildDirectory.dir("compose_compiler")
- metricsDestination = layout.buildDirectory.dir("compose_compiler")
-}
-```
-
-Reports should remain build artifacts unless a small summarized baseline is intentionally committed. Review:
-
-- Restartable versus skippable composables.
-- Unstable parameter types.
-- Large generated composable groups.
-- Feature models that force unnecessary invalidation.
-
-### 10.3 Gradle structure
-
-- Introduce convention plugins for Android library, Compose feature, and non-Compose feature configuration after two feature modules prove the desired pattern.
-- Keep annotation processors and KSP only in modules that need them.
-- Use `implementation` by default; use `api` only for deliberate public contracts.
-- Keep module public APIs small to minimize downstream recompilation.
-- Preserve configuration-cache compatibility.
-- Avoid dynamic dependency versions.
-- Audit direct `:app` dependencies after feature extraction.
-
-### 10.4 Dependency cleanup candidates
-
-- Remove direct Media3 dependencies from `:app` after all direct Media3 use is behind `:player` APIs.
-- Remove direct Room/Retrofit/OkHttp dependencies from UI feature modules.
-- Evaluate replacing `material-icons-extended` with a small set of vector resources or core icons.
-- Centralize and audit regular Material 3 versus TV Material usage.
-- Keep debug-only tooling and LeakCanary out of release dependency graphs.
-- Compare APK and dex composition before and after dependency cleanup; do not assume source dependency removal equals packaged size reduction.
-
-## 11. Phased delivery plan
-
-Every phase should be independently mergeable. Avoid a long-running branch that moves all features at once.
-
-### Phase 0 - Baseline and guardrails
-
-Purpose: establish evidence and prevent regressions.
-
-Deliverables:
-
-- Build benchmark scenarios and recorded baseline.
-- Compose compiler reports for `:app`.
-- A `:benchmark` or equivalent Macrobenchmark module plan, followed by implementation.
-- Runtime traces for startup, dashboard scroll, Live TV navigation, EPG navigation, player controls, and settings.
-- Representative device matrix.
-- Baseline APK/dex and memory measurements.
-
-Exit criteria:
-
-- Results are reproducible across at least five warm/incremental runs and three runtime iterations per scenario.
-- Debug-only Compose overhead is not used to judge release runtime.
-- The team agrees on baseline-relative targets in Section 13.
-
-Phase 0 execution outcome (2026-08-16): the five-run build repeatability checks, five-iteration release-like cold-start benchmark, five-iteration interaction journeys, release APK/dex snapshot, emulator memory snapshot, and two-channel long-duration playback evidence are recorded in `COMPOSE_REDUCTION_BASELINE.md`. Macrobenchmark interaction journeys use the seeded debug fixture because the release-like benchmark target has no provider data; those results are diagnostic only. The CBSN channel completed the full playback window, while the F1 channel received an external HTTP 403 after initially rendering video. Full playback validation therefore remains a Phase 1 gate for playback-facing changes.
-
-### Phase 1 - Source decomposition without behavior change
-
-Purpose: make later changes reviewable and lower merge risk.
-
-Deliverables:
-
-- Split `ProviderSetupScreen.kt` into provider-specific forms, dialogs, import helpers, and components.
-- Split `PlayerScreen.kt` into lifecycle, input, overlay host, and modal host files. For the modal slice, preserve the current two composition locations and source order rather than forcing every dialog through one relocated call site: top-level program-history/MultiView dialogs remain before the root player `Box`, while controls-related dialogs remain inside it after `PlayerResumePrompt`. The detailed contract is defined in [`superpowers/specs/2026-08-17-player-modal-host-design.md`](superpowers/specs/2026-08-17-player-modal-host-design.md).
-- Split `AppNavigation.kt` into startup/external navigation policy and graph registration files.
-- Keep packages and Gradle modules unchanged.
-- Add or move tests around extracted pure policies.
-- During source-only player splits, preserve every existing visibility predicate, composition order, callback sequence, Back/input/focus rule, effect key, state owner, and ViewModel scope. In particular, do not add a new Picture-in-Picture restriction to MultiView, and preserve the audio/video-offset dialog's sync-enabled and cast-disconnected gates.
-- Do not introduce Phase 2 modal exclusivity or grouped modal state merely to simplify a Phase 1 host API. Callback-heavy, placement-preserving interfaces are acceptable temporary boundaries when they make the mechanical diff safer.
-
-Exit criteria:
-
-- No intended UI behavior change.
-- Screenshots/golden tests remain equivalent.
-- Existing unit and applicable instrumentation suites pass; if a repository baseline test fails, record the exact command and failure and keep the Phase 1 gate open rather than attributing it to the extraction without evidence.
-- Live TV validation passes if player composition or lifecycle code moved.
-- Manual TV smoke coverage passes for affected dialogs, Back/focus behavior, controls auto-hide, and Picture-in-Picture behavior. Unavailable scenarios are recorded as incomplete gates, not counted as passes.
-
-### Phase 2 - Runtime state isolation
-
-Purpose: reduce broad recomposition before modularization.
-
-Deliverables:
-
-- Player state grouped by consumer and frequency.
-- Position/duration state isolated to transport controls.
-- Player modal state represented explicitly.
-- Shared channel ticker ownership reduced.
-- Stable keys and `contentType` audited for hot lazy lists.
-- Large UI models audited with Compose compiler reports.
-- Recomposition traces compared to Phase 0.
-
-Exit criteria:
-
-- No increase in player errors or stuck playback.
-- Measurable reduction in recomposed scopes or frame cost on at least one identified hot flow.
-- No material regression on reference devices.
-
-### Phase 2.5A - Hardened live-category flow caching
-
-Purpose: isolate the live-category replay optimization from Compose state
-isolation and preserve fresh command-style reads.
-
-Deliverables:
-
-- Cache one replayed `ChannelRepository.getCategories(providerId)` flow per
- provider in the data layer, backed by Room and evicted when the provider is
- removed.
-- Keep `getCategoriesSnapshot(providerId)` as an explicit fresh one-shot API
- for command-style callers; do not make `first()` read from the replay cache.
-- Cover replay, active Room updates, duplicate suppression, preference changes,
- provider deletion/recreation, and fresh snapshot behavior with focused tests.
-- Measure category entry/re-entry or query work separately from Phase 2
- recomposition and frame-cost evidence.
-
-Exit criteria:
-
-- No stale one-shot category reads or cached data surviving provider deletion.
-- Shared collectors reuse one upstream category computation.
-- Existing grouping, counts, parental filtering, and decorative-row behavior
- remain unchanged.
-- Benchmark evidence demonstrates reduced repeated category work or faster
- category re-entry. If the available emulator cannot measure query work
- reliably, record the limitation rather than treating rendering metrics as a
- cache win.
-
-Implementation/report: [`COMPOSE_REDUCTION_PHASE2_5A_REPORT.md`](COMPOSE_REDUCTION_PHASE2_5A_REPORT.md).
-
-### Phase 2.5B - Baseline-profile and startup optimization
-
-Purpose: separate startup refactoring from baseline-profile generation and
-measure each benefit independently. This is the maintained replacement for the
-baseline-profile portion of PR #162; PR #162's generated files are not reused.
-
-Deliverables:
-
-- Capture a ten-iteration cold-start `CompilationMode.None()` baseline before
- the startup refactor, including TTID, artifacts, DEX counts, device details,
- and trace references.
-- Move process maintenance behind an internal singleton
- `AppStartupCoordinator`: process-only work starts once from
- `StreamVaultApp`, while Watch Next, launcher recommendations, and TV-input
- refresh wait for a one-shot `MainActivity` first-frame signal on televisions.
- Expensive dependencies are injected through `Provider<T>`, task failures are
- isolated with supervisor semantics, and coordinator/task trace sections are
- named.
-- Use the existing `:benchmark` module as the only profile producer. Apply the
- AndroidX Baseline Profile plugin at `1.4.1` to `:app` and `:benchmark`, merge
- one maintained profile into the main app, save generated sources only under
- `app/src/main/generated/baselineProfiles/`, and keep ordinary builds from
- generating profiles automatically.
-- Guard profile collection with seeded Home and `All Channels` assertions.
- Keep `startup` limited to cold launch and ready Home; keep Home, Live TV,
- EPG, and player-control traversal in the independently collected
- `criticalJourneys` profile.
-- Provide controlled `coldStartupNoCompilation()` and
- `coldStartupWithBaselineProfile()` benchmarks with ten iterations and
- `StartupTimingMetric`; the treatment must use
- `BaselineProfileMode.Require`.
-- Verify source separation/subset rules, compiled beta/release profiles,
- startup-optimized R8 metadata, secret-free shipping artifacts, and the
- artifact-size budget.
-
-Exit criteria:
-
-- Code and tooling checks pass, generated sources are nonempty and distinct,
- and profile packaging is proven for beta and release.
-- Post-refactor/no-profile and with-profile measurements are compared against
- the pre-2.5B baseline. Emulator results are diagnostic only.
-- A constrained physical TV completes two ten-iteration comparisons: no
- profile does not regress median/P90 TTID by more than 5%, with-profile median
- is at least 5% faster in both comparisons, with-profile P90 is not more than
- 5% slower, and traces show TV integration work after the first UI frame.
-- Keep generated profile output uncommitted until packaging and the physical
- performance gate succeed. Record open gates in the phase report.
-
-Implementation/report: [`COMPOSE_REDUCTION_PHASE2_5B_REPORT.md`](COMPOSE_REDUCTION_PHASE2_5B_REPORT.md).
-
-### Phase 3 - Core UI boundary
-
-Purpose: create a stable reusable presentation foundation.
-
-Deliverables:
-
-- Add `:core:ui`.
-- Move theme, design tokens, generic focus/input primitives, and truly generic components.
-- Separate `AppShell` visuals from app navigation and activity behavior.
-- Establish dependency rules with a build-time or static check where practical.
-
-Exit criteria:
-
-- `:core:ui` does not depend on `:app`, `:data`, or feature implementations.
-- A core UI change intentionally recompiles dependents; a feature UI change does not recompile unrelated features.
-- Golden coverage exists for critical shared TV components, with committed
- baselines rather than a first-run self-baseline.
-
-### Phase 4 - Navigation contracts
-
-Purpose: allow feature extraction without central navigation growth.
-
-Deliverables:
-
-- Add `:core:navigation` if typed contracts justify a separate module.
-- Extract typed destinations and player navigation requests.
-- Add `NavGraphBuilder` registration extensions per feature area.
-- Move startup and external navigation orchestration out of the root composable.
-
-Exit criteria:
-
-- `AppNavHost` is primarily a graph registry.
-- Features do not access the root `NavController` directly.
-- Deep links, startup player requests, return routes, and external navigation are covered by tests.
-
-Completed report: [COMPOSE_REDUCTION_PHASE4_REPORT.md](COMPOSE_REDUCTION_PHASE4_REPORT.md)
-
-Review hardening note (2026-08-25): startup landing navigation no longer waits
-for optional player-target lookup, deep-link detail returns replace the current
-detail entry when their typed target is absent from the back stack, catalog
-layout changes reconcile the current top-level route, and golden helpers now
-require checked-in assets. App and playback baseline assets are present in the
-current hardening working tree; recording-disabled golden, navigation-contract,
-and playback-overlay runs previously passed on the API 36
-`Television_1080p` emulator. The hardening slice remains pending its own commit.
-
-### Phase 5 - Feature module extraction
-
-Detailed delivery artifacts:
-
-- [Phase 5 feature extraction roadmap](COMPOSE_REDUCTION_PHASE5_ROADMAP.md)
-- [Phase 5 accepted architecture design](superpowers/specs/2026-08-24-phase-5-feature-module-extraction-design.md)
-- [Playback feature extraction implementation plan](superpowers/plans/2026-08-24-playback-feature-extraction.md)
-- [Provider feature extraction implementation plan](superpowers/plans/2026-08-26-provider-feature-extraction.md)
-- [Settings and Live regression audit](COMPOSE_REDUCTION_PHASE5_SETTINGS_LIVE_REGRESSION_AUDIT.md)
-- [Catalog feature extraction design](superpowers/specs/2026-09-02-phase-5-catalog-feature-extraction-design.md)
-- [Catalog feature extraction implementation plan](superpowers/plans/2026-09-02-catalog-feature-extraction.md)
-- [Catalog feature extraction execution report](COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md)
-- [System feature extraction implementation plan](superpowers/plans/2026-09-04-system-feature-extraction.md)
-
-Recommended extraction order:
-
-1. `:feature:playback` because it has the highest runtime and state-isolation value.
-2. `:feature:provider` because it has the largest composable source file and a clear boundary.
-3. `:feature:settings` because it is large and highly connected.
-4. `:feature:live` because Live TV and EPG share player-preview and channel concepts.
-5. `:feature:catalog` because movies, series, VOD, favorites, and search share media UI.
-6. `:feature:system` only if measurement shows value.
-
-For each module:
-
-- Move source, resources, tests, and Hilt bindings needed by the feature.
-- Expose only route registration and destination contracts.
-- Remove app-package imports from feature implementation.
-- Track temporary `:data` implementation dependencies.
-- Measure feature edit build time before and after extraction.
-
-Exit criteria:
-
-- A change inside one feature does not recompile unrelated feature source.
-- Feature tests can run independently.
-- `:app` contains orchestration and platform entry points, not feature presentation implementations.
-
-Phase 5 execution status (2026-08-27): the structural playback extraction and
-feature-boundary work through Task 9 is complete, and the Task 10 report is
-recorded in [COMPOSE_REDUCTION_PHASE5_PLAYBACK_REPORT.md](COMPOSE_REDUCTION_PHASE5_PLAYBACK_REPORT.md).
-Feature overlay connected coverage passed 7/7 (six golden cases plus the
-missing-golden guard); the app connected suite passed 22/27, with five existing
-fixture/provider/focus failures recorded in the report. The fresh ADB startup
-check confirms the current app's horizontal top
-navbar. Seeded two-channel playback produced real video and 61 unique frames
-per channel, but both channels hit repeated recoverable HLS live-window source
-errors during the required window, so runtime acceptance remains open. Full
-manual journey coverage, macrobenchmark performance comparison, and physical-
-device validation also remain open. The release-like profile gate itself passed
-after correcting the live-controls probe: the full
-`generateBaselineProfile` collection, merge, source copy, and install completed
-successfully on 2026-08-26; the focused `criticalJourneys` run also passed.
-An isolated rerun added a clean 61/61-frame Dare To Dream HLS window, but the
-French channel reproduced stalls, an unintended MPEG-TS fallback to a malformed
-URL, and a final source error (56 unique frames); the two-channel stability
-gate remains open. Detailed evidence is in the linked playback report.
-
-Provider execution status (2026-08-27): the provider module, setup/edit/import/
-pairing presentation state, resources, tests, graph registration, ownership
-cleanup, boundary checks, feature lint, unit tests, and debug assembly are
-complete. SDK-local ADB reached the connected API 36 TV emulator: focused
- provider completion (1/1), app navigation (3/3), platform compatibility
- (4/4), golden (1/1), and playback overlay (7/7) checks passed. Manual source switching, D-pad focus, document
-choosers, Drive account-picker handoff, QR pairing start/stop, and a sanitized
-cold-start route smoke check, and seeded existing-provider edit/cancel journey
-also passed without a fatal app error; full provider completion journeys remain
-open. The release-like profile regeneration was rerun successfully on the now-idle
-host (10 profile tests passed, eight unrelated macrobenchmarks skipped, stale
-app-provider paths removed, and `:app:assembleRelease` passed). Paired
-incremental-performance samples are still open because the provider extraction
-and related navigation/golden changes are uncommitted; a valid before/after
-comparison requires a committed post-extraction ref. See
-[COMPOSE_REDUCTION_PHASE5_PROVIDER_REPORT.md](COMPOSE_REDUCTION_PHASE5_PROVIDER_REPORT.md).
-
-Live execution status (2026-08-30): the Home Live TV and EPG presentation
-extraction is structurally complete under `:feature:live`, with the app
-retaining route/platform/player-request/MultiView adapters. The Live boundary,
-feature checks, translated-resource parity audit, six reviewed feature golden
-baselines, connected suite, populated MultiView planner journey, and required
-two-channel long-playback protocol are recorded in
-[COMPOSE_REDUCTION_PHASE5_LIVE_REPORT.md](COMPOSE_REDUCTION_PHASE5_LIVE_REPORT.md).
-The follow-up device passes also cover RTL/reduced-motion interaction, D-pad
-activation of category/channel/Guide search inputs, quick-filter form input,
-hidden category/channel restore, populated Favorites reorder/save-order, and
-the populated MultiView placement/removal/replacement/clear and two-slot launch
-paths, plus reversible quick-filter save/reload/remove persistence. The full
- fixture-dependent Home/EPG journey matrix, formal paired performance and
- clean-build guardrails remain open. Current-checkout clean/warm samples and
- Beta/Release packaging are recorded. Profile generation now passes on the
- seeded API 36 TV emulator, with fresh sources containing zero stale app
- Home/EPG descriptors and nonzero `feature/live` descriptors. The explicit
- Macrobenchmark selector now runs all eight journeys: 8/8 passed, 0 skipped,
- including both 10-iteration release cold-start methods and six seeded
- interaction journeys; the emulator execution gate is closed. Matched paired
- performance and physical-device measurements remain open. The app-owned Live
- route golden was reviewed and regenerated with its connected assertion
- passing. Catalog is now structurally
- extracted under `:feature:catalog`; its execution report records passing
- connected goldens, a public-M3U Home/Live/Movies/Series/Search smoke pass,
- and a diagnostic temporary Xtream fixture run covering 63-item VOD/series
- catalogs, VOD/series details, episodes, saved filters, and cross-type Search.
- The checked-in `tools/catalog_xtream_fixture.py` now provides a rerunnable
- development seed, and `tools/catalog_connected_validation.py` provides a
- semantic ADB journey for the fixture-backed Home/Movies/Series/detail/
- favorite/saved/Search path, Movies full-library entry/back,
- `Load more (60/63)` → `Pagination Movie 63`, Infinite scroll off/on
- restoration, and the Settings-owned Dashboard shelf customization
- cancel/save/reset flow. Series pagination, browse reorder, movie download
- completion, and direct Favorites test-host rendering/Back cancellation are
- covered; Catalog RTL/large-text semantics and reduced-motion instrumentation
- also pass. Cast receiver chooser, touch/device variants, and accessibility
- service journeys remain open. The successful diagnostic Dashboard benchmark
- rerun after populating a scrollable seeded Home shelf, the open paired
- benchmark gate, and refreshed release-like profiles remain explicitly
- tracked; current app lint now passes with the maintained baseline covering
- 374 existing findings. System is now
- structurally extracted under `:feature:system`; its connected acceptance
- passes 19/19, paired incremental thresholds pass, and the steady-state
- clean/warm guardrail passes. The System profile refresh is promoted through
- the supported Gradle workflow; physical-device/accessibility gates and
- authorized production journeys remain open. Playback, Provider, and Settings
- gates remain governed by their separate reports.
-
-### Phase 6 - Optional Views migrations
-
-Purpose: reduce Compose where evidence and maintenance characteristics support it.
-
-Candidate order:
-
-1. Backup/import flows.
-2. Provider setup forms.
-3. Settings sections.
-4. Simple onboarding.
-
-For each candidate, build a small benchmarkable vertical slice first. Compare:
-
-- Initial render and interaction frame time.
-- Memory and allocations.
-- TV focus behavior.
-- Accessibility.
-- Source complexity and testability.
-- Incremental build impact.
-- Navigation/interoperability cost.
-
-Exit criteria:
-
-- Conversion demonstrates a measurable benefit or a clear maintenance win.
-- Converted modules omit the Compose plugin when they contain no Compose code.
-- The change does not create a fragile Compose/View navigation boundary.
-
-### Phase 7 - Dependency and API cleanup
-
-Purpose: finish the architecture rather than leaving transitional dependencies permanent.
-
-Deliverables:
-
-- Domain-facing settings/preferences contracts.
-- Player capability API without presentation casts to `Media3PlayerEngine`.
-- Provider setup UI models without direct Stalker implementation types.
-- Removal of unused direct dependencies from `:app` and feature modules.
-- Final package and naming cleanup.
-- Updated architecture documentation and diagrams.
-
-Exit criteria:
-
-- No presentation package imports DAOs or remote provider implementation classes.
-- Concrete Media3 use is confined to player/platform implementation code.
-- Temporary migration dependencies are removed or explicitly documented with owners.
-
-Phase 7 checkpoint 1 - provider boundary (complete):
-
-- `:feature:provider` depends on `:domain`, `:core:ui`, and `:core:navigation`, with no direct `:data` dependency.
-- Provider setup owns its editable Stalker advanced-options models and compatibility-preserving JSON codec.
-- Reusable Stalker compatibility metadata and provider input normalization are domain APIs rather than remote-provider implementation details.
-- Data-layer provider and credential exceptions implement the domain-facing `ProviderSetupFailure` contract; presentation maps that contract to user-facing messages without importing concrete data exceptions.
-- The automated provider boundary rejects future Gradle or source-level data coupling.
-
-Current dependency direction for this checkpoint:
-
-```text
-:feature:provider ---> :domain <--- :data
- | |
- +----> :core:ui +----> concrete provider implementations
- +----> :core:navigation
-```
-
-Validation recorded for this checkpoint:
-
-- `:domain:test`
-- `:feature:provider:testDebugUnitTest`
-- Focused data tests for Stalker compatibility, credential handling, and Xtream requests
-- `:data:compileDebugKotlin :data:compileDebugUnitTestKotlin`
-- `:app:assembleDebug`
-
-The provider, settings, and playback boundary tasks now use typed task inputs and configuration-time dependency snapshots, so the boundary verification path is configuration-cache compatible. A warm rerun of all three verification tasks reused the configuration cache successfully on 2026-09-06.
-
-Phase 7 checkpoint 2 - settings boundary (complete):
-
-- `:feature:settings` depends on `:domain`, `:core:ui`, `:core:navigation`, and `:player`, with no direct `:data` dependency.
-- Settings presentation consumes the domain `SettingsPreferences` and `SettingsOperations` contracts; Room, DataStore, and provider-sync implementations remain in `:data`.
-- The automated settings boundary rejects direct `com.streamvault.data` imports and fully qualified references in Kotlin and Java presentation sources while retaining fixtures for both forms.
-
-Current dependency direction for this checkpoint:
-
-```text
-:feature:settings ---> :domain <--- :data
- | |
- +----> :core:ui +----> concrete persistence and provider-sync implementations
- +----> :core:navigation
- +----> :player
-```
-
-Phase 7 checkpoint 3 - player capability boundary (automated checkpoint complete):
-
-- `PlayerEngine` is the approved capability API for playback presentation code; `:feature:playback` retains its `:player` dependency without reaching into `Media3PlayerEngine`.
-- Nullable subtitle text (`String?`) and the player-owned `PlayerPcmEncoding` model replace Media3 subtitle cue and PCM encoding values at the presentation boundary.
-- `:feature:playback` has no direct Media3 dependency and no `Media3PlayerEngine` or `androidx.media3` source references.
-- The automated playback boundary rejects future direct Media3 or data-layer declarations and Kotlin or Java source regressions while retaining the existing app and root-navigation guards.
-
-Current dependency direction for this checkpoint:
-
-```text
-:feature:playback ---> :player ---> Media3 implementation
- |
- +----> PlayerEngine capability API
-```
-
-Phase 7 checkpoint 4 - playback data boundary (complete):
-
-- `:feature:playback` depends only on `:core:navigation`, `:core:ui`, `:domain`, and `:player`; the transitional `:data` dependency is removed.
-- `PlayerPreferences` owns the player/multiview preference contract; `PreferencesRepository` remains the data-backed implementation bound by the app.
-- `PlayerPlaybackResolver` exposes domain `ResolvedPlayback` results; the app adapter translates data resolver metadata and known provider/credential failures.
-- The internal playback token codec used by live recovery is domain-owned, so playback has no `com.streamvault.data` imports in main or test source.
-
-Current dependency direction for this checkpoint:
-
-```text
-:feature:playback ---> :domain <--- :data
- | ^ |
- +----> :player | +----> persistence/provider implementations
- |
- :app adapter and Hilt bindings
-```
-
-Automated validation recorded for this checkpoint on 2026-09-06:
-
-- Cold and warm `:feature:playback:verifyFeaturePlaybackBoundary --configuration-cache` runs passed; the warm run reused configuration cache.
-- Playback boundary, domain, data, settings, player, and live unit-test suites passed with zero failures/errors/skips (1,742 tests total).
-- `:app:testDebugUnitTest --tests com.streamvault.app.playback.AppPlayerPlaybackResolverTest` passed.
-- `:app:assembleDebug` passed.
-
-Live TV device acceptance passed on 2026-09-06 using the `Television_1080p` AOSP TV emulator. Two channels were each captured for 61 screenshots with a two-second sleep interval between captures; both had 61 unique frame hashes and a healthy `PLAYING` media session with `error=null`. Fresh startup logs recorded HLS prepare and first-frame success for both channels. Sanitized evidence is recorded in [`validation/phase7_player_capability/README.md`](../validation/phase7_player_capability/README.md) and [`validation/phase7_player_capability/live-validation.log`](../validation/phase7_player_capability/live-validation.log).
-
-The manual player smoke gate is now closed by the live-control focus and rewind
-validation recorded in [`validation/phase7_player_capability/README.md`](../validation/phase7_player_capability/README.md).
-External PR packaging and the broader baseline follow-up items remain listed in
-the plan checklist.
-
-### Phase 7 success-target audit (2026-09-07)
-
-The initial targets in Section 13 were reviewed against the committed baseline,
-the paired Phase 5 System measurements, and the Phase 7 playback acceptance
-evidence. A target is marked **open** when the available result is diagnostic or
-has no comparable before/after measurement; this is not a failure result.
-
-| Metric | Current evidence | Status |
-|---|---|---|
-| Incremental build after feature-only UI edit | System source edit improved 64.56% against its paired warm baseline (`validation/phase5_system/performance-after.md`). | Met for the measured feature extraction; no new Phase 7-only pair was collected. |
-| Incremental unit-test compile for extracted feature | System unit-test compile improved 57.88% against its paired warm baseline. | Met for the measured feature extraction; no new Phase 7-only pair was collected. |
-| Clean debug build | Paired current/rollback clean invocation was faster; cache reuse is documented. | Guardrail met; uncached machine-build improvement is not claimed. |
-| Warm no-change build | Steady-state median was 2.12% slower, within the 10% guardrail. | Met. |
-| Slow/janky frames in selected hot flows | Macrobenchmark journeys pass, but the available run is diagnostic and has no matched post-Phase-7 comparison. | Open measurement gate. |
-| Player controls open latency | The controls journey passes, but no comparable latency measurement is recorded. | Open measurement gate. |
-| Startup time | Release cold-start/profile journeys pass, but no paired before/after startup comparison is recorded for this phase. | Open measurement gate. |
-| Memory during Live TV/player | Baseline PSS exists; no paired post-Phase-7 memory sample is committed. | Open measurement gate. |
-| Player stability | Two channels completed 61 two-second captures each with changing frames, `PLAYING(3)`, `error=null`, HLS first-frame success, and no fatal/stuck/fallback evidence. | Met for the validated Phase 7 channels; broader provider coverage remains separate. |
-| `:app` presentation source | A 2026-09-07 snapshot counted 345 Kotlin files / 79,288 lines under `feature/*/src/main/java` versus 20 files / 3,433 lines under `app/src/main/java/com/streamvault/app/ui`. | Met for this source scope; app shell/navigation/platform UI outside `app/ui` remains app-owned. |
-
-Conclusion: the Phase 7 implementation and playback acceptance gates are
-supported by evidence. The remaining success-target work is measurement
-follow-up, not a newly identified playback or dependency-boundary defect.
-
-## 12. Validation strategy
-
-### 12.1 Build validation
-
-Record for every architectural phase:
-
-- Clean build median and p95.
-- Warm no-change build median.
-- Incremental feature-edit build median and p95.
-- Kotlin compilation and KSP task durations.
-- Number of tasks executed versus up-to-date/from-cache.
-- Gradle configuration time.
-- CI build duration when available.
-
-Do not compare one clean build against one warm build.
-
-### 12.2 Runtime benchmark flows
-
-Benchmark at least:
-
-- Cold startup to welcome/home content.
-- Warm startup to the configured landing destination.
-- Dashboard vertical and horizontal scrolling.
-- Live TV category change and repeated channel focus movement.
-- Live preview start, stop, and replacement.
-- EPG horizontal time movement and vertical channel movement.
-- Player controls open/close.
-- Player channel list and EPG overlay navigation.
-- Repeated channel zapping.
-- MultiView open, focus movement, and slot replacement.
-- Provider setup typing and provider-tab switching.
-- Settings scrolling and dialog opening.
-
-Collect:
-
-- Frame timing and jank.
-- Main-thread slices in Perfetto.
-- Composition/recomposition traces where available.
-- Java/Kotlin allocations.
-- PSS and heap behavior on constrained devices.
-- Time to initial display and time to fully drawn.
-
-### 12.3 Device matrix
-
-At minimum:
-
-- A constrained Android TV/Google TV class device.
-- A Fire TV class device when available.
-- The Television 1080p emulator used by existing tests.
-- A phone/tablet reference device if those form factors remain supported.
-- One modern high-performance device as a regression control.
-
-### 12.4 Player and Live TV validation
-
-Any phase that changes player composition, lifecycle, state collection, surface ownership, navigation, or overlays must use the repository's full live playback protocol:
-
-- Validate more than one live channel.
-- Capture at a 2-second cadence.
-- Capture at least 45 screenshots for about 90 seconds; prefer 61 screenshots for about two minutes.
-- Confirm changing screenshot hashes through the full window.
-- Confirm the media session remains `PLAYING` with `error=null`.
-- Check logs for fatal player errors, stuck-player timeout, and unintended MPEG-TS fallback.
-- Record HLS prepare/read/first-frame or expected recovery evidence.
-- Report channel names, screenshot count, interval, unique hash count, media-session result, and log findings.
-
-Build success, installation, launch, or a single screenshot is not sufficient player validation.
-
-### 12.5 Functional regression coverage
-
-- Navigation and deep links.
-- Startup landing preference.
-- Provider create/edit/import.
-- Backup and restore.
-- TV remote focus and back behavior.
-- Mouse and touch behavior.
-- RTL layout and navigation.
-- Reduced motion.
-- Accessibility labels and focus order.
-- PiP.
-- Cast.
-- TV input and recommendations.
-- Parental controls.
-- Playback recovery and timeshift.
-
-## 13. Success metrics
-
-Phase 0 should produce the final approved numbers. The following are initial target ranges, all relative to the recorded baseline:
-
-| Metric | Initial target | Guardrail |
-|---|---:|---:|
-| Incremental build after feature-only UI edit | At least 25% faster | Must not regress |
-| Incremental unit-test compile for extracted feature | At least 20% faster | Must not regress |
-| Clean debug build | Improvement is optional | No more than 10% slower without documented reason |
-| Warm no-change build | Maintain or improve | No more than 10% slower |
-| Slow/janky frames in selected hot flows | At least 20% reduction where a hotspot is confirmed | No regression on other benchmark flows |
-| Player controls open latency | At least 10% improvement if currently measurable | No visible focus delay |
-| Startup time | Maintain or improve during structural phases | No more than 5% regression |
-| Memory during Live TV/player | Maintain or improve | No sustained growth or new leak |
-| Player stability | Equivalent or better | Zero new fatal/stuck/fallback regressions |
-| `:app` presentation source | Majority moved to feature modules | `:app` must not regain feature implementations |
-
-Targets should be revised if Phase 0 shows measurement noise larger than the proposed improvement.
-
-## 14. Proposed pull-request sequence
-
-This sequence keeps reviews focused and creates rollback points:
-
-1. Add build/runtime benchmark infrastructure and capture baseline.
-2. Add Compose compiler report configuration and stability audit notes.
-3. Extract pure player input policies with tests, and extract modal composition through placement-preserving hosts with explicit visibility and callback contracts.
-4. Split `PlayerScreen` files without behavior changes.
-5. Isolate player real-time state and validate Live TV fully.
-6. Split provider setup files without behavior changes.
-7. Move provider draft and import behavior to explicit route/ViewModel state.
-8. Add `:core:ui` and move only theme/design/focus primitives.
-9. Separate AppShell visuals from navigation and activity dependencies.
-10. Split navigation policy and feature graph registration.
-11. Extract `:feature:playback`.
-12. Extract `:feature:provider`.
-13. Extract `:feature:settings`.
-14. Extract `:feature:live`.
-15. Extract `:feature:catalog`.
-16. Run Views proof of concept for backup/import or one provider form.
-17. Make toolkit decision per candidate feature.
-18. Remove transitional data/player implementation dependencies.
-19. Final performance report and architecture documentation update.
-
-Large moves should use history-preserving file moves where possible. Mechanical moves and behavior changes should be separate PRs.
-
-## 15. Risks and mitigations
-
-| Risk | Mitigation |
-|---|---|
-| Feature modules increase clean-build task overhead | Optimize for incremental builds, keep module count modest, measure clean-build guardrail |
-| Core UI becomes a new god module | Keep it generic, small, stable, and dependency-restricted |
-| Moving state changes behavior | First extract files without behavior changes; then change state ownership with tests |
-| Player refactor introduces playback regressions | Separate UI architecture from recovery policy; use full multi-channel long-duration validation |
-| Compose/View interop complicates navigation | Convert only after a vertical-slice proof and explicit navigation contract |
-| `@Immutable` hides mutable behavior | Annotate only deeply immutable contracts and validate with compiler reports/tests |
-| Feature extraction exposes too many APIs | Default to internal visibility and expose route registration plus small contracts |
-| Temporary feature-to-data dependencies become permanent | Track them in the migration PR and require removal in Phase 7 |
-| Golden tests become brittle during file moves | Preserve semantics and test tags; avoid visual changes in structural PRs |
-| Long-running refactor causes merge conflicts | Use small sequential PRs and move one feature at a time |
-
-## 16. Rollback strategy
-
-- Every PR must compile and pass its scoped validation independently.
-- Preserve old route entry points until a feature module is fully wired and tested.
-- For state isolation, keep adapters that can map old ViewModel flows into the new state model during transition.
-- For Views experiments, keep the existing Compose destination behind a debug or development switch until the comparison is complete.
-- Do not delete the original implementation in the same commit that introduces an unvalidated replacement.
-- Playback policy and recovery code must remain unchanged during presentation-only refactors unless a separate bug-fix scope explicitly authorizes it.
-
-## 17. Definition of done
-
-The modernization is complete when:
-
-- `:app` is primarily the platform and navigation composition root.
-- Major UI areas are isolated in a small number of feature modules.
-- Route, Screen, UiState, Action, and Effect responsibilities are consistently applied.
-- Player high-frequency state is isolated from infrequently changing screen state.
-- Provider setup no longer owns a flat collection of draft fields and platform work inside one composable.
-- Settings presentation does not directly depend on DAOs or concrete data implementations.
-- Feature presentation does not cast to `Media3PlayerEngine`.
-- `:core:ui` is generic and stable.
-- Navigation is registered per feature and the root nav host is readable.
-- Compose remains where it provides clear product value.
-- Views are used only where measurement or maintenance evidence supports them.
-- Incremental build targets are met or documented with evidence.
-- Runtime benchmark flows show no regressions and confirmed hotspots improve.
-- Full Live TV/player validation passes after all playback-facing phases.
-- Documentation and graph output reflect the final architecture.
-
-## 18. Immediate next actions
-
-- [ ] Approve this plan's target module grouping.
-- [ ] Assign an owner for Phase 0 measurements.
-- [x] Capture initial clean-task-graph and warm Gradle baselines. See `docs/COMPOSE_REDUCTION_BASELINE.md`.
-- [x] Add Compose compiler reports and inspect player/provider stability output. See `docs/COMPOSE_REDUCTION_BASELINE.md`.
-- [x] Complete the first no-behavior-change slice by extracting `PlayerControlsOverlayHost`.
-- [x] Extract and unit-test the player Back-navigation priority policy without changing event handling.
-- [x] Extract player lifecycle/window effects into `PlayerLifecycleHost` without changing effect keys or callbacks.
-- [x] Extract and unit-test the player root input decision policy while keeping Android event adaptation, state mutation, focus, pointer, and Back execution in `PlayerScreen.kt`.
-- [x] Add event-time freshness guards for player input and Back decisions, including numeric-channel buffer regressions.
-- [x] Complete the second no-behavior-change `ProviderSetup` decomposition slice; package it as a review PR after the remaining manual gates.
-- [x] Complete the Phase 1 `AppNavigation` policy, adapter, external-request, and graph-registration split.
-- [x] Run a manual app smoke test for player launch, controls, remote/back handling, seeking, and overlay actions; see `validation/phase7_player_capability/README.md`.
-- [x] Run full multi-channel Live TV validation before marking the player phase complete; see `validation/phase7_player_capability/README.md`.
-- [x] Add or define the Macrobenchmark module and constrained-device benchmark flows.
-- [ ] Create the first no-behavior-change PR for `PlayerScreen` decomposition after the manual smoke test.
-- [ ] Package the completed second no-behavior-change `ProviderSetupScreen` decomposition as a review PR.
-- [x] Revisit the initial success targets after baseline data is available; see the Phase 7 success-target audit above.
-
-## 19. PR #162 assessment
-
-PR [#162](https://github.com/Davidona/StreamVault-IPTV/pull/162) is directionally aligned with this plan, but it should be reviewed as four separate changes. It currently targets `master`; if `develop` is the integration branch, retarget it or apply the accepted commits to `develop` before synchronizing `feature/improveCompose`.
-
-| PR commit/change | Plan relationship | Recommendation |
-|---|---|---|
-| Split Home preview state from `HomeUiState` | Directly advances Phase 2 runtime state isolation and the current route/state boundary work | Good merge candidate after Home preview, fullscreen handoff, cleanup, and loading/error tests pass |
-| Cache live category flows per provider | Useful data/runtime optimization, but not specifically a Compose refactor | Keep as a separate data-layer change; add replay, refresh, preference-change, provider-removal, and one-shot `first()` tests before merging |
-| `compose-stability.conf` for all `domain.model` classes | Supports the stability goal, but the wildcard is broader than the proof currently provides | Do not accept unchanged; audit every UI-facing model and replace the wildcard with a verified allowlist or genuinely immutable UI models |
-| Baseline profile module and generated profiles | Fits Phase 2.5B startup/profile optimization | Keep the Gradle/profile-installation concept, but use the existing `:benchmark` producer, separate startup from critical journeys, regenerate on current code, and validate release/beta behavior before committing output |
-
-The Home preview split is the part most directly connected to this Compose plan. The stability configuration is the highest correctness risk because Kotlin `List`/`Map` properties are not immutable by type, even when the containing model uses `val` properties. The category cache also needs explicit freshness tests because `shareIn(replay = 1)` changes the behavior of callers that use `first()`.
-
-The four generated profile files in PR #162 are identical: beta and release contain the same rules, and `baseline-prof.txt` is also identical to `startup-prof.txt`. The generator currently sets `includeInStartupProfile = true` around startup, top-navigation traversal, and list scrolling, causing the whole journey to be treated as startup code. Startup-profile generation should cover only startup-critical paths; general Home/navigation/scrolling journeys belong in the Baseline Profile without being promoted into the Startup Profile.
-
-Recommended integration order:
-
-1. Merge the accepted Home preview change into `develop` after adding focused tests.
-2. Review the category cache as an independent data PR.
-3. Narrow and validate the stability configuration separately.
-4. Add baseline-profile generation separately after the benchmark/device setup is available.
-5. Synchronize `feature/improveCompose` with the resulting `develop` and rerun Phase 0 measurements.
-
-Execution update (2026-08-22): the category cache is implemented and reported
-as Phase 2.5A, and the startup coordinator/profile tooling is implemented and
-reported as Phase 2.5B. The API 36 TV emulator proves profile collection and
-shipping packaging for beta, release APK, and release AAB, and provides a
-diagnostic startup benefit. The Phase 2.5B physical-device and release-APK-size
-gates remain open; generated profile output is therefore not yet a
-release-approved artifact.
-
-Phase 2.5A measurement update (2026-08-23): the category re-entry
-Macrobenchmark now counts category presentation builds plus shared-flow
-upstream starts and stops. The cache keeps each provider's Room observation
-alive for 30 seconds after its last subscriber, while provider removal cancels
-the entry immediately. The deterministic unit test confirms that a subscriber
-returning within that window replays the latest value without a second upstream
-start.
-
-The connected seeded-emulator task-reset journey was rerun after a clean debug
-process reset. Five iterations captured `2/2/1` for build/upstream-start/
-upstream-stop (min/median/max) and 83/87/90 frames. A controlled diagnostic run
-with the stop timeout set to zero captured the same `2/2/1` lifecycle counts,
-so this Activity-task boundary did not demonstrate a measurable cache win. The
-result is useful lifecycle evidence but not a quantified runtime reduction; a
-direct Room query-count or category-ready-latency benchmark remains open. The
-emulator and debuggable seeded target are diagnostic only, and the separate
-`.debug` interaction package is intentionally used to provide deterministic
-seeded data.
-
-### 19.1 Baseline profile maintenance policy
-
-- Treat the generator tests and reusable critical-user-journey helpers as the source of truth. Never hand-maintain generated profile rules.
-- Cherry-pick or recreate the baseline-profile module and build configuration when this phase begins; regenerate the text files from the then-current `develop` code instead of carrying PR #162's generated files forward.
-- Keep startup generation separate from general performance journeys. Startup, Home, Live TV/EPG, and player-controls journeys should be explicit and independently reviewable.
-- Use a pinned Gradle-managed device for reproducible generation where practical. Use physical constrained-TV hardware for before/after benchmarking.
-- If beta and release execute the same code paths, set `mergeIntoMain = true` and maintain one generated profile. Keep variant-specific profiles only when their code or journeys genuinely differ.
-- Regenerate before release candidates and after material startup/navigation changes, feature-module moves, package or method-signature changes, major Compose/Kotlin/AGP upgrades, or changes to the profiled journeys.
-- Commit generated output only after the generator succeeds, the release artifact contains the compiled profile, and macrobenchmarks show a benefit without unacceptable APK, memory, or compilation cost.
-- Do not enable profile generation on every ordinary developer build. A dedicated release-candidate or scheduled CI job avoids doubling normal build time while still keeping profiles current.
-
-Official maintenance references: [Create Baseline Profiles](https://developer.android.com/topic/performance/baselineprofiles/create-baselineprofile), [Configure Baseline Profile generation](https://developer.android.com/topic/performance/baselineprofiles/configure-baselineprofiles), and [Startup Profiles](https://developer.android.com/topic/performance/startupprofiles/dex-layout-optimizations).
-
-## Appendix A - Initial file mapping
-
-The following is a proposed migration map. Exact names may change during implementation.
-
-| Current file or package | Proposed destination |
-|---|---|
-| `MainActivity.kt` | `:app` |
-| `navigation/AppNavigation.kt` | `:app` plus feature `navigation/*NavGraph.kt` files |
-| `ui/theme` | `:core:ui` |
-| `ui/design` | `:core:ui` |
-| Generic parts of `ui/interaction` | `:core:ui` |
-| Generic parts of `ui/components` | `:core:ui` |
-| App-specific parts of `ui/components/shell` | `:app` or feature navigation shell |
-| `ui/screens/player` | `:feature:playback` |
-| `ui/screens/multiview` | `:feature:playback` |
-| `ui/screens/provider` | `:feature:provider` |
-| `ui/screens/settings` | `:feature:settings` |
-| `ui/screens/home` | `:feature:live` |
-| `ui/screens/epg` | `:feature:live` |
-| `ui/screens/movies` | `:feature:catalog` |
-| `ui/screens/series` | `:feature:catalog` |
-| `ui/screens/vod` | `:feature:catalog` |
-| `ui/screens/favorites` | `:feature:catalog` |
-| `ui/screens/search` | `:feature:catalog` |
-| `ui/screens/dashboard` | `:feature:catalog` initially; reconsider if it becomes an app-level aggregator |
-| `ui/screens/welcome` | `:feature:system` or `:app` initially |
-| `ui/screens/downloads` | `:feature:system` or dedicated feature if it grows |
-| `ui/screens/plugins` | `:feature:system` or dedicated feature if plugin UI grows |
-
-## Appendix B - Suggested local commands
-
-Examples use PowerShell from the repository root.
-
-Build profile:
-
-```powershell
-.\gradlew.bat --stop
-.\gradlew.bat clean :app:assembleDebug --profile --no-daemon
-.\gradlew.bat :app:assembleDebug --profile
-```
-
-Focused checks during early source decomposition:
-
-```powershell
-.\gradlew.bat :app:compileDebugKotlin
-.\gradlew.bat :app:testDebugUnitTest
-.\gradlew.bat :app:lintDebug
-```
-
-After feature extraction, run the feature's checks directly before the app integration checks:
-
-```powershell
-.\gradlew.bat :feature:playback:compileDebugKotlin
-.\gradlew.bat :feature:playback:testDebugUnitTest
-.\gradlew.bat :app:assembleDebug
-```
-
-After modifying code files, refresh the repository graph:
-
-```powershell
-graphify update .
-```
-
-## Appendix C - Architecture review checklist
-
-Use this checklist for every migrated feature:
-
-- [ ] The feature has one clear Route entry point.
-- [ ] The Screen can render without Hilt or a `NavController`.
-- [ ] UI state is immutable by contract.
-- [ ] High-frequency state is isolated near its consumer.
-- [ ] Lazy collections use stable keys.
-- [ ] Mixed lazy content uses `contentType`.
-- [ ] Business sorting/filtering is outside composition.
-- [ ] Android launchers and platform services are at the route boundary.
-- [ ] No DAO or remote implementation type appears in presentation APIs.
-- [ ] No concrete Media3 engine cast appears in presentation code.
-- [ ] Shared components are truly generic before moving to `:core:ui`.
-- [ ] Navigation is expressed through callbacks or destination contracts.
-- [ ] Focus, accessibility, RTL, touch, mouse, and remote behavior are tested.
-- [ ] Build impact is measured.
-- [ ] Runtime impact is measured for hot flows.
-- [ ] Player-facing changes pass full live playback validation.
diff --git a/docs/COMPOSE_REDUCTION_BASELINE.md b/docs/COMPOSE_REDUCTION_BASELINE.md
deleted file mode 100644
index 42e0bada..00000000
--- a/docs/COMPOSE_REDUCTION_BASELINE.md
+++ /dev/null
@@ -1,148 +0,0 @@
-# Compose Reduction Baseline
-
-Date: 2026-08-16
-Repository state: existing user changes in `gradle.properties` and `player/build.gradle.kts` were preserved
-Purpose: Phase 0 baseline for UI architecture and Compose reduction work
-
-## Execution update
-
-The first behavior-preserving decomposition has been applied: `PlayerControlsOverlayHost` now lives in `app/src/main/java/com/streamvault/app/ui/screens/player/PlayerControlsOverlayHost.kt`, while its inputs, callbacks, and `PlayerControlsOverlay` wiring remain unchanged. `PlayerScreen.kt` decreased from 1,561 to 1,432 lines.
-
-Verification after the extraction:
-
-- `:app:compileDebugKotlin` — passed.
-- `graphify update .` — passed; graph refreshed to 12,893 nodes and 24,601 edges.
-- `:app:testDebugUnitTest` — passed after the test fixture was updated to provide the already-required `m3uClassificationRepository` mock. Existing coroutine opt-in warnings remain.
-
-This document records the Phase 0 baseline and the first Phase 1 source decomposition. Phase 0 is complete for the available emulator/development environment; physical-device and provider-auth limitations are recorded below.
-
-## Phase 0 completion measurements
-
-Validation environment:
-
-- SDK: `E:\androidSdk`
-- Emulator: `emulator-5554`, `AOSP_TV_on_x86`, 1920x1080, API 36
-- Release-like benchmark target: `com.streamvault.app`, non-debuggable `benchmark` build type
-- Seeded interaction fixture: `com.streamvault.app.debug`
-
-### Repeated builds
-
-| Scenario | Runs (seconds) | Result |
-|---|---:|---|
-| Warm `:app:assembleDebug --no-daemon` | 286.610, 12.046, 11.673, 11.445, 11.692 | Passed; run 1 included variant/configuration work, subsequent runs were stable |
-| Touch `PlayerScreen.kt` + `:app:compileDebugKotlin --no-daemon` | 19.671, 9.334, 9.428, 9.447, 9.396 | Passed; four stable incremental runs were about 9.3–9.4s |
-
-### Runtime benchmark and artifacts
-
-- Macrobenchmark module builds and installs successfully.
-- Release-like cold startup: 5 iterations, median `975.8 ms`, min `888.4 ms`, max `1,487.6 ms` on the emulator. Android reported the expected emulator warning; do not use this as a physical-device target.
-- Seeded interaction journeys: dashboard scroll, Live TV navigation, EPG navigation, player-controls navigation, and settings navigation all completed 5 iterations. Dashboard diagnostic snapshot: median 87 frames, P50 frame overrun 9.5 ms, P90 27.4 ms, P99 79.2 ms.
-- Release APK: `18,138,676` bytes; 3 dex files totaling `14,015,356` bytes.
-- Release cold `am start -W`: `648 ms` on the same emulator.
-- Release-like process memory after 5 seconds: `91,441 kB` total PSS.
-- Seeded debug process memory after 5 seconds: `265,167 kB` total PSS. This is diagnostic only and includes debug/runtime overhead.
-
-### Live TV long-run evidence
-
-Each channel used 61 screenshots at a 2-second cadence (120 seconds nominal), with SHA-256 frame-hash progression checked.
-
-| Channel | Frames | Unique hashes | Final media session | Result |
-|---|---:|---:|---|---|
-| F1 Channel | 61 | 32 | `ERROR`, `Source error` | Failed after initially rendering video; sanitized logs show provider HTTP 403 and no recovery candidate |
-| US CBSN New York (D) | 61 | 61 | `PLAYING`, `error=null` | Passed; no fatal/stuck-player error observed |
-
-The F1 result is an external stream/authentication failure, not evidence of a Compose regression. Full multi-channel playback validation remains required after playback-facing Phase 1 changes.
-
-## Build baseline
-
-| Scenario | Result | Cache/task details |
-|---|---:|---|
-| `clean :app:assembleDebug --profile --no-daemon` | 2m15.47s | 93 actionable tasks; 40 executed; 53 from cache; configuration cache stored |
-| Warm `:app:assembleDebug --profile --no-daemon` | 36.264s | 89 actionable tasks; 1 executed; 88 up-to-date; configuration cache reused |
-| `:app:compileDebugKotlin --profile --no-daemon` | 33s command duration | 49 actionable tasks; 1 executed; 48 up-to-date |
-| Forced `:app:compileDebugKotlin --rerun-tasks` | Incomplete after 184s timeout | Compose reports were emitted; duration is not a valid benchmark |
-
-The clean command still reused Gradle/build-cache outputs for several compilation tasks. It should therefore be described as a clean task-graph run with cache reuse, not a fully uncached machine build.
-
-The warm command includes the cost of starting a single-use Gradle process because `--no-daemon` was used. Future comparisons should include both a normal developer-daemon scenario and a CI-style no-daemon scenario.
-
-## Packaging snapshot
-
-| Artifact | Size |
-|---|---:|
-| `app/build/outputs/apk/debug/app-debug.apk` | 50,778,921 bytes |
-
-This is a debug APK and is not an install-size or release-size target. A release APK and per-ABI/package analysis should be added before dependency cleanup.
-
-## Compose compiler snapshot
-
-Reports were generated under `app/build/reports/compose-compiler/` using the diagnostic configuration added to [`app/build.gradle.kts`](../app/build.gradle.kts).
-
-| Metric | Value |
-|---|---:|
-| Generated composables | 2,273 |
-| Restartable composables | 2,240 |
-| Skippable composables | 1,243 |
-| Known unstable arguments | 1,060 |
-| Inferred unstable classes | 156 of 268 analyzed classes |
-| Memoized lambdas | 3,070 |
-| Compose feature flags | Strong skipping, intrinsic remember, optimized non-skipping groups, pausable composition |
-
-The generated composable count is larger than the source-level `@Composable` count because the compiler report also includes generated and nested composable groups/lambdas.
-
-## First findings
-
-### Shared media components
-
-- `ChannelCard` receives an unstable `Channel`.
-- `MovieCard` receives an unstable `Movie`.
-- `SeriesCard` receives an unstable `Series`.
-- `CategoryRow` receives an unstable `List<T>`.
-- `ChannelCard` also observes the shared 30-second progress ticker from inside each card.
-
-Initial implication: do not immediately annotate domain models as stable. Introduce feature-facing immutable UI models where the rendering contract is smaller and genuinely immutable. Revisit ticker ownership after runtime tracing.
-
-### Player
-
-- `PlayerScreen` itself is restartable/skippable, but its root ViewModel exposes a large state surface.
-- `PlayerControlsOverlay` receives unstable `Program` and `Channel` values in addition to realtime position/duration.
-- `PlayerRenderView` receives the player abstraction and should remain isolated from the rest of the composition.
-
-Initial implication: split player state by frequency and consumer before changing player module boundaries. Position, duration, and seek preview should remain in the transport subtree. Live channel/EPG state should remain in the live overlay subtree.
-
-### Provider setup
-
-- `ProviderSetupScreen` receives an unstable ViewModel and owns many local draft values.
-- `ProviderSetupCompletionLayer` receives unstable provider state, sets, and file values.
-
-Initial implication: extract provider-specific form models and move persistent draft ownership toward the ViewModel. Keep launcher callbacks at the route boundary.
-
-### Settings and navigation
-
-- `SettingsScreen` receives an unstable ViewModel and passes unstable settings state into dialog/overlay hosts.
-- `AppNavigation` receives an unstable `MainActivity` object containing repositories, managers, and mutable flows.
-
-Initial implication: root navigation should receive small platform capability interfaces and navigation state rather than a concrete activity. Settings should eventually depend on use-case/contracts instead of concrete DAOs and preference implementations.
-
-## Baseline limitations
-
-The following are still required before performance targets are finalized:
-
-- Repeated Gradle Profiler incremental scenarios.
-- Compose recomposition tracing for player, Live TV, EPG, dashboard, and settings.
-- Macrobenchmark or equivalent startup/interaction measurements.
-- At least one constrained TV device and one phone/tablet reference device.
-- Release/beta APK size and per-ABI analysis.
-- Long-duration Live TV validation before and after player-facing changes.
-
-## Immediate next implementation
-
-The first behavior-preserving UI change should be player source decomposition and state-boundary preparation:
-
-1. **Complete for this slice:** split the controls overlay host out of `PlayerScreen.kt`.
-2. **Next:** split lifecycle, input, transport-host, and modal-host code from `PlayerScreen.kt`.
-3. Keep the same ViewModel, routes, callbacks, and playback policies.
-4. Run the existing player tests and full Live TV validation.
-5. Compare the Compose compiler report and runtime trace with this baseline.
-
-The next candidate is provider setup decomposition, followed by feature-module extraction once the internal boundaries are stable.
diff --git a/docs/COMPOSE_REDUCTION_PHASE2_5A_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE2_5A_REPORT.md
deleted file mode 100644
index ba648a04..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE2_5A_REPORT.md
+++ /dev/null
@@ -1,141 +0,0 @@
-# Compose Reduction Phase 2.5A Report
-
-Date: 2026-08-23
-
-## Scope
-
-Phase 2.5A hardens the live-category flow path exposed by `ChannelRepository`.
-It is an independent data-layer optimization. Movie, series, VOD, and
-`CategoryRepository` caching are not included, and no baseline-profile work is
-included.
-
-## Implementation
-
-- Added the singleton `ChannelCategoryFlowCache` at
- `data/src/main/java/com/streamvault/data/repository/ChannelCategoryFlowCache.kt`.
-- The cache stores one `shareIn(replay = 1)` flow per provider ID in a
- `ConcurrentHashMap`, uses the application repository scope, applies
- `distinctUntilChanged()`, and observes `ProviderDao.getAll()` to evict deleted
- providers.
-- The shared flow uses a 30-second `WhileSubscribed` stop timeout. That keeps
- the Room observation alive across a short screen/task re-entry gap while
- provider deletion still cancels the per-provider scope immediately.
-- `ChannelRepositoryImpl` now keeps the original Room-backed category
- construction in `buildCategoriesFlow(providerId)`. `getCategories()` uses the
- shared replaying flow, while `getCategoriesSnapshot()` builds a fresh flow and
- calls `first()` without consulting the cache.
-- `MultiViewViewModel` now uses `getCategoriesSnapshot()` for its command-style
- picker load, removing the stale one-shot read from the replayable UI API.
-- Hardened the existing Live TV/category macrobenchmark setup so it uses the
- seeded debug fixture, follows the TV focus path to enter Live TV, and asserts
- that the seeded `All Channels` category is present before measuring
- navigation. Re-entry returns to Home through the app's existing `ACTION_VIEW`
- landing intent so the route transition does not depend on retained TV focus.
-- Added a separate `liveTvCategoryReentryCategoryBuild` journey. It leaves and
- re-enters Live TV inside the measured block and reports both the
- `StreamVault.CategoryFlow.Build` trace-section count and the new
- `StreamVault.CategoryFlow.UpstreamStart` and
- `StreamVault.CategoryFlow.UpstreamStop` counts alongside frame timing. The
- lifecycle hooks are emitted at the shared-flow upstream boundary, so they can
- distinguish a restart from a retained subscription. These traces are still
- not a substitute for a direct Room query-count measurement.
-
-## Test coverage
-
-Focused tests cover:
-
-- same-provider flow reuse;
-- independent flows for different providers;
-- replay to a second subscriber;
-- active Room/upstream refreshes;
-- duplicate suppression through `distinctUntilChanged()`;
-- provider deletion eviction and recreation without the old replay;
-- fresh snapshot reads after the replayed result has become stale;
-- category visibility/count changes after parental-control preference changes;
-- quick subscriber re-entry inside the stop timeout, proving that it does not
- start a second Room observation;
-- existing grouping, count, parental-filtering, and decorative-row behavior.
-
-Validation completed:
-
-```text
-:data:testDebugUnitTest
-:app:testDebugUnitTest
-:app:compileDebugKotlin
-:app:assembleDebug
-:benchmark:compileBenchmarkKotlin
-:benchmark:connectedBenchmarkBenchmarkAndroidTest
-```
-
-All completed successfully on 2026-08-22.
-
-The repository-wide `gradlew test` task was also run. It stopped on the
-pre-existing `domain:test` failure
-`AppHomeDashboardShelfTest.defaultOrder matches the existing home layout`
-(`AppHomeDashboardShelfTest.kt:10`); this test and its shelf implementation
-are outside the Phase 2.5A changes. The Phase 2.5A-targeted module checks above
-remain green.
-
-## Runtime benchmark status
-
-ADB was validated through `E:\androidSdk\platform-tools\adb.exe` on
-`Television_1080p(AVD) - 16`. The installed debug APK before the change was
-treated as the pre-2.5A baseline (`com.streamvault.app.debug`, versionCode 18,
-SHA-256 `64B39335C0FD3810CD71AFB2463915C0067596120F5535EBAFF02F71EDDA11A9`).
-The current debug APK was then installed and verified by device hash
-(`C1119B6B2857FC0301EE208271C4BA2EED85ADE2DD831E6C5E7E6487176EAEB1`).
-
-Both runs used five iterations of
-`liveTvCategoryAndChannelNavigation`, `FrameTimingMetric`, and an explicit
-`All Channels` readiness assertion:
-
-| Metric | Pre-2.5A | Current 2.5A |
-| --- | ---: | ---: |
-| Frame counts | 38, 44, 48, 46, 46 | 48, 49, 42, 44, 46 |
-| Frame-count median | 46 | 46 |
-| CPU frame P50 | 27.553 ms | 31.120 ms |
-| CPU frame P90 | 37.713 ms | 41.799 ms |
-| Frame overrun P50 | 19.703 ms | 24.895 ms |
-| Frame overrun P90 | 36.915 ms | 38.832 ms |
-
-The journey passed before and after and confirmed category availability, but
-this metric measures rendering/input navigation, not Room query count. It does
-not demonstrate a cache win; the current run was noisier/slower on this
-emulator/debug target. The benchmark also carries the standard warnings that
-an emulator and debuggable target are diagnostic only, not representative
-release-device performance.
-
-The strengthened re-entry benchmark was rerun on
-`Television_1080p(AVD) - 16` with five iterations using
-`:benchmark:connectedBenchmarkBenchmarkAndroidTest` and the seeded debug
-fixture. The harness starts from Home without killing the debug process,
-performs the first Live TV entry inside the measured block, and leaves
-`targetPackageOnly` disabled because the interaction fixture is the separate
-`.debug` package. The latest run used the 30-second stop timeout and captured:
-
-| Metric | Min | Median | Max |
-| --- | ---: | ---: | ---: |
-| `categoryBuildCountCount` | 2 | 2 | 2 |
-| `categoryUpstreamStartCountCount` | 2 | 2 | 2 |
-| `categoryUpstreamStopCountCount` | 1 | 1 | 1 |
-| `frameCount` | 83 | 87 | 90 |
-
-For comparison, the same task-reset journey with the stop timeout temporarily
-set to zero also captured `2/2/1` for build/start/stop in every iteration. The
-30-second setting therefore did not produce a measurable reduction in this
-specific Activity-task reset; the boundary is longer or otherwise different
-from the short subscriber gap covered by the deterministic unit test. This is
-useful lifecycle evidence, but it is not a quantified cache win. A direct Room
-query-count or category-ready-latency benchmark remains open before claiming a
-runtime reduction. Both runs are diagnostic only because they use an emulator
-and a debuggable seeded target.
-
-The focused unit test is the stronger proof for the cache contract: a second
-subscriber arriving before the timeout observes the replay without a second
-upstream start or stop. Provider deletion and preference changes remain covered
-by the same test suite.
-
-The benchmark harness now explicitly launches `com.streamvault.app.debug` for
-interaction journeys while leaving release startup/profile checks on the
-release target. This fixes the previous target-package mismatch that could
-silently measure an unseeded release app.
diff --git a/docs/COMPOSE_REDUCTION_PHASE2_5B_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE2_5B_REPORT.md
deleted file mode 100644
index 1290091d..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE2_5B_REPORT.md
+++ /dev/null
@@ -1,232 +0,0 @@
-# Compose Reduction Phase 2.5B Report
-
-Date: 2026-08-22
-
-## Status
-
-The startup coordinator, route-guard, async-tracing, and redesigned profile
-generator are present. A clean connected-device collection now passes both the
-startup and critical-journey tests on the API 36 TV emulator. Phase 2.5B is
-**not exit-ready**: the release APK is 1.44% larger than the pre-2.5B APK and
-the required two ten-iteration comparisons on constrained physical TV hardware
-were not available. Shipping packaging is now validated for beta, release APK,
-and release AAB. Generated profile sources remain local and uncommitted pending
-the remaining size and physical-device gates.
-
-## Scope and implementation
-
-- Added `AppStartupCoordinator` as an internal singleton. Process maintenance
- starts once from `StreamVaultApp`, independently of an Activity. Expensive
- task graphs are injected through `Provider<T>` and each task runs under
- supervisor isolation with named Android trace sections.
-- `MainActivity` uses a one-shot `doOnPreDraw` plus posted callback to release
- television-only Watch Next, launcher recommendation, and TV-input refresh
- work after the first UI frame. Non-TV and service-only launches do not run
- those tasks.
-- Applied AndroidX Baseline Profile plugin `1.4.1` to `:app` and `:benchmark`.
- The app uses `mergeIntoMain = true`, `saveInSrc = true`, and disables
- automatic generation during ordinary builds. The existing `:benchmark`
- module is the only profile producer.
-- Added guarded `startup` and `criticalJourneys` collections. Startup contains
- only cold launcher-to-ready-Home. Critical journeys cover Home, Live TV,
- EPG, and live-player controls, with destination assertions before each
- continuation.
-- Added source verification and controlled startup benchmarks. The treatment
- requires `BaselineProfileMode.Require`; a missing ProfileInstaller or
- packaged profile therefore fails instead of silently measuring a control.
- Source verification is a typed, read-only Gradle task and remains compatible
- with the repository's configuration-cache build path. A separate typed
- merge/install task unions startup rules into `baseline-prof.txt` after the
- profile plugin copies fresh captures, while preserving `startup-prof.txt` as
- the startup-only subset.
-- Stabilized Kotlin's module name across release-like variants so internal JVM
- method names in a profile collected from `nonMinifiedRelease` also match beta
- and release. The merge task normalizes legacy variant-suffixed rules,
- canonicalizes flags, and deduplicates both maintained sources. Its explicit
- task ordering prevents Gradle from consuming copied profile outputs before
- they have been merged.
-
-## Pre-2.5B baseline
-
-Captured before the startup coordinator changes on the API 36 TV emulator:
-
-| Field | Value |
-| --- | --- |
-| Device | `Television_1080p(AVD) - 16`, AOSP TV on x86 |
-| Android | API 36, fingerprint `google/sdk_google_atv_x86/emulator_x86_arm:16/BT2A.251018.001.A1/14340881` |
-| Hardware model | 4 cores, max 2 GHz, approximately 2 GiB RAM |
-| Compilation | `CompilationMode.None()` |
-| Iterations | 10 cold starts |
-| TTID median | 2,077.1 ms |
-| TTID min/max | 1,958.2 / 2,490.4 ms |
-| TTID P90 | Not emitted by the original result; min/max were retained |
-
-The original ten Perfetto trace files were captured, then removed by a later
-workspace `clean` to recover disk space. Their names and measurements were
-recorded before cleanup.
-
-## Generated profile sources
-
-The redesigned generator completed a clean connected-device collection on the
-seeded API 36 TV emulator. `startup()` contains only cold launcher-to-ready-Home
-work; `criticalJourneys()` covers Home, Live TV, EPG, and a seeded HLS player.
-The profile plugin emitted the general and startup captures separately. The
-typed merge/install task then added the startup capture's rules to the
-maintained baseline source, which enforces the Android requirement that a
-startup profile is a subset of the baseline profile without reusing PR #162's
-generated files.
-
-| Source | Size | Rules |
-| --- | ---: | ---: |
-| `baseline-prof.txt` (after startup union) | 5,433,664 bytes | 47,782 |
-| `startup-prof.txt` | 3,156,029 bytes | 29,852 |
-
-The latest raw captures contained 48,006 general and 29,857 startup lines.
-Normalization canonicalizes Kotlin variant suffixes, profile flag strength, and
-duplicate keys, producing the maintained 47,782/29,852 rule sets. The general
-capture already covered every startup method/class; the merge adds only
-genuinely missing startup keys if a future collection exposes one. No
-beta/release generated-profile subdirectories were created.
-
-The collection command used the final generator class explicitly:
-
-```text
-gradlew.bat :app:generateBaselineProfile ^
- -Pandroid.testInstrumentationRunnerArguments.androidx.benchmark.enabledRules=BaselineProfile ^
- -Pandroid.testInstrumentationRunnerArguments.class=com.streamvault.benchmark.BaselineProfileGenerator
-```
-
-## Artifact validation
-
-The final shipping-variant packaging run completed successfully:
-
-```text
-gradlew.bat :app:assembleBeta :app:assembleRelease :app:bundleRelease --max-workers=1
-```
-
-Pre-2.5B artifact hashes retained from the capture step were:
-
-```text
-benchmark APK 37,454,211 bytes C3112180B567AC232CAFA909BC067B668D8D173928029490E9CD783ED500E8BF
-beta APK 37,454,231 bytes 1CD0D0597A727B0EDAE88319DCAC9570111336E556B886F660CE103837F38328
-release APK 18,270,972 bytes 8B311F02AC31043C72F14D09A771680B0413BF25F9A7C61D99F20C30CE4F8E86
-release AAB 24,695,046 bytes 6AB694F96FEDF5C1CAC6B0CCBC04FAEC1CF5A6A1FAE86980D3F98F1BA8488D31
-```
-
-| Artifact | Pre-2.5B bytes | Post-2.5B bytes | Change | Post SHA-256 |
-| --- | ---: | ---: | ---: | --- |
-| Release APK | 18,270,972 | 18,533,311 | +1.436% | `117AD397DA585A1E03585EBF51177B0302218C1F6905A6F4B0F59CD1B2FE2FCF` |
-| Release AAB | 24,695,046 | 24,880,653 | +0.752% | `C459414A8F36C9B1A21DE8DD9C73ADE88D30EFB561CD27D88D3BDB68A63202DD` |
-| Beta APK | 37,454,231 | 38,486,423 | +2.756% | `E66DD73D03262588DD7866075D4F9F0C7E5AA060C12B66EF50A4515EE229E066` |
-
-Both beta and release APKs contained:
-
-```text
-assets/dexopt/baseline.prof
-assets/dexopt/baseline.profm
-```
-
-The release AAB contained the corresponding
-`BUNDLE-METADATA/com.android.tools.build.profiles/baseline.prof` and
-`baseline.profm` entries. Release R8 metadata recorded two DEX files, with the
-startup-optimized entry set to `"startup":true`. Release had three DEX files.
-
-The AAB is inside the 1% size budget. The APK is not, so the artifact-size
-acceptance gate remains open. Beta/release build constants were checked to have
-empty dev-seed fields; the non-shipping `nonMinifiedRelease` target alone reads
-local.properties seed fields. No dev-server URL, username, password, or M3U
-seed value was found in the beta/release shipping artifacts.
-
-The D8 profile merge still emits non-fatal startup-class placement diagnostics
-for some external dependency rules. The previous app-variant suffix misses are
-gone after the stable module-name fix. Packaging and `Require` both succeeded;
-the remaining dependency diagnostics should be reviewed before committing the
-generated output.
-
-## Controlled startup measurements
-
-The valid post-refactor runs used the non-shipping `nonMinifiedRelease` target,
-ten cold iterations, `StartupTimingMetric`, and the API 36 TV emulator. The
-Macrobenchmark runner needed the explicit `enabledRules=Macrobenchmark`
-override because the baseline-profile plugin otherwise selects its
-`BaselineProfile` rule for this variant.
-
-| Run | Compilation | Median TTID | P90 TTID* | Min | Max |
-| --- | --- | ---: | ---: | ---: | ---: |
-| Post-refactor control | `None()` | 1,808.0 ms | 1,955.8 ms | 1,758.2 ms | 2,216.6 ms |
-| Post-refactor treatment | `Partial(Require)` | 1,700.9 ms | 1,809.6 ms | 1,639.6 ms | 1,818.4 ms |
-
-The required-profile treatment was successful. On this emulator it was 5.9%
-faster by median and 7.5% faster by P90 than the post-refactor control. P90 is
-the linear percentile calculated from the ten raw run values; the benchmark
-message itself reports median/min/max.
-
-The controlled commands were:
-
-```text
-gradlew.bat :benchmark:connectedNonMinifiedReleaseAndroidTest ^
- -Pandroid.testInstrumentationRunnerArguments.androidx.benchmark.enabledRules=Macrobenchmark ^
- -Pandroid.testInstrumentationRunnerArguments.class=com.streamvault.benchmark.StreamVaultMacrobenchmark#coldStartupNoCompilation
-
-gradlew.bat :benchmark:connectedNonMinifiedReleaseAndroidTest ^
- -Pandroid.testInstrumentationRunnerArguments.androidx.benchmark.enabledRules=Macrobenchmark ^
- -Pandroid.testInstrumentationRunnerArguments.class=com.streamvault.benchmark.StreamVaultMacrobenchmark#coldStartupWithBaselineProfile
-```
-
-Both prior runs generated ten Perfetto traces. The treatment trace contained
-the named coordinator sections in order, including
-`AppStartupCoordinator:process-maintenance` and the later
-`AppStartupCoordinator:tv-integrations-after-first-frame`, with task sections
-for process work and TV integrations. Those traces were collected before the
-async trace-section fix, so the ordering must be re-captured before using them
-as final evidence; physical-TV confirmation is still required.
-
-## Validation record
-
-Passed in the final verification pass:
-
-```text
-:app:testDebugUnitTest
-:benchmark:compileBenchmarkKotlin
-:app:compileDebugKotlin
-:app:assembleBeta
-:app:assembleRelease
-:app:bundleRelease
-:app:generateBaselineProfile (guarded generator; both tests passed)
-:app:installMergedBaselineProfile
-:verifyBaselineProfileSources
-:data:testDebugUnitTest (focused cache/repository tests)
-:app:testDebugUnitTest (focused startup coordinator test)
-:benchmark:compileBenchmarkKotlin
-:benchmark:connectedNonMinifiedReleaseAndroidTest (control)
-:benchmark:connectedNonMinifiedReleaseAndroidTest (Require treatment)
-```
-
-The earlier combined command reached `:app:testDebugUnitTest` successfully and
-then failed during `:app:assembleDebug` because the C: volume exhausted space
-while dexing Compose dependencies. That failure is unrelated to the focused
-tests or shipping packaging; a full debug assemble remains a separate local
-workspace check.
-
-The repository-wide `gradlew.bat test --max-workers=1` was also rerun. App and
-data test tasks passed, but the suite stopped at the existing domain failure:
-`AppHomeDashboardShelfTest.defaultOrder matches the existing home layout`
-(`AppHomeDashboardShelfTest.kt:10`), with 132 domain tests completed and one
-failure. This is outside the Phase 2.5B changes.
-
-## Limitations and exit decision
-
-- Only the API 36 TV emulator was available. The emulator warns that its
- performance is not representative of physical TV hardware.
-- The plan requires two constrained physical-TV ten-iteration comparisons,
- with a no-profile regression limit and a with-profile benefit threshold.
- Those comparisons were not run.
-- The release APK exceeds the 1% artifact-growth limit by approximately 0.436
- percentage points, although the AAB is within budget.
-- The pre-run P90 was not emitted by the original baseline result, so the
- three-way P90 comparison needs a fresh controlled baseline if the physical
- gate is pursued.
-- Generated profile sources are intentionally not committed yet.
-
-Decision: **code/tooling and shipping packaging complete; Phase 2.5B exit gate
-open** pending APK-size investigation and two physical-TV startup comparisons.
diff --git a/docs/COMPOSE_REDUCTION_PHASE2_LIST_AUDIT.md b/docs/COMPOSE_REDUCTION_PHASE2_LIST_AUDIT.md
deleted file mode 100644
index 3fe8654e..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE2_LIST_AUDIT.md
+++ /dev/null
@@ -1,37 +0,0 @@
-# Phase 2 lazy-list audit
-
-This audit covers the Phase 2 hot paths: Home and Dashboard shelves, Search
-results, the EPG channel rail, reusable media rows, and player/dialog lists.
-The rule is to keep a stable domain identifier as the key and add
-`contentType` when a list mixes item layouts or when a dynamic dialog list is
-otherwise difficult for Compose to classify.
-
-| Area / list | Key | `contentType` | Result |
-| --- | --- | --- | --- |
-| Dashboard sections | `AppHomeDashboardShelf.storageValue` | shelf storage value | Fixed; sections render materially different shelf layouts. |
-| Dashboard recent/favorite/media `CategoryRow` shelves | Existing item IDs | CategoryRow supports an optional selector; omitted for homogeneous item renderers | Reviewed; stable keys are supplied by each caller and no mixed item layouts are present inside a shelf. |
-| Dashboard shortcuts, stats, favorite-logo row | Existing shortcut/stat label or channel ID | Not required | Reviewed; each list is homogeneous and already keyed. |
-| Home visible categories | Category ID | `live_category` | Fixed. |
-| Home filtered channel list | Channel ID | `live_channel` | Fixed; this is also the owner of the shared channel clock. |
-| Search result rails | Item ID through `CategoryRow` | Not required | Reviewed; each rail is homogeneous and uses the existing ID selector. |
-| Search LIVE/MOVIES/SERIES result rows | Joined stable item IDs | `search_live_row`, `search_movie_row`, `search_series_row` | Fixed; the selected tab chooses a different row renderer. |
-| Search tabs and recent queries | Tab name / query text | Not required | Reviewed; both lists are homogeneous and keyed. |
-| EPG channel rows | `epgChannelKey(channel, index)` | `epg_channel` | Fixed; the index remains part of the existing collision-safe key policy. |
-| `CategoryRow` reusable media shelf | Caller key selector | Caller selector, when needed | Reviewed; the reusable primitive already forwards keys and content types; homogeneous shelves only need the stable caller key. |
-| `ContinueWatchingRow` | Playback-history ID | History content type | Already compliant. |
-| Player channel/category/EPG overlay lists | Existing player item-key helpers | Existing channel/category/program types | Already compliant from the overlay audit. |
-| Player track/variant/format/speed/timer/episode lists | Track/variant/format/speed/minutes/season/episode IDs | Homogeneous list types | Reviewed; stable keys are present and each list has one renderer. |
-| Program history dialog | Channel/start/end/ID composite | `program` | Fixed; prevents index-key fallback when EPG data refreshes. |
-| Add-to-group dialog | Category ID | `category` | Fixed. |
-| M3U series-assignment dialog | Channel ID | `series_assignment` | Fixed. |
-| Player quick-action row | Semantic `PlayerActionSpec.id` | `player_quick_action` | Fixed; labels can change for mute, casting, playback speed, aspect ratio, or timer state without changing item identity. |
-| Channel-info action tray | Existing `label:index` composite | `channel_info_action` | Reviewed; the private action type has no semantic ID, so the short homogeneous tray retains its collision-safe composite key. |
-| Settings navigation rail | Stable `SettingsNavEntry.id` | `settings_navigation_entry` | Fixed; the static rail now has explicit item identity and content type. |
-
-No display-text key was introduced where a stable domain ID was available. The
-private channel-info action type has no domain ID, so its short tray keeps the
-existing collision-safe `label:index` key.
-The channel progress clock is collected once by Dashboard, Home, and Search;
-`ChannelCard` and `LiveChannelRowCard` now receive the timestamp as a value and
-do not collect a flow from inside a lazy-list item. `CategoryRow` now requires a
-key selector, preventing future callers from silently opting into index keys.
diff --git a/docs/COMPOSE_REDUCTION_PHASE2_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE2_REPORT.md
deleted file mode 100644
index a16cb534..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE2_REPORT.md
+++ /dev/null
@@ -1,183 +0,0 @@
-# Compose Reduction Phase 2 Report
-
-Date: 2026-08-22
-Status: Implementation slices complete; live-TV validation passed; performance comparison pending
-
-This report records the Phase 2 work against the goals in
-[`COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md`](COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md).
-The purpose of Phase 2 is to reduce unnecessary invalidation in confirmed
-Compose hotspots by making state ownership follow consumer and update
-frequency, while preserving playback and navigation behavior. It is not a
-request to annotate unstable domain models or to rewrite the UI in Views.
-
-## Implementation completed
-
-### Player state and modal ownership
-
-- Replaced the broad mutable modal-state shape with one nullable
- `PlayerModalState.active` value and typed `PlayerModal` variants.
-- Migrated the existing PlayerScreen modal transitions to the typed open and
- dismiss API without changing modal placement or callback behavior.
-- Kept position and duration in `PlayerControlsOverlayHost` and moved other
- player state toward the host that consumes it:
- - controls and transport state in `PlayerControlsOverlayHost`;
- - diagnostics and player statistics in `PlayerDiagnosticsHost`;
- - live channel, category, EPG, numeric-input, and channel-info state in
- `PlayerLiveOverlayHost`;
- - track, speed, timer, offset, and related dialog state in
- `PlayerModalHosts`;
- - program-history state is collected only while the top-level history modal
- is visible;
- - current and next program state is collected by the controls and live
- overlay hosts that render it.
-- Preserved the two modal-host placement boundaries: top-level modal content
- remains before player preparation, and controls modal content remains inside
- the player surface layer after the resume prompt.
-- Consolidated the audio/video sync flow collection so it is collected once in
- the modal branch that needs it.
-
-### Shared channel clock
-
-- Added one shared 30-second channel-progress ticker per screen owner.
-- Dashboard, Home, and Search collect the timestamp once and pass it to their
- channel cards.
-- `ChannelCard` and `LiveChannelRowCard` no longer collect a flow from inside
- lazy-list items. Progress calculation is now a pure clamped function.
-
-### Lazy-list audit
-
-The hot-list review is recorded in
-[`COMPOSE_REDUCTION_PHASE2_LIST_AUDIT.md`](COMPOSE_REDUCTION_PHASE2_LIST_AUDIT.md).
-It covers Dashboard, Home, Search, EPG, media shelves, player overlays, and
-dialogs. The changes add stable domain/composite keys and `contentType` where
-lists mix layouts or refresh independently. Player quick actions now use
-semantic IDs instead of display labels, so changing mute, cast, speed, aspect,
-or timer labels does not change item identity.
-
-## Compose compiler audit
-
-The report was regenerated with:
-
-```text
-.\gradlew.bat :app:compileDebugKotlin --rerun-tasks --no-daemon --console=plain
-```
-
-The current aggregate is:
-
-| Metric | Phase 0 | Fresh Phase 2 report |
-| --- | ---: | ---: |
-| Generated composables | 2,273 | 2,311 |
-| Restartable composables | 2,240 | 2,278 |
-| Skippable composables | 1,243 | 1,266 |
-| Known unstable arguments | 1,060 | 1,122 |
-| Inferred unstable classes | 156 of 268 | 160 of 331 |
-| Memoized lambdas | 3,070 | 3,115 |
-
-The aggregate compiler counts are not a Phase 2 performance win. The increase
-is consistent with the added host boundaries and the existing unstable domain
-arguments; it is not evidence that the refactor regressed runtime behavior,
-but runtime traces are required to establish that. No blanket `@Stable` or
-`@Immutable` annotations were added.
-
-The highest-value remaining stability hotspots are the domain-shaped
-`Channel`, `Movie`, and `Series` arguments in shared cards, collection-shaped
-media shelf arguments, and large player-facing arguments such as `PlayerViewModel`,
-`PlayerEngine`, `Program`, `Channel`, and `PlayerTimeshiftUiState`. These should
-be addressed with smaller immutable UI models only when their contracts can be
-proved, rather than by suppressing the diagnostic.
-
-## Build and test evidence
-
-The following checks passed after the Phase 2 changes:
-
-- `:app:testDebugUnitTest` — passed; 69 actionable tasks, 8 executed and 61
- up-to-date.
-- `:app:assembleDebug` — passed; 89 actionable tasks, 5 executed and 84
- up-to-date.
-- `:app:compileDebugKotlin --rerun-tasks` — passed; all 49 tasks executed and
- the fresh Compose compiler report was emitted.
-- Focused modal, input, and Back-policy tests — passed.
-- Focused channel-progress ticker tests — passed.
-
-The build output contains existing deprecation, opt-in, and JVM-sharing
-warnings; no new compilation or unit-test failure was observed.
-
-## Runtime and benchmark evidence
-
-ADB is available at `E:\androidSdk`. Runtime validation used the connected
-`Television_1080p` emulator (`emulator-5554`, API 36) with the debug build and
-the documented public US M3U seed configured through `local.properties`.
-
-The Phase 0 baseline contains emulator macrobenchmark and long-run playback
-measurements. No comparable Phase 2 frame-cost or recomposition trace has been
-captured yet.
-
-The task `:benchmark:connectedBenchmarkAndroidTest` was started. It built the
-benchmark and release-like target artifacts. The first run was rejected by the
-benchmark library because the target was an emulator and the seeded journeys
-target the debuggable debug APK. The benchmark configuration now suppresses
-only those known `EMULATOR` and `DEBUGGABLE` policy errors for diagnostic local
-runs. A rerun entered instrumentation but hung before completing a test or
-writing metrics and was terminated. It is therefore an incomplete runtime
-attempt, not a benchmark result.
-
-A manual release-target `am start -W` launch completed successfully with a
-single `TotalTime` of approximately 1.1 seconds. This is a smoke check only,
-not a comparable replacement for the Phase 0 macrobenchmark median.
-
-Compiler diagnostics are useful for identifying ownership and stability
-hotspots, but they do not replace recomposition tracing, frame-cost
-measurement, or a before/after macrobenchmark.
-
-## Live playback gate
-
-The required Phase 2 live-TV validation passed on the connected emulator. Each
-channel was opened in the full player and captured at a two-second cadence for
-61 screenshots (roughly two minutes):
-
-The same protocol was rerun after the follow-up player state-isolation slice;
-the final exact-build captures are prefixed `final_` in
-`validation/phase2_live/` and produced the same passing results below. The
-fresh preflight log contains HLS prepare and first-frame success for `00s Replay`;
-both final capture logs contain zero matches for the prohibited error,
-stuck-player, and fallback patterns.
-
-| Channel | Screenshots | Unique hashes | Final media session | Healthy log evidence | Error/fallback evidence |
-| --- | ---: | ---: | --- | --- | --- |
-| 00s Replay | 61 | 61 | `PLAYING`, `error=null` | HLS prepare and first-frame success | No fatal error, stuck timeout, or MPEG-TS fallback |
-| 3ABN Dare To Dream Network | 61 | 61 | `PLAYING`, `error=null` | HLS prepare and first-frame success | No fatal error, stuck timeout, or MPEG-TS fallback |
-
-The raw screenshots, hashes, media-session dumps, and sanitized log captures
-are in the local `validation/phase2_live/` directory. The public playlist and
-provider streams can change independently of the application; these results
-are evidence for this build and validation window.
-
-The validation protocol covered:
-
-- two live channels;
-- 61 screenshots at a two-second cadence per channel;
-- changing-frame hash counts across the capture window;
-- media session still `PLAYING` with `error=null`;
-- sanitized logs showing healthy HLS prepare/read/first-frame or recovery;
-- no fatal error, stuck-player timeout, or unintended MPEG-TS fallback.
-
-The Phase 0 baseline remains the reference: CBSN passed that protocol, while
-the F1 channel failed because the provider returned HTTP 403. Those baseline
-results must not be reused as Phase 2 evidence.
-
-## Exit assessment
-
-Phase 2 is implementation-complete for the state-isolation, shared-clock, and
-lazy-list slices, and the repository builds and tests successfully. The live-TV
-runtime gate passed. It is not fully exit-ready yet because the performance
-comparison remains unmeasured. Completion still requires:
-
-1. a completing Phase 2 macrobenchmark or equivalent controlled frame-cost
- measurement;
-2. a Phase 2 recomposition trace against the Phase 0 baseline; and
-3. before/after runtime evidence showing whether the isolated consumers reduce
- invalidated scopes or frame cost.
-
-Until those measurements are available, the correct status is **implementation
-complete, live validation passed, performance comparison pending**, not a
-claimed performance improvement.
diff --git a/docs/COMPOSE_REDUCTION_PHASE3_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE3_REPORT.md
deleted file mode 100644
index db6d48d6..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE3_REPORT.md
+++ /dev/null
@@ -1,122 +0,0 @@
-# Compose Reduction Phase 3 Report
-
-Date: 2026-08-23
-
-## Status
-
-Phase 3 is complete. The reusable presentation foundation now lives in
-`:core:ui`; app navigation, resources, repositories, and Activity behavior
-remain in `:app` behind an adapter. The dependency guard, JVM tests, connected
-TV instrumentation, shell goldens, premium-route goldens, and debug APK build
-all pass.
-
-## New boundary
-
-- Added `:core:ui` and included it from the root Gradle settings.
-- `:app` depends on `:core:ui`; `:core:ui` has no project dependencies.
-- Added `verifyCoreUiBoundary` at the module and root levels. It rejects
- project dependencies and references to app/data/domain/player/navigation/Hilt
- APIs from core source.
-- Core source contains no app resources, navigation controllers, domain models,
- repositories, Activity references, or player references.
-
-The guard command and result were:
-
-```text
-.\gradlew.bat verifyCoreUiBoundary
-Verified :core:ui boundary: no project dependencies and no banned source references.
-```
-
-## Moved or extracted source
-
-- Design tokens, typography, focus helpers, theme, and Inter font resources move
- to `com.streamvault.core.ui.design` and `com.streamvault.core.ui.theme`.
-- TV interaction primitives and mouse/remote activation handling move to
- `com.streamvault.core.ui.interaction`.
-- The television-device predicate moves to
- `com.streamvault.core.ui.device`, with the original feature/UI-mode
- short-circuit behavior preserved.
-- Generic shell visuals move to
- `com.streamvault.core.ui.components.shell`, including the core scaffold,
- navigation chrome, headers, status/message states, load-more card, metadata
- strip, and close action.
-- App-owned media cards and browse-specific panels remain in `:app`.
-- `AppShellNavigation` remains in `:app` and maps app routes, localized labels,
- resources, repository-backed configuration, and Activity callbacks into the
- core shell API. The existing VOD catalog merge rule is covered by unit tests.
-
-## Verification
-
-| Check | Result |
-| --- | --- |
-| `verifyCoreUiBoundary` | Pass |
-| `:core:ui:compileDebugKotlin` | Pass |
-| `:core:ui:testDebugUnitTest` | Pass |
-| `:app:compileDebugKotlin` | Pass |
-| `:app:testDebugUnitTest` | Pass; 363 tests |
-| `:app:assembleDebug` | Pass |
-| `:core:ui:compileDebugAndroidTestKotlin` | Pass |
-| `:core:ui:connectedDebugAndroidTest` | Pass; 1/1 on `Television_1080p(AVD) - 16` |
-| `ShellGoldenTest` | Pass; 3/3 on `Television_1080p(AVD) - 16` |
-| `PremiumRouteGoldenTest` | Pass; 6/6 on `Television_1080p(AVD) - 16` |
-| `git diff --check` | Pass |
-
-The app connected tests installed the current checkout's debug APK for
-`com.streamvault.app` over the same-package app instance on the emulator. No
-release or differently sourced APK was used for these results.
-
-## Task impact and graph refresh
-
-The build graph now makes the intended boundary explicit: changes in core UI
-compile `:core:ui` and its app consumer, while app-only UI remains downstream
-of core. The verification run completed `:core:ui` compilation/tests, app
-compilation, and `assembleDebug`. No before/after build-time benchmark was
-claimed.
-
-`graphify update .` completed successfully:
-
-```text
-Rebuilt: 13613 nodes, 26500 edges, 362 communities
-graph.json and GRAPH_REPORT.md updated in graphify-out
-```
-
-`graph.html` was skipped by graphify because the graph exceeds its 5,000-node
-visualization limit.
-
-## Warnings and resolved issue
-
-The extracted device predicate initially read `resources.configuration`
-eagerly, which broke two mocked `WatchNextManagerTest` cases. The original app
-implementation returned early for known TV feature flags; that short-circuit
-was restored in core. The complete app unit suite then passed.
-
-Gradle reports configuration-cache warnings for the custom boundary task
-because it intentionally inspects `project` during task execution. The task
-still passes and is marked incompatible with configuration cache by design.
-
-## Scope confirmation
-
-This phase changed UI module ownership, package imports, and the app-to-core
-shell contract. It did not change player composition, playback lifecycle,
-stream recovery, or Live TV playback behavior; the long-duration Live TV
-validation protocol was therefore not newly required for this structural UI
-move.
-
-## Review hardening follow-up (2026-08-25)
-
-The original golden result above was produced by a helper that silently recorded
-the first screenshot when no baseline existed. The helper is now strict and
-loads committed assets from `app/src/androidTest/assets/ui-goldens/`. The
-app shell and premium-route PNG baselines are now checked in and were
-revalidated with recording disabled on the API 36 `Television_1080p` emulator.
-This prevents a clean emulator from accepting a visual regression as its own
-baseline.
-
-Implementation commits:
-
-- `ef421345` — add core UI module boundary
-- `9ce82152` — extract television device predicate
-- `66dfac60` — move design system into core UI
-- `c6853787` — move TV interaction primitives into core UI
-- `517b272f` — extract core shell visuals and app navigation adapter
-- `d2a9a69b` — preserve shell focus stacking
diff --git a/docs/COMPOSE_REDUCTION_PHASE4_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE4_REPORT.md
deleted file mode 100644
index 651655f4..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE4_REPORT.md
+++ /dev/null
@@ -1,121 +0,0 @@
-# Compose Reduction Phase 4 Report
-
-Date: 2026-08-24
-Phase: 4 — Navigation contracts
-Status: Architecture exit criteria met; one unrelated existing player smoke fixture remains red.
-
-## Scope and outcome
-
-Phase 4 established the navigation boundary needed for Phase 5 feature extraction without changing feature module ownership or player runtime behavior. The root navigation composable now wires the host and lifecycle effects, while typed contracts, startup/external coordination, route compatibility, and graph registration are separated behind explicit interfaces.
-
-The implementation does not change player composition, playback lifecycle, stream recovery, or stream policy. It changes how destinations and navigation requests reach the existing screens.
-
-## Module and package structure
-
-- `:core:navigation` is a pure Kotlin module with no project dependencies. It owns `AppDestination`, player navigation requests, navigation actions, and navigation request contracts.
-- `:core:ui` remains independent of `:app`, `:data`, and feature implementations. Its existing boundary and golden checks remain active.
-- `app/navigation` owns Android intent parsing, route strings, AndroidX Navigation adapters, domain presentation-hint transport, and the application coordinator.
-- `app/navigation/graph` contains controller-free `NavGraphBuilder` registration extensions for Welcome, Provider, Home, Live, Catalog, Player, and System areas.
-- Build-time guards verify the core and feature navigation boundaries. `FeatureGraphBoundaryTest` covers the source shape in the app test suite.
-
-## Contracts and compatibility
-
-The typed contracts cover:
-
-- top-level and detail destinations through `AppDestination`;
-- player launches, payload identity, and typed return destinations through `PlayerNavigationRequest`;
-- internal navigation actions and queued navigation commands;
-- external URI, legacy route, and Android intent requests;
-- navigation options such as single-top and pop-up behavior.
-
-`AppRouteCodec` and the compatibility `Routes` facade preserve the existing route strings, argument names, URI validation, legacy route parsing, typed extras, playlist/backup/search/action-view handling, and presentation-hint transport. Existing app-owned domain adapters remain in `app/navigation` until feature extraction.
-
-## Coordinator and graph responsibilities
-
-`AppNavigationCoordinator` owns startup resolution, external request submission, command queue/state, acknowledgement, top-level navigation requests, and resumed-destination handoff. `MainActivity` parses platform intents and submits typed requests. `AppNavigation.kt` is now a narrow host/wiring boundary with lifecycle effects that execute and acknowledge commands exactly once.
-
-`AppNavHost` supplies the `NavHost` and delegates destination registration to the seven graph extensions. Those extensions receive `NavigationActions` and feature-specific ports; they do not receive or reference the root `NavController`.
-
-The Phase 5 move units are therefore clear: playback/player graph and screen implementation first, followed by provider, settings, live/EPG, and catalog. Each future feature can retain its registration extension while consuming only typed navigation contracts and an injected action port.
-
-## Verification evidence
-
-### Static, JVM, and build checks
-
-The exact Phase 4 gate passed:
-
-```text
-.\gradlew.bat verifyCoreNavigationBoundary verifyFeatureNavigationBoundary verifyCoreUiBoundary :core:navigation:test :core:ui:testDebugUnitTest :app:testDebugUnitTest :app:compileDebugKotlin :app:assembleDebug --console=plain
-```
-
-Result: `BUILD SUCCESSFUL`.
-
-- `:core:navigation:test`: 6 tests, 0 failures.
-- `:core:ui:testDebugUnitTest`: 6 tests, 0 failures.
-- `:app:testDebugUnitTest`: 391 tests, 0 failures.
-- `:app:compileDebugKotlin` and `:app:assembleDebug`: passed.
-- All three boundary verification tasks: passed.
-
-The build reported 9 configuration-cache problems representing 4 unique warnings. They are from the existing `:core:navigation:verifyCoreNavigationBoundary` and `:core:ui:verifyCoreUiBoundary` tasks accessing `Task.project` during execution; the new `verifyFeatureNavigationBoundary` task was not implicated.
-
-### Connected TV coverage
-
-The suites ran on `Television_1080p(AVD) - 16` (API 36) using the current checkout's debug APK over the same application package:
-
-- `:core:ui:connectedDebugAndroidTest`: 1 passed.
-- `AppNavigationContractTest`: 2 passed.
-- `ShellGoldenTest`: 3 passed, 0 skipped.
-- `PremiumRouteGoldenTest`: 6 passed.
-- `PlayerSmokeTest`: 1 passed, 3 failed.
-
-The three player smoke failures are fixture/UI-test assertions outside the Phase 4 navigation contract. Phase 4 did not modify `PlayerSmokeTest` or the directly composed rail, controls-overlay, and track-dialog sources:
-
-- `categoryRailPanel_searchField_acceptsInitialFocusAndInput`: no text-action node was focused;
-- `playerControlsOverlay_playButton_canReceiveFocus`: the `>` node was not focused;
-- `playerTrackSelectionDialog_selectsAudioTrack`: the click did not update the test callback's selected track id.
-
-The passing player smoke case was `playerControlsOverlay_showsMuteActionWhenMuted`. The exact failure details are in the generated connected-test report under `app/build/reports/androidTests/connected/debug`.
-
-### Source-shape and graph checks
-
-- No `NavHostController` or `NavController` references were found under `app/src/main/java/com/streamvault/app/ui` or `app/src/main/java/com/streamvault/app/navigation/graph`.
-- No app, data, domain, player, Compose, Hilt, AndroidX Navigation, `NavController`, or `NavHost` references were found in `core/navigation/src/main`.
-- `AppNavHost.kt` contains the host plus graph registration calls and no feature screen implementations.
-- `graphify update .` completed successfully: 13,854 nodes and 26,866 edges. HTML visualization was skipped because the graph exceeds its 5,000-node limit.
-
-## Gates not exercised
-
-The long-duration live-TV playback protocol and manual player behavior matrix were not rerun for this phase because Phase 4 did not move or alter player composition, playback lifecycle, stream recovery, or stream policy. The player smoke suite was run and its known fixture failures are recorded above rather than counted as passes.
-
-## Phase 5 handoff
-
-Phase 5 can extract feature modules behind the established typed contracts. The first extraction should preserve `registerPlayerGraph` and the player request/return contracts as the seam, then move the provider, settings, live/EPG, and catalog registration units without reintroducing root-controller access.
-
-## Review hardening follow-up (2026-08-25)
-
-The Phase 4 review found two navigation regressions that the original report did
-not cover:
-
-- startup landing routes were held behind the optional favorite/history player
- lookup; landing navigation is now published immediately and the player
- request is completed asynchronously;
-- typed detail returns used a `popUpTo(Player)` fallback even when the deep-link
- stack contained no Player destination; the fallback now replaces the current
- detail entry and navigates to the typed target (or Home).
-
-The review also found that the app and extracted playback golden helpers were
-accepting a first-run screenshot as its own baseline. Both helpers now require
-checked-in `androidTest/assets/ui-goldens/*.png` files unless the explicit
-`recordGoldens=true` instrumentation argument is supplied. The app and playback
-baseline PNGs are present in the current hardening working tree, and the
-recording-disabled shell, premium-route, navigation-contract, missing-baseline,
-and playback-overlay runs previously passed on the API 36
-`Television_1080p` emulator. The hardening files still need their own commit;
-once committed, this makes the golden and connected navigation gates
-reproducible for future reviewers.
-
-The follow-up also hardens startup player delivery against event ordering: a
-Live TV destination resume is recorded even if startup navigation acknowledgement
-arrives later, and acknowledgement retries the deferred player enqueue. Focused
-coordinator coverage now includes resume-before-acknowledgement, deferred player
-lookup completion, and catalog-layout route reconciliation.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md
deleted file mode 100644
index 824c9671..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md
+++ /dev/null
@@ -1,219 +0,0 @@
-# Compose Reduction Phase 5 — Catalog execution report
-
-Date: 2026-09-04
-Status: structurally extracted; connected Catalog goldens now pass
-
-## Outcome
-
-The Dashboard, Movies, Series, VOD, Favorites, Search, and media-detail
-presentation now live in `:feature:catalog`. The app remains the composition
-root: it supplies platform/service adapters, player-request and payload
-compatibility, shell/Settings composition, and root graph registration. Route
-strings, arguments, return destinations, callbacks, focus/test semantics, and
-the four golden test names were preserved; the feature-only baselines were
-re-recorded after the original app-shell fixtures proved non-equivalent.
-
-This report covers the Catalog slice only. Playback, Provider, Settings, Live,
-and System status remain governed by their existing Phase 5 reports; Phase 5 is
-not complete because cross-slice acceptance and paired-performance gates remain
-open. `:feature:system` is now structurally extracted.
-
-## Delivery sequence and rollback
-
-The extraction was delivered in small commits from rollback point
-`d303e2aa` (pre-extraction design/plan links):
-
-| Task | Commit | Result |
-|---:|---|---|
-| 0 | `626903f0` | inventory, dependency ledger, focused baseline, before measurements |
-| 1 | `70bb99ea` | module and fail-closed boundary |
-| 2–4 | `ed3f819b`, `70023eea`, `9ca8cf0d` | contracts, app adapters, localized resources |
-| 5–6 | `6a84d503`, `91782978` | shared primitives and browse presentation |
-| 7–9 | `10c3a9c4`, `3fc94ea0`, `68423dba` | detail, Dashboard, Search/Favorites |
-| 10 | `e3238404` | feature-owned graph registration |
-| 11 | `4c7871fd` | goldens, legacy ownership cleanup, structural gates |
-| 12 | `5f1e615e` | connected acceptance evidence and semantics fix |
-| 13 | current documentation commit | isolation/build measurements and status report |
-
-To roll back the Catalog work, return to `d303e2aa`; the intermediate commits
-are intentionally independently revertible.
-
-## Ownership and boundary
-
-`feature/catalog` production code contains 50 Kotlin/Java files, 24 unit-test
-files, and 3 instrumentation Kotlin/Java files, plus feature resources and
-four route goldens. The six app Catalog screen directories are absent. The
-Catalog boundary reports exactly these project dependencies:
-
-```text
-:core:navigation, :core:ui, :domain, :data
-```
-
-The boundary scan passes with zero forbidden main-source references (`:app`,
-`:player`, sibling features, `MainActivity`, and root NavController types).
-The temporary `PreferencesRepository` and `ProviderSyncStateSource` imports
-are recorded in the transitional dependency ledger. Catalog has no `:player`
-or sibling-feature dependency.
-
-App-owned adapters cover Cast route selection, stream preparation, downloads,
-update/install messaging, navigation payloads, player requests, and the shell
-and Dashboard customization dialog. Favorites intentionally remains a direct
-feature-host surface; no new Favorites route was introduced.
-
-## Resources, tests, and goldens
-
-The locale audit passed with 351 expected keys, zero missing/value/format
-mismatches, zero unexpected keys, and 525 explicitly allowed source-locale
-fallbacks. No app locale key was removed because app-shell and compatibility
-consumers still resolve shared strings.
-
-The four Catalog goldens are feature-owned and reviewed at 1920x1080:
-
-```text
-route_dashboard_default.png
-route_movies_landing.png
-route_series_detail.png
-route_search_results.png
-```
-
-The feature unit, lint, Android-test compilation, resource processing, boundary,
-and debug assembly gates passed. The app navigation-boundary task also passes
-after removing obsolete `HomeGraph.kt`/`CatalogGraph.kt` expectations.
-
-## Connected/device evidence
-
-Target: `emulator-5554`, `Television_1080p(AVD) - 16`, API 36, AOSP TV x86,
-1920×1080 physical display, 320 dpi, animation scales 1.0.
-
-- Feature connected suite: 5/5 passed after correcting the golden harness to
- capture the full unpadded 1920x1080 Canvas surface and re-recording reviewed
- feature-only baselines.
-- Latest direct-Favorites-host-inclusive Catalog connected rerun: 6/6 passed;
- this added the real Favorites composable host and Activity Back cancellation
- regression on the same target.
-- App `AppNavigationContractTest`: 3/3 passed.
-- App `PlatformCompatibilityMatrixTest`: 4/4 passed.
-- No Catalog-caused fatal exception was reported by these runs.
-
-The partial seeded production smoke pass on the API 36 TV emulator loaded Home,
-Live TV (1,319 synced channels in the latest run), Movies, Series, and Search.
-Movies and Series
-correctly reported `Sync needed` because the configured public M3U fixture has
-live channels only; Search accepted `3ABN` and rendered seven Live TV results
-(Movies 0, Series 0), while the earlier `CNN` query rendered its no-match
-state.
-
-As a follow-up, a temporary local Xtream-compatible fixture was used with the
-same debug `MainActivity` on this emulator. It returns one live channel, 63
-movies, 63 series, movie metadata, and two episodes per series. The first two
-items are stable fixture titles; generated pagination-category items make the
-second-page boundary deterministic. The run rendered Dashboard media shelves,
-Movies and Series browse cards, the Movies full-library entry/back path, movie
-and series details, season/episode rows, movie and series favourite toggles,
-saved filters, and a five-result cross-type Search query (Live TV 1, Movies 2,
-Series 2). The semantic route/title/count evidence and the exact fixture
-limitations are recorded in
-`validation/phase5_catalog/task14-xtream-fixture-journeys.md`. The fixture is
-now available as the checked-in development tool
-`tools/catalog_xtream_fixture.py`; the repeatable semantic journey is exposed
-by `tools/catalog_connected_validation.py`. This remains diagnostic evidence,
-not a production-provider pass.
-
-The same semantic harness now also opens Settings Browsing, switches Infinite
-scroll off, proves the selected Movies and Series libraries reach `Load more
-(60/63)`, activates the controls to render `Pagination Movie 63` and
-`Pagination Series 63`, and restores Infinite scroll on. It then opens the
-Settings-owned Dashboard shelf customization dialog through TV focus, proves
-cancel leaves the persisted seven-shelf order unchanged, saves a six-shelf
-edit, and resets/saves the default order again. The latest rerun also enters
-the Favorites reorder surface for both Movies and Series, moves each fixture
-pair in both directions, verifies the saved order after leaving edit mode,
-re-enters to prove persistence, and restores the original order. Its report
-is `build/catalog-validation-download-completion-retry/report.json`; the full
-surface list is recorded in
-`validation/phase5_catalog/task14-xtream-fixture-journeys.md`. Movie and
-Series pagination, browse reorder, the Movies browse return path, and
-download completion are covered. A direct test host now renders the real
-Favorites composable and verifies Activity Back cancels reorder; the host is
-not a production route. A focused Catalog behavior test now covers merged card
-semantics and RTL placement at 1.5x font scale, and the full Catalog
-instrumentation suite passes with animator scale 0. Cast chooser,
-touch/phone/tablet, the full production accessibility-service journey, and
-the paired performance matrix remain open. The same 7-test Catalog
-instrumentation suite also passed with the API 36 AOSP TalkBack service
-enabled. The fixture journey reached the Catalog reorder surface under
-TalkBack, but the semantic ADB harness could not maintain its D-pad focus
-path consistently (`Reorder Items` on one run, the first browse card on the
-next); direct TalkBack instrumentation evidence is therefore positive, while
-the production fixture accessibility journey remains unclosed.
-The direct Favorites composable remains intentionally unregistered by the
-Catalog architecture plan; its direct host now exercises the real reorder
-panel and BackHandler, while the ViewModel cancel/Back path restores the
-preview without persisting it, with focused host and unit regressions.
-Copy URL is independently covered by a focused Robolectric clipboard-read
-regression; the API 36 shell limitation remains documented as a device-shell
-evidence limitation.
-
-## Build isolation and performance
-
-Five reversible feature source-edit samples and five feature test-compile
-samples all passed. Warm source-edit median was 8.3s (8.2–8.5s); warm
-test-compile median was 8.7s (8.7–9.0s). Raw profile names and protocol are in
-`validation/phase5_catalog/performance-before.md` and
-`validation/phase5_catalog/performance-after.md`.
-
-An actual Catalog-only edit followed by `:app:assembleDebug` executed the
-Catalog Kotlin task; sibling feature Kotlin tasks were `UP-TO-DATE` and did
-not execute. A clean debug assembly passed in 80.7s, warm debug assembly in
-14.9s, and Beta/Release packaging passed in 332.3s. These are not a
-cache-equivalent five-run clean-build comparison. A focused Dashboard
-`dashboardVerticalScroll` rerun on the same emulator passed 1/1 with five warm
-iterations after public-M3U live-channel activity populated a scrollable Home
-recent-channel shelf. It emitted `frameCount` min/median/max 88/91/98,
-`frameDurationCpuMs` P50/P90/P95/P99 60.0/70.7/78.7/92.7, and
-`frameOverrunMs` P50/P90/P95/P99 63.1/78.9/89.7/106.1. The earlier empty Home
-fixture had no scrollable content and failed before metrics with
-`Observed no renderthread slices in trace` from `FrameTimingQuery`. This rerun
-is diagnostic after-run evidence only: no cache-equivalent pre-extraction run
-was captured, and the public M3U fixture still lacks VOD/series content, so the
-formal paired performance gate remains open.
-That benchmark comparison is explicitly deferred until the computer/device has
-stable capacity for matched baseline and current runs.
-
-## Profiles and known failures
-
-The generated profile workflow was rerun on 2026-09-03 with
-`:app:generateBaselineProfile` on the API 36 TV emulator. It passed in 19m
-04s; the connected profile suite completed 18 tests, with the eight configured
-macrobenchmark cases skipped by configuration. The refreshed sources contain
-feature Catalog descriptors and no legacy app Catalog descriptors:
-
-| Source | legacy app Catalog descriptors | feature Catalog descriptors |
-|---|---:|---:|
-| `baseline-prof.txt` | 0 | 885 |
-| `startup-prof.txt` | 0 | 763 |
-
-Generated files were not hand-edited. The run exercised seeded Home/Live
-journeys; the public M3U fixture has no VOD/movie/series content. The separate
-temporary Xtream journey run is documented as diagnostic and does not change
-the profile input or close the full Catalog production journey gate.
-
-The repository-wide `:app:check` attempt reached the app lint task but failed
-on 675 lint findings (first: API-level `Trace.beginAsyncSection` in
-`AppStartupCoordinator.kt`; 901 baseline-filtered findings were also reported).
-This is outside Catalog ownership and is not claimed as a Catalog pass. The
-Catalog lint task itself passed with 802 warnings and one hint. Gradle also
-reported 24 configuration-cache diagnostics for the existing boundary task's
-execution-time project access.
-
-## Next acceptance work
-
-1. Use the checked-in Xtream fixture (or approved provider data) to execute the
- remaining browse, detail-action, and production accessibility-service journeys,
- and keep the fixture invocation in the validation protocol. Direct TalkBack
- Catalog instrumentation is already passing; the end-to-end fixture harness
- still needs a stable TalkBack-compatible focus path.
-2. Capture a cache-equivalent pre-extraction run and pair it with the
- successful Dashboard benchmark evidence, using the same seeded content,
- device, iteration count, and compilation mode. Keep the 374 accepted lint
- findings and all other Phase 5 feature gates tracked separately.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_LIVE_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE5_LIVE_REPORT.md
deleted file mode 100644
index b7bd3711..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_LIVE_REPORT.md
+++ /dev/null
@@ -1,332 +0,0 @@
-# Compose Reduction Phase 5: Live report
-
-Date: 2026-09-02
-Slice: `:feature:live` (Phase 5 order 4 of 6)
-Rollback point: `ea349d83dceb8baa45ef1a18b7a45fa295a264f5`
-
-## Outcome
-
-Home Live TV, categories, channel preview, EPG/Guide presentation, state, and
-dialogs now compile under `:feature:live`. The app remains the composition root
-for route registration, platform/scaffold adapters, typed player request
-mapping, and Playback MultiView composition. The slice is not a Phase 5
-completion claim: the remaining archive-replay fixture, formal paired
-performance/clean-build guardrails, and neighboring slice gates remain open.
-
-## Ownership and boundaries
-
-Live-owned production includes Home and Guide screens, ViewModels/state,
-preview pane and lifecycle calls, category/channel/EPG hosts, dialogs, remote
-shortcut policy, default resources, unit tests, and connected presentation
-tests. `:app` owns `LiveGraph`, route compatibility, `AppLivePlaybackRequestMapper`,
-the Home/EPG scaffold and platform adapters, and the Playback MultiView planner
-adapter. Preview handoff remains delegated to the existing Playback singleton.
-
-The feature boundary verifier passes with only the approved project
-dependencies: `:core:navigation`, `:core:ui`, `:domain`, `:data`, and `:player`.
-No app or feature-to-feature implementation imports remain. The two temporary
-`:data` imports (`PreferencesRepository` and `ProviderSyncStateSource`) remain
-documented for the Phase 7 domain-contract migration.
-
-## Delivered checkpoints
-
-The implementation was built incrementally from the rollback point through the
-following Live checkpoints (plus the final lint fix):
-
-- `bd4f2649`, `96e9c9b9`: move Home and EPG presentation units;
-- `108ca79f`, `aeb7cea7`, `6766c528`, `d27fff05`, `597f9706`, `e2422a6b`:
- Home composition ports, adapters, and remote dispatch;
-- `abc05272`, `6d2d6d9d`, `ccf06258`, `c8957551`, `62124ade`, `844a1f19`,
- `a71ec037`, `979cf11c`, `b715d026`, `57b9389c`, `218d6b4a`, `95d40b96`,
- `e7864086`, `19b95766`: extracted Home/EPG policies, focus, grid, and
- presentation seams;
-- `ffcf710b`, `23c26d64`, `8a3e8cd1`, `354d2842`, `28971f89`, `7de15c6a`:
- preview/route/graph/request contracts and connected presentation coverage;
-- `0cf3927c`: Compose lint-safe resource lookups and Crossfade state usage.
-
-## Verification
-
-- `:feature:live:verifyFeatureLiveBoundary`: passed.
-- `:feature:live:connectedDebugAndroidTest`: final full-suite rerun passed 24
- tests, 0 failures, 0 errors, and 0 skipped (the original 13 tests plus the
- Home 6-test and Guide 5-test screen-host suites). XML is under
- `feature/live/build/outputs/androidTest-results/connected/debug/`.
-- `LivePresentationGoldenTest`: passed 6/6 at 1920x1080 after visual review.
- The checked-in feature baselines cover selection chips, locked/unlocked
- category rows, progress/locked channel surfaces, the app-parity row fixture,
- the selected/unavailable source switcher, and the reorder bar. The separate
- app-owned route golden is covered by the review and regeneration checkpoint
- below.
-- Focused feature/app structural bundle (unit tests, lint, Android-test
- compilation, app compilation/tests, and debug assembly): passed after the
- lint fix. Gradle reported 235 actionable tasks, 23 executed, 212 up-to-date.
-- Final fresh verification (`:feature:live:verifyFeatureLiveBoundary`,
- `:feature:live:check`, the focused app-adapter and Playback-handoff tests,
- app debug assembly, and `verifyBaselineProfileSources`): passed in 7m 39s
- (375 actionable tasks; 64 executed, 37 from cache, 274 up-to-date). The
- focused XML results contain 4 app-adapter and 9 Playback-handoff tests, all
- passing with no failures, errors, or skips.
-- Fixture-flow follow-up on 2026-08-30: the reversible Live quick-filter
- save/reload/remove journey passed. `Movies` survived force-stop/relaunch,
- was removed through Settings > Browsing > Live TV Quick Filters, and the
- final Live drawer reported no saved filters. The selected synthetic
- schedule/PIN contract suite also passed: 39 tests, 0 failures, 0 errors,
- and 0 skips across `:feature:live` and `:domain`.
-- Direct app adapter coverage was completed test-first for stream-result
- preservation, exactly-once surface refresh, MultiView occupied-slot/capacity
- mapping, and preview-origin mapping. The four app tests plus the existing
- Playback handoff suite passed together in 40s (163 actionable tasks; 12
- executed, 2 from cache, 149 up-to-date).
-- Graphify refreshed at 15,838 nodes, 30,884 edges, 440 communities. The
- legacy app Home/EPG source directories and imports are absent.
-- The locale audit compared all 245 moved defaults and placeholder sequences
- against every app locale. It corrected two mojibake feature defaults
- (`live_preview_loading` and `time_range_format`) and found no remaining
- default or placeholder mismatch. All 25 translated feature locale files now
- have `featureMissing=0`, `featureValueMismatch=0`, and
- `featureUnexpected=0`; app copies remain for shared consumers.
-- The app-owned Live route golden was reviewed and regenerated from a fresh
- 1920x1080 recorder capture after confirming the layout/content matched and
- the drift came from the current `AppColors.Canvas`/gradient palette. The
- non-recording `live_route_matchesGolden` assertion now passes 1/1.
-- The extracted Home/Guide behavior coverage is now present in
- `HomeScreenBehaviorTest` (6 tests) and `EpgScreenBehaviorTest` (5 tests).
- These validate presentation-host contracts; they do not close the
- fixture-dependent ViewModel journey matrix below.
-- The documented post-extraction clean debug command passed twice: 4m 46s
- after configuration-cache invalidation (136 executed, 73 from cache) and 1m
- 01s on the final tree with configuration-cache reuse (95 executed, 114 from
- cache). Both had 209 actionable tasks. Five final-tree warm no-change builds
- passed with a 13.690s median and 13.984s average; every run had 1 executed
- and 197 up-to-date tasks. The 3.7x clean-run spread makes the formal clean
- guardrail inconclusive without cache-equivalent pre/post snapshots. Exact
- samples and limitations are in `validation/phase5_live/performance-after.md`.
-- Current-tree guardrail rerun (2026-09-01) also passed two warm no-change
- `:app:assembleDebug --profile` samples (48.435s and 20.162s); the formal
- paired target remains open because the pre-extraction record is no-change,
- not an equivalent source-edit snapshot.
-- Fresh `:app:generateBaselineProfile` passed on the `Television_1080p` API 36
- emulator in 25m 13s: both `BaselineProfileGenerator` tests passed and the
- eight separate macrobenchmark tests were skipped by configuration. It
- generated 49,409 baseline and 32,151 startup rules. The follow-up
- `verifyBaselineProfileSources`, `:app:assembleBeta`, and
- `:app:assembleRelease` passed in 10m 27s (387 actionable tasks; 142
- executed, 14 from cache, 231 up-to-date). Fresh descriptor scans found zero
- stale app Home/EPG matches and 1,533 baseline / 43 startup `feature/live`
- matches. The earlier no-device and native-memory failures remain recorded as
- historical attempts; generated profiles were never hand-edited. A direct
- `:benchmark:connectedNonMinifiedReleaseAndroidTest` attempt also passed in
- 13m 09s, but its XML again reports the eight macrobenchmark cases as
- `ignored (-)`. Direct runner probes then produced emulator-only startup
- diagnostics: `coldStartupNoCompilation` median TTID 991.7 ms and
- `coldStartupWithBaselineProfile` median 947.4 ms over 10 iterations. A
- FrameTiming probe failed because the emulator trace contained no RenderThread
- slices, so the broader macrobenchmark performance gate remains open.
-
-- The authorized post-reboot rerun of
- `:benchmark:connectedNonMinifiedReleaseAndroidTest` also completed with
- `BUILD SUCCESSFUL` in 23m 59s. Its XML recorded 10 tests, 0 failures, 0
- errors, and 8 skipped: `startup` and `criticalJourneys` passed, while every
- `StreamVaultMacrobenchmark` case was skipped by the current instrumentation
- configuration. This confirms the rerun is healthy but does not produce the
- missing FrameTiming/navigation measurements; the macrobenchmark gate remains
- open and no generated profile was hand-edited.
-
-- The explicit Macrobenchmark rerun on 2026-09-05 completed 8/8 tests with
- 0 failures and 0 skipped after the cold-start methods began force-stopping
- the release target in per-iteration setup. Both release startup methods
- completed 10 iterations; the six seeded interaction journeys completed five
- iterations each. This closes the emulator execution gate. The matched paired
- performance comparison and physical-device measurement remain open. Full
- metrics are recorded in `validation/phase5_system/macrobenchmark-validation.md`.
-
-## Runtime acceptance
-
-Using `E:\androidSdk\platform-tools\adb.exe` on `emulator-5554`
-(`Television_1080p`, API 36, `AOSP_TV_on_x86`), both channels were opened from
-Live TV preview into full playback and captured for approximately two minutes:
-
-| Channel | Frames / cadence | Unique hashes | Media session | Logs |
-| --- | --- | ---: | --- | --- |
-| `01 00s Replay` | 61 / ~2 s / ~120 s | 61 | `PLAYING`, `error=null` | HLS prepare + `first-frame-success`; no fatal/stuck/MPEG-TS markers |
-| `02 3ABN Dare To Dream Network` | 61 / ~2 s / ~124 s | 61 | `PLAYING`, `error=null` | HLS prepare + `first-frame-success`; no fatal/stuck/MPEG-TS markers; two non-fatal video-stall/reprepare warnings |
-
-Full PNG, hash, media-session, and logcat evidence is in
-`validation/phase5_live/task10-live-playback/`. The connected and route smoke
-record is `validation/phase5_live/task10-connected-validation.md`.
-
-The bounded Guide journey moved focus right/right through the timeline,
-down/down through program rows, and left/up back toward channel content while
-remaining on `streamvault.destination:epg`; Back restored
-`streamvault.destination:home`. The corresponding local screenshots and
-hierarchies are under `build/adb-validation/guide-journey-*` and
-`guide-back.xml`.
-
-A follow-up D-pad-first pass opened Live TV from Home, selected the `Movies`
-category (77 channels rendered), moved into the preview pane, and opened Guide
-through the top navigation. Valid PNG/XML artifacts are in
-`validation/phase5_live/task10-manual.md` and the adjacent
-`task10-manual/` directory. Touch taps on Guide `Program Search`/`Options` did
-not change state on this emulator, although both controls were later reached
-through D-pad focus. In that earlier pass, visible channels had no schedule
-data, so schedule-dependent actions were not yet exercisable. A
-second Enter handed the
-selected Guide channel to the existing full-screen player; its overlay exposed
-EPG, Multiview, Stats, Record, and Pause, and Back returned to the Guide route.
-Activating Multiview returned to the Live route with the empty-favorites/queue
-state. A follow-up populated-planner journey switched to `All Channels`,
-long-pressed `01 00s Replay`, added it through the management dialog, assigned
-it to slot 1, launched MultiView, and returned with Back. The planner evidence
-is under `validation/phase5_live/task10-multiview/`; the remaining manual matrix
-is still fixture- or input-path-dependent. A follow-up fixture pass also
-validated adding/removing two channels from Favorites and entering populated
-Favorites reorder mode; the final device state was restored to zero favorites.
-The D-pad-activated category search accepted `Movies` and filtered the
-category rows. The D-pad-activated All Channels search accepted the `Replay`
-input path (received as `Rep`) and reported exactly one result, `01 00s
-Replay`; both queries were cleared and All Channels returned to `1,467`.
-The Add quick filter dialog accepted `Movies` and enabled `Save filter`; it
-was then saved, survived a force-stop/relaunch, and was removed through
-Settings. The final Live drawer and Settings row reported no saved filters;
-the device artifacts are under
-`validation/phase5_live/task11-fixture-flows/`. The same fixture pass
-then opened the Movies `Lock Group` action and
-reached the focused `Enter PIN` keypad; no PIN was submitted because the
-configured value was unavailable, Back canceled the dialog, and Movies
-remained unlocked (`live_lock_options.*`, `live_pin_dialog.*`, and
-`live_after_pin_cancel.xml`).
-The populated Favorites reorder path was also exercised: `00s Replay` was
-grabbed and moved above the other fixture favorite with D-pad Up, then the
-unsaved reorder was canceled and both favorites were removed; the final count
-was zero (`live_reorder_dragging.xml`, `live_reorder_moved_up.xml`, and
-`live_fav_restored_again.xml`).
-The Save Order callback was also validated in a reversible round trip: the
-swapped order was saved, the original order was restored and saved again, and
-both favorites were removed (`live_save_order_saved.xml`,
-`live_save_order_restored.xml`, and `live_save_order_final.xml`).
-
-A further reversible device pass changed the locale from `en-US` to `ar-SA`:
-Live TV and Guide stayed interactive, Arabic labels rendered, and the Live
-sidebar plus Guide grid mirrored correctly before the locale was restored.
-The three Android animation scales were then set to `0.0`; Guide remained
-interactive through D-pad input with no app fatal marker in the bounded scan,
-and all scales were restored to `1.0`. D-pad also opened Program Search and
-Guide Options; Guide mode was switched and restored. The Compose text field
-accepted Home category/channel text input after D-pad activation. Guide
-Program Search was also activated through D-pad and accepted `Movies`; the
-dialog was dismissed and Live TV restored, but the loaded provider's visible
-guide rows still reported `No schedule`, so a distinct filtered program-result
-assertion remains unavailable. Screenshots and hierarchies are recorded in
-`validation/phase5_live/task10-manual/`.
-
-The planner edge follow-up populated two slots, removed slot 2, replaced the
-occupied slot 1 with the pending channel, and used Clear All. The planner
-dismissed as designed, and the final Live fixture state remained Favorites `0`
-and All Channels `1,467`. Evidence is under
-`validation/phase5_live/task10-multiview/`.
-
-A further planner pass launched the two-slot configuration with `00s Replay`
-and `3ABN Dare To Dream Network`; the MultiView surface showed both channel
-labels and two empty slots. The planner was reopened from a queued channel,
-Clear All removed both active slots, and Live TV was restored with no split
-badge, Favorites `0`, and All Channels `1,467`.
-
-A provider-backed EPG pass on 2026-09-01 then exercised the previously blocked
-schedule/PIN paths. In `VIP | 4K ULTRA HD`, `VIP - TNT SPORT ULTIMATE 4K`
-displayed the program `This is TNT Sports Ultimate`; Guide moved between the
-current and future timeline, and the future-program dialog exposed Watch Live,
-Remind Me, Record, recurring record, Program Details, and Cancel. With exact
-alarm access enabled for the app, `Remind Me` changed to `Cancel Reminder` and
-registered an exact alarm; cancellation removed it and restored `Remind Me`.
-Privacy PIN `1234` was saved, the protection level was temporarily changed to
-`LOCKED`, and the same PIN successfully unlocked the locked
-`VIP | GOLDEN EVENTS` category. The temporary category lock was removed and
-the original `HIDDEN` protection level restored. Favorites and saved quick
-filters ended empty. Evidence is in `build/pin-screen.png`,
-`build/unlock-success.xml`, `build/remind-success-clean.xml`,
-`build/remind-cancelled.xml`, and `build/final-favorites.xml`.
-
-A fresh D-pad follow-up after the profile/package validation reopened Guide in
-the currently installed debug state and confirmed the provider currently
-exposed to that package is the M3U fixture without Guide schedule coverage.
-Guide rows rendered `No schedule` / `No schedule data available`; activating
-Program Search with `TNT` returned `No matching programs in this guide window`.
-Back returned to the Home destination with the top navigation focused. The
-captured hierarchies are `build/current-guide.xml`,
-`build/current-guide-search-no-results.xml`, and
-`build/current-guide-after-search.xml`. This is recorded as fixture-limited
-evidence, not as a filtered-result or focus-restoration pass.
-
-A temporary local XMLTV fixture was then used on 2026-09-02 because the native
-Xtream Guide endpoint returned `auth=0` and no usable schedule rows. The fixture
-parsed seven programmes and matched cached channel IDs/names across the
-restored Xtream and M3U providers. It exercised the remaining Guide branches:
-
-- Program Search accepted `VIP`, returned the three matched fixture programmes,
- and restored focus to the selected VIP channel row after dismissal
- (`guide-search-vip-fixture.png`, `guide-search-focus-before.png`, and
- `guide-search-focus-after.png`). This closes the filtered-result and search
- focus-restoration checks.
-- A future fixture programme scheduled successfully after exact-alarm access
- was enabled (`guide-record-scheduled-allow.png`). Two overlapping schedules
- from different providers produced the real Recording Conflict dialog naming
- both conflicting recordings; it was dismissed with Cancel so no replacement
- occurred (`guide-recording-conflict.png`). The scheduled fixture rows were
- subsequently cancelled in Settings > Recording.
-- The `-3h` Guide window exposed a fixture archive programme
- (`guide-before-archive.png`). Activating it reached the archive callback but
- returned to Home because the current provider has no replay stream/IDs; no
- archive playback claim is made. The archive gate remains open until a provider
- fixture with catch-up metadata is available.
-- Existing MultiView evidence already covers empty and populated planner entry,
- slot-1 placement, slot removal, occupied-slot replacement, Clear All,
- two-slot launch, and Back restoration under
- `validation/phase5_live/task10-multiview/`. No additional provider-backed
- MultiView branch was left unexercised by this fixture run.
-
-The temporary XMLTV source, provider assignments, scheduled fixture recordings,
-and emulator backup were cleaned up afterward. The active provider was restored
-to Xtream and both provider Guide source policies were returned to Auto; the
-local fixture server was stopped.
-
-## Open gates
-
-- Deferred — provider fixture: the remaining Home/EPG manual gate is archive
- playback. The temporary XMLTV
- fixture proved archive-window rendering and callback routing, but the active
- provider supplied no replay stream/IDs, so archive playback could not be
- verified. Search focus restoration, filtered results, recording scheduling,
- and the real cross-provider conflict dialog now pass. Hidden category/channel
- hide and restore paths, PIN submission/unlock, reminder scheduling/cancellation,
- RTL/reduced-motion smoke checks, and the core MultiView
- placement/removal/replacement/clear and two-slot launch paths are covered.
- Quick-filter save/reload/remove is no longer open.
-- The feature presentation golden review and the app-owned Live route golden
- review are complete. The app baseline was regenerated from the reviewed
- current capture; the old baseline is preserved as
- `route_live_browse_baseline_before.png` in the ignored follow-up evidence.
-- Deferred — performance follow-up: the formal paired performance target and
- clean-build guardrail remain open.
- Five
- post-extraction Live source-edit and five moved-test samples are recorded in
- `validation/phase5_live/performance-after.md`; they establish the
- post-extraction envelope but are not a comparable pre/post pair because the
- available pre record is no-change rather than source-edit.
- Two current clean task-graph runs passed at 4m46s and 1m01s under materially
- different cache states, so the clean guardrail is inconclusive; five
- final-tree warm no-change samples passed with a 13.690s median.
- `verifyBaselineProfileSources`, Beta assembly, and Release assembly pass
- against the regenerated 49,409 baseline / 32,151 startup rules. The
- baseline-profile producer's aggregate task still selects only its profile
- rules, but the explicit Macrobenchmark rerun now passes all eight journeys.
- The matched paired macrobenchmark comparison remains open even though
- profile generation, descriptor freshness, and emulator execution now pass.
- Details are in
- `validation/phase5_live/profile-validation.md`.
-- Deferred — neighboring slices: Playback, Provider, and Settings
- acceptance/performance gates remain governed by their existing reports. All
- six feature unit-test and lint tasks now pass; the current `:app:lintDebug`
- gate also passes with the existing baseline. Catalog and System extraction slices are now
- structurally complete; Phase 5 is not complete until the deferred
- acceptance, paired-performance, physical-device, and production-credential
- gates are resolved.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_PLAYBACK_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE5_PLAYBACK_REPORT.md
deleted file mode 100644
index 38585376..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_PLAYBACK_REPORT.md
+++ /dev/null
@@ -1,327 +0,0 @@
-# Phase 5 Playback Feature Extraction Report
-
-Date: 2026-08-27
-Branch: `feature/improveCompose`
-Implementation endpoint: `7df21b86` (`refactor: move playback graph registration to feature`)
-
-## Outcome
-
-The playback implementation is now owned by `:feature:playback` for the
-Player, MultiView, support services, presentation resources, tests, and graph
-registration. The app remains the composition root: it supplies platform
-adapters and registers the feature graph through
-`registerPlaybackGraph(...)`.
-
-The structural Phase 5 work is complete through Task 9 and the Task 10
-connected checks that were available in the initial environment. That initial
-emulator session had no seeded provider or channels; a later seeded-provider
-run is recorded below. The full live-TV acceptance gate remains open because
-the seeded run encountered repeated recoverable live-window source errors. No
-live-playback acceptance pass is claimed.
-
-## Ownership and public boundary
-
-- `:feature:playback` owns Player and MultiView screens, coordinators, overlays,
- playback support services, feature resources, tests, and
- `PlaybackGraph.kt`.
-- `:app` owns `MainActivity`, `AppNavHost`, app navigation orchestration,
- platform/service adapters, and the callback that consumes typed player
- requests.
-- `registerPlaybackGraph(actions, platformHost, consumePlayerRequest)` is the
- feature graph entry point.
-- `PlaybackNavigationPolicy` owns supported playback URI schemes and
- navigation-request safety. `PlaybackRoutePatterns` owns the feature route
- constants.
-- The feature boundary has no production imports of `com.streamvault.app`,
- `NavController`, `NavHostController`, or `Media3PlayerEngine`.
-- The six intentional temporary `:data` imports remain ledgered in
- `COMPOSE_REDUCTION_PHASE5_TRANSITIONAL_DEPENDENCIES.md`; removing them now
- would break active provider/playback contracts.
-
-Rollback for the Task 4 graph move is `git revert 7df21b86`; the earlier
-feature commits are independently reversible in the Phase 5 extraction
-history.
-
-## Files, resources, and tests moved
-
-The extraction moved the Player and MultiView implementation trees, playback
-resources/localizations, support services, navigation graph, and their unit
-and Android tests into `feature/playback`. The transparent EPG guide remains
-app-owned because it is an EPG integration concern.
-
-Task 4 also moved the navigation safety policy and route constants into the
-feature. The new feature policy test covers null, blank, unsupported schemes,
-all ten supported schemes, field preservation, and request-instance
-preservation.
-
-## Build and boundary evidence
-
-Task 9's five-sample measurements are recorded in
-`validation/phase5_playback/build-after/README.md`. They establish feature
-local recompilation and app compile isolation; they do not justify a speed-up
-claim because the before/after commands and cache states differ.
-
-The following post-Task-4 gate passed:
-
-```text
-.\gradlew.bat :feature:playback:testDebugUnitTest :app:testDebugUnitTest --tests "com.streamvault.app.navigation.*" verifyFeaturePlaybackBoundary :app:compileDebugKotlin --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 12s
-```
-
-The fresh debug APK was rebuilt after cleaning the old output and installed
-from:
-
-```text
-app/build/outputs/apk/debug/app-debug.apk
-SHA-256: 0B346848DE643483914E54FC3EA98471C61F13DBB46922FAF25DEE2042E05DB1
-```
-
-## Connected validation
-
-Device: `Television_1080p(AVD) - 16`, API 36, emulator `emulator-5554`.
-
-Feature playback overlay goldens:
-
-```text
-:feature:playback:connectedDebugAndroidTest
-7 tests, 7 passed
-```
-
-The seven tests include the six checked-in overlay baselines and the
-recording-disabled missing-baseline guard.
-
-App connected suite:
-
-```text
-:app:connectedDebugAndroidTest
-27 tests, 22 passed, 5 failed, 0 errors
-```
-
-The five existing fixture/environment failures were recorded and not counted
-as passes:
-
-1. `DownloadForegroundServiceQuotaInstrumentationTest.reducedDataSyncTimeoutReleasesDownloadLease`
- — Android timeout callback did not release the shared quota lease.
-2. `LauncherProviderInstrumentationTest.watchNextProgram_supportsInsertUpdateDeleteRoundTrip`
- — the API 36 AOSP TV provider lacks the required `type` column.
-3. `PlayerSmokeTest.playerControlsOverlay_playButton_canReceiveFocus`
- — no `>` node was focused.
-4. `PlayerSmokeTest.categoryRailPanel_searchField_acceptsInitialFocusAndInput`
- — no text-entry node was focused.
-5. `PlayerSmokeTest.playerTrackSelectionDialog_selectsAudioTrack`
- — the existing check failed at `PlayerSmokeTest.kt:178`.
-
-Relevant passing connected coverage included the app navigation contracts
-(3/3), premium route goldens (6/6), shell goldens (3/3), and the recording-
-disabled golden capture. The XML result is retained at
-`app/build/outputs/androidTest-results/connected/debug/TEST-Television_1080p(AVD) - 16-_app-.xml`.
-
-The connected suite was rerun on 2026-08-26 after the navigation hardening;
-it reproduced the same 22/27 result and the same five failures above. The
-focused navigation-contract (3/3) and app golden-capture (1/1) runs passed,
-and the full playback overlay suite passed 7/7.
-
-The release-like profile gate was rerun with the seeded debug fixture installed
-alongside `com.streamvault.app`. The benchmark helpers now drive the release
-target during baseline collection (ordinary macrobenchmarks retain the debug
-fixture), and player-control detection uses the app's `MENU` control-toggle
-shortcut rather than the live channel-info `DPAD_CENTER` shortcut. The earlier
-full `:app:generateBaselineProfile` attempt failed at the one-shot player
-controls assertion; after this harness correction, the focused connected
-`BaselineProfileGenerator#criticalJourneys` test passed (1/1, 2026-08-26).
-The clean full `:app:generateBaselineProfile` rerun then passed on 2026-08-26
-in 39m28s: 10 profile-collection tests passed, 8 unrelated macrobenchmark
-tests were skipped, generated beta profiles were copied, startup rules were
-merged, and the merged profile was installed. The separate macrobenchmark
-performance and physical-device gates remain open. Follow-up
-`verifyBaselineProfileSources`, `:app:assembleBeta`, `:app:assembleRelease`,
-and `:app:bundleRelease` checks also passed.
-
-## ADB smoke evidence
-
-The fresh APK was installed with `E:\androidSdk\platform-tools\adb.exe` and
-launched as `com.streamvault.app.debug/com.streamvault.app.MainActivity`.
-The captured startup frame is
-`validation/phase5_playback/adb-smoke/task10-home-top-navbar.png`.
-
-The UI hierarchy reported `streamvault.destination:home` and a
-`HorizontalScrollView` containing `Home`, `Live TV`, `Movies`, `Series`,
-`Downloads`, `Guide`, `Search`, `Plugins`, and `Settings` across the top of
-the screen. This is direct evidence of the current top-navbar startup state;
-it does not validate player playback.
-
-## Manual, live, and profile gates
-
-- Provider-dependent manual journeys (player launch, return/Back priority,
- controls, seeking, overlays, numeric entry, zapping, PiP, Cast, MultiView,
- touch/mouse, RTL, and reduced motion) were not fully accepted; the seeded
- ADB run covered channel launch and sustained video, while the remaining
- journeys still need explicit manual coverage.
-- Two-channel live validation was run and is recorded below. Both channels
- rendered video and recovered to final `PLAYING`, but neither passed the
- stability gate because repeated live-window source-error transitions were
- observed during the required window.
-- `:benchmark:compileBenchmarkKotlin` passed. The focused release-like
- `criticalJourneys` connected run and the subsequent full
- `:app:generateBaselineProfile` generation/merge/install task passed after the
- benchmark key/selector fix (10 profile tests passed, 8 unrelated
- macrobenchmarks skipped). `verifyBaselineProfileSources`, beta/release APK
- assembly, and release AAB bundling passed afterward. A macrobenchmark
- performance comparison and physical-device validation remain open.
-
-### Fresh seeded-provider ADB validation (2026-08-26)
-
-The configured public M3U seed was rebuilt into the debug APK, installed with
-`E:\androidSdk\platform-tools\adb.exe`, and allowed to sync 1,459 channels.
-Both channels below rendered real video and produced 61 unique screenshots at
-two-second cadence. Neither is an acceptance pass because the stream entered
-repeated recoverable `BehindLiveWindowException`/`Source error` transitions
-during the two-minute window; the final media session recovered to
-`PLAYING` with `error=null`.
-
-| Channel | Frames / unique SHA-256 | Final media session | HLS evidence | Recovery/error evidence | Result |
-|---|---:|---|---|---|---|
-| 3ABN English | 61 / 61 | `PLAYING`, `error=null` | 8 prepares, 6 first frames | 5 live-window retries, 35 `state=ERROR` records | Open/fail |
-| 3ABN French | 61 / 61 | `PLAYING`, `error=null` | 4 prepares, 4 first frames | 3 live-window retries, 21 `state=ERROR` records | Open/fail |
-
-No `fatal-error`, stuck-player timeout, MPEG-TS fallback, or malformed-HLS
-fallback marker appeared in either isolated log. The captured frames visibly
-contain the two live broadcasts, so this is a playback-recovery stability
-failure rather than an empty catalog or navigation failure. Raw local
-artifacts are under `validation/phase5_playback/live-validation/` and remain
-uncommitted.
-
-### Cross-upstream ADB validation follow-up (2026-08-26)
-
-The capture transport was then moved onto the emulator so screenshot timing
-was not dominated by Windows-side ADB startup. Each run captured 61 frames and
-recorded on-device timestamps; the frame hashes and final media-session state
-were checked after pulling the completed directory.
-
-| Channel | Upstream family | Frames / unique SHA-256 | Timestamp intervals | Final media session | HLS/recovery evidence | Result |
-|---|---|---:|---|---|---|---|
-| 00s Replay | Pluto HLS | 61 / 61 | 1.41–2.74s, avg 2.09s | `PLAYING`, `error=null` | Continuous HLS playlist responses; 0 fatal/stuck/live-window/MPEG-TS markers | Stability pass |
-| 3ABN Dare To Dream Network | 3ABN HLS | 61 / 61 | 1.23–6.08s, avg 2.08s | `PLAYING`, `error=null` | 3 HLS prepares, 2 first frames, 1 live-window retry, 7 `state=ERROR`, 9 source-error records; no fatal/stuck/MPEG-TS fallback | Not accepted |
-
-The two channels use different upstream families, so the live-window behavior
-is not limited to the earlier English/French pair. The 3ABN run still renders
-video and recovers, but the repeated source-error transitions keep the Phase 5
-stability gate open. The long startup gaps on that run also make it unsuitable
-as a clean cadence pass. The raw local artifacts remain uncommitted under
-`validation/phase5_playback/live-validation/`.
-
-### Isolated two-channel rerun (2026-08-27)
-
-Each channel was isolated with `adb logcat -c` before launch and captured at
-the requested two-second cadence for 61 screenshots. Raw screenshots were
-transient and removed after hashing; no credentials, tokens, or other private
-payloads were retained.
-
-| Channel | Screenshots | Unique hashes | Final media session | Log findings | Result |
-|---|---:|---:|---|---|---|
-| 3ABN Dare To Dream Network | 61 | 61 | `PLAYING`, `error=null` | HLS prepare/first-frame; recoverable video-stall/reprepare and release-timeout retry; no fatal error, `BehindLiveWindowException`, MPEG-TS fallback, or stuck-player marker | Stability pass for this window |
-| 3ABN French | 61 | 56 | `ERROR`, `error=Source error` | Video stalls selected `XTREAM_TS_FALLBACK`; subsequent `MPEG_TS_LIVE` prepares targeted a malformed `1/live/...` URL and ended in `fatal-error` | Not accepted |
-
-The Dare-to-Dream run satisfies the screenshot, final-session, and no-fatal/
-no-fallback criteria for its isolated window. The French run does not: it
-reproduces the source/recovery failure and loses frame progression before the
-window ends. The two-channel Phase 5 stability gate therefore remains open;
-the recovery-policy implementation was not changed in this validation slice.
-
-### Manual seeded playback journey (2026-08-26)
-
-On the same debug APK/emulator, a manual remote-input pass completed the
-provider-backed route handoff: Home → Live TV → All Channels → `00s Replay`
-preview → fullscreen player. The fullscreen player rendered video, the MENU
-shortcut opened the extracted controls chrome (including Mute, PiP, Audio,
-Video Quality, and Split Screen), and the controls auto-hide returned to an
-unobstructed video frame after the configured timeout. BACK returned to the
-Live TV route with the channel preview still present; the media session was
-left in a recoverable paused state after leaving fullscreen.
-
-Captured evidence is under
-`validation/phase5_playback/manual-validation/20260826/`, including
-`manual-00s-controls-last.png` (controls visible),
-`manual-controls-autohide-end.png` (video after auto-hide),
-`manual-preview.xml`/`manual-full.xml` (preview/fullscreen route checks), and
-`manual-back-route.xml` (return to Live TV). This is partial manual coverage;
-seeking, numeric entry/zapping, track/quality dialogs, PiP, Cast, MultiView,
-touch/mouse, RTL, and reduced-motion journeys remain unaccepted.
-
-### Connected smoke and macrobenchmark follow-up (2026-08-26)
-
-The focused `:app:connectedDebugAndroidTest` rerun of `PlayerSmokeTest` used
-the same seeded emulator and reproduced the three known fixture/focus failures
-(controls play-button focus, category-rail search-field focus, and audio-track
-fixture setup); the mute-action test passed. This matches the previously
-recorded baseline and is not attributed to the extraction.
-
-The two playback-relevant Macrobenchmark journeys also completed five of five
-iterations on `Television_1080p(AVD) - 16`:
-
-| Journey | Frame-count runs | Frame-count median | CPU frame P50/P90 | Overrun P50/P90 |
-|---|---|---:|---:|---:|
-| `liveTvCategoryAndChannelNavigation` | 30, 31, 30, 30, 31 | 30 | 606.8 / 939.8 ms | 922.4 / 1,278.7 ms |
-| `playerControlsOpenAndNavigate` | 21, 31, 22, 32, 27 | 27 | 804.7 / 2,476.7 ms | 1,313.2 / 4,890.4 ms |
-
-The JSON and Perfetto artifacts are under
-`benchmark/build/outputs/connected_android_test_additional_output/` (the exact
-device directory is ignored build output). These runs prove that the seeded
-journeys execute after extraction, but the emulator timings are materially
-noisier than the earlier 2.5A snapshot and include live HLS preparation and
-decoder work. They are therefore diagnostic rather than a clean before/after
-performance comparison; the Phase 5 performance gate remains open pending a
-controlled paired baseline and, ideally, a constrained physical-TV run. Because
-the host was under heavy load during this capture, no performance conclusion
-should be drawn from these numbers; repeat the paired run on an idle host.
-
-### Navigation and golden hardening follow-up (2026-08-26)
-
-The in-progress review-hardening slice adds coordinator coverage for startup
-resume-before-acknowledgement, deferred player lookup completion, and
-catalog-layout route reconciliation. It also corrects the startup handoff so a
-resume event recorded before acknowledgement still triggers the deferred player
-command after acknowledgement. These changes are separate from the playback
-extraction endpoint and do not alter player preparation or recovery policy.
-The focused app unit rerun for this hardening slice passed 8/8 tests
-(`AppNavigationCoordinatorTest` 7/7 and `PlaybackProgressGuardrailTest` 1/1).
-
-### Attribution boundary
-
-The Phase 5 extraction is not the source-level change that introduced this
-behavior. `git diff --name-status 3b1becba..HEAD --
-player/src/main/java/com/streamvault/player/Media3PlayerEngine.kt
-player/src/main/java/com/streamvault/player/playback` is empty: the
-`Media3PlayerEngine` and retry-policy implementations were not modified by
-the Phase 5 extraction commits. The observed path is the existing
-`onPlayerError` → `handlePlaybackError` → live-window retry flow in `:player`.
-The two tested channels are also related streams from the same public 3ABN
-source, so a source-window interaction remains possible; this evidence does
-not yet justify changing the recovery policy.
-
-### Playback lint and connected golden follow-up (2026-09-05)
-
-The extracted Playback module now passes `:feature:playback:lintDebug` after
-targeted Compose/resource fixes: the golden fixture caches its
-`FocusRequester`, MultiView uses `LocalActivity` and a recomposition-aware
-`stringResource`, and the historically default-locale-only strings and
-intentional `one`/`other` plurals carry narrow lint annotations. The focused
-connected `PlayerOverlayGoldenTest` then passed 6/6 on the API 36 TV emulator.
-The full feature sweep also passed all six feature unit-test tasks and all six
-feature lint tasks. The current `:app:lintDebug` gate now passes with the
-maintained baseline covering 374 current findings; stale historical entries have
-been removed.
-
-## Graph and final status
-
-`graphify update .` was run after this report and plan update. It rebuilt
-`graphify-out/graph.json` and `GRAPH_REPORT.md` with 14,272 nodes, 27,916
-edges, and 362 communities; the refreshed graph includes the extracted
-feature sources and no longer places Player/MultiView implementation under
-the app source tree.
-
-Phase 5 structural extraction: **complete**.
-Runtime playback acceptance: **open pending live-recovery stability, manual
-journeys, matched macrobenchmark comparison, and physical-device validation**. The focused and full
-release-like profile journeys passed after the benchmark harness fix, including
-baseline-profile generation, merge, and installation.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_PROVIDER_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE5_PROVIDER_REPORT.md
deleted file mode 100644
index c0ae0ea0..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_PROVIDER_REPORT.md
+++ /dev/null
@@ -1,87 +0,0 @@
-# Compose Reduction Phase 5 — Provider Feature Report
-
-Date: 2026-08-27
-
-## Current status
-
-Provider extraction is **structurally complete** through the automated gates.
-The provider setup/edit/import/pairing UI, state, resources, tests, and route
-registration now live in `:feature:provider`. The app remains the composition
-root and supplies the settings-owned backup-preview renderer through the
-feature request contract.
-
-Runtime acceptance and paired incremental-performance gates remain open. The
-idle-host release-like profile gate was rerun successfully on 2026-08-27, but
-that is a packaging/profile check rather than a before/after timing result.
-This does not close Phase 5 overall; playback stability/manual journeys and
-later Phase 5 slices remain independent gates.
-
-## Ownership and boundary evidence
-
-- `:feature:provider` depends only on `:core:navigation`, `:core:ui`,
- `:domain`, and the ledgered temporary `:data` types.
-- The provider boundary task rejects app imports, root navigation controllers,
- and `MainActivity`; its fixture coverage passes.
-- No legacy provider package or old provider graph references remain in app or
- feature source.
-- Provider-only `setup_*` and `stalker_*` resources were removed from app
- resources and copied with tracked locale translations into the feature.
-- Generated baseline profiles were regenerated by the profile tool after the
- provider move; both generated files now contain feature-owned playback and
- provider symbols and zero legacy app-provider paths. They were not
- hand-edited.
-
-## Automated checks
-
-Passed:
-
-- provider ViewModel and core progress-fraction unit tests (the feature owns
- one provider unit-test source file plus five boundary fixtures);
-- the complete `:feature:provider:check` lifecycle (debug/release unit tests,
- lint, Kover verification, and the provider boundary guard);
-- app route compatibility tests (`AppRouteCodecTest`, `ExternalDestinationTest`);
-- provider instrumentation-test compilation;
-- focused connected provider completion test (1/1), app navigation contracts
- (3/3), platform compatibility matrix (4/4), recording-disabled golden guard
- (1/1), and playback overlay tests (7/7) on `Television_1080p(AVD)`;
-- the full app-connected suite completed 26 tests with 17 passed and 9 failed
- (one fewer app test than the historical 27-test run because the provider
- completion test now belongs to `:feature:provider`);
- the nine are the two known environment/fixture failures, three known
- PlayerSmoke failures, and four mixed-run timeout failures. The navigation
- class and all six premium golden methods passed in isolated reruns.
-- provider lint and debug assembly;
-- fresh `:app:assembleRelease` on 2026-08-27 (release lint/R8 completed with
- existing baseline/tooling warnings);
-- app Kotlin compilation and debug assembly;
-- core navigation/UI, playback, provider, and app navigation boundary tasks;
-- provider runtime dependency report;
-- `graphify update .` (14,445 nodes / 28,231 edges).
-
-The app graph-boundary regression found during the 2026-08-27 focused rerun is
-resolved: `FeatureGraphBoundaryTest` now includes the extracted provider graph
-root, and the isolated plus combined unit reruns pass.
-
-`:app:lintDebug` remains non-gating with 200 repository-wide errors (API-level,
-Compose, translation, and unused-resource debt); no provider-specific lint
-errors are present. The detailed automated evidence is in
-`validation/phase5_provider/task8-automated-validation.md`.
-
-## Open gates
-
-- Connected provider instrumentation passed. Manual acceptance has progressed
- through source switching, D-pad focus, M3U file and backup document-chooser
- launchers, Drive account-picker handoff, QR pairing start/stop, a seeded
- existing-provider edit/cancel journey, and a sanitized cold-start route
- smoke check. No credentials or provider submission were performed. Full
- create/edit save, preview/import completion, cold-start URI import,
- second-device QR, and exhaustive focus-restoration journeys remain open. See
- `validation/phase5_provider/task9-runtime-validation.md`.
-- Five paired compile/profile performance samples remain open because the
- provider extraction is not represented by a committed post-extraction ref;
- do not compare timings from the mixed dirty tree. See
- `validation/phase5_provider/performance-rerun.md`.
-- Release-like/profile regeneration and stale-path verification passed on the
- idle host; `:app:assembleRelease` also passed after regeneration.
-- Existing Phase 5 playback live-TV stability/manual/profile gates remain open
- under the playback report.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_ROADMAP.md b/docs/COMPOSE_REDUCTION_PHASE5_ROADMAP.md
deleted file mode 100644
index 54303ab3..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_ROADMAP.md
+++ /dev/null
@@ -1,163 +0,0 @@
-# Compose Reduction Phase 5 Roadmap
-
-Date: 2026-08-24
-
-## Purpose
-
-Phase 5 moves feature presentation from the monolithic :app module into a small
-number of independently compiled and tested feature modules. It is a sequence
-of behavior-preserving extractions, not one large migration.
-
-The governing design is
-docs/superpowers/specs/2026-08-24-phase-5-feature-module-extraction-design.md.
-
-## Delivery sequence
-
-| Order | Module | Primary move unit | Special gate |
-|---:|---|---|---|
-| 1 | :feature:playback | Player, overlays, MultiView, playback graph | Full two-channel long-duration live validation |
-| 2 | :feature:provider | Provider setup/edit/import/pairing | Create/edit/import, file launchers, focus and typing |
-| 3 | :feature:settings | Settings, parental control, backup/restore presentation | Backup/restore, dialogs, DAO/concrete dependency audit |
-| 4 | :feature:live | Home Live TV, categories, EPG, preview presentation | Preview handoff, EPG navigation, multi-channel live validation |
-| 5 | :feature:catalog | Dashboard, movies, series, VOD, favorites, search | Detail return routes, shelves, search, lazy-list behavior |
-| 6 | :feature:system | Welcome, downloads, plugins | Extract only if measured isolation exceeds module overhead |
-
-## Standard extraction shape
-
-Every feature follows the same review sequence:
-
-1. Record source ownership, app/project imports, resources, Hilt bindings,
- tests, and representative incremental build measurements.
-2. Create the Android library module and a boundary task that rejects :app,
- root controller, and unapproved project dependencies.
-3. Move typed route patterns and controller-free graph registration.
-4. Introduce the minimum platform/service ports needed to remove app imports.
-5. Move source, resources, ViewModels, Hilt bindings, unit tests, and
- feature-owned instrumentation tests with behavior unchanged.
-6. Wire AppNavHost to feature registration and remove the old implementation
- only after the new destination passes.
-7. Run independent feature checks, app integration checks, functional/device
- gates, build measurements, profile maintenance, and graph refresh.
-8. Write docs/COMPOSE_REDUCTION_PHASE5_<FEATURE>_REPORT.md and update the
- transitional dependency ledger.
-
-## Shared guardrails
-
-- Features may depend on :core:navigation, :core:ui, :domain, and relevant
- technical modules such as :player.
-- A temporary :data dependency is allowed only when every imported type is
- recorded in docs/COMPOSE_REDUCTION_PHASE5_TRANSITIONAL_DEPENDENCIES.md.
-- Feature implementations never depend on each other or on :app.
-- AppNavHost remains the root graph registry.
-- Routes, arguments, focus, semantics/test tags, UI output, and callback
- sequence remain unchanged during extraction.
-- Playback recovery, timeshift, and stream policy changes require a separate
- bug-fix scope.
-- Generated baseline/startup profiles are regenerated after package/signature
- moves and are never hand-maintained.
-
-## Measurement gates
-
-For each extracted feature, record:
-
-- five before/after incremental feature-edit build runs;
-- five before/after incremental feature-test compile runs;
-- clean debug and warm no-change guardrails;
-- the exact task set triggered by a feature-only edit;
-- feature and app source/file counts before and after;
-- runtime or functional evidence for the feature's hot flows;
-- device, unavailable-gate, and existing-failure details.
-
-The phase succeeds when a feature edit no longer recompiles unrelated feature
-source, feature tests run independently, and most presentation code has moved
-out of :app without behavior or stability regressions.
-
-## Current plan
-
-- Playback:
- docs/superpowers/plans/2026-08-24-playback-feature-extraction.md
-- Playback execution report:
- docs/COMPOSE_REDUCTION_PHASE5_PLAYBACK_REPORT.md
-- Provider extraction implementation plan:
- docs/superpowers/plans/2026-08-26-provider-feature-extraction.md
-- Provider execution report:
- docs/COMPOSE_REDUCTION_PHASE5_PROVIDER_REPORT.md
-- Settings extraction implementation plan:
- docs/superpowers/plans/2026-08-27-settings-feature-extraction.md
-- Settings execution report:
- docs/COMPOSE_REDUCTION_PHASE5_SETTINGS_REPORT.md
-- Live extraction implementation plan:
- docs/superpowers/plans/2026-08-28-live-feature-extraction.md
-- Live execution report:
- docs/COMPOSE_REDUCTION_PHASE5_LIVE_REPORT.md
-- Catalog extraction design:
- docs/superpowers/specs/2026-09-02-phase-5-catalog-feature-extraction-design.md
-- Catalog extraction implementation plan:
- docs/superpowers/plans/2026-09-02-catalog-feature-extraction.md
-- Catalog execution report:
- docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md
-- System extraction implementation plan:
- docs/superpowers/plans/2026-09-04-system-feature-extraction.md
-- Transitional dependency ledger:
- docs/COMPOSE_REDUCTION_PHASE5_TRANSITIONAL_DEPENDENCIES.md
-
-The settings implementation plan and execution report document the structural
-extraction, resource ownership cleanup, dependency audit, and open runtime
-gates. The Live implementation plan and report now document the Home/EPG
-extraction, translated-resource parity, reviewed feature goldens, connected
-checks, populated MultiView planner journey, and two-channel runtime evidence.
-The exhaustive fixture-dependent journeys and formal paired performance and
-clean-build guardrails remain explicit. Current-checkout clean/warm samples
-and successful Beta/Release packaging are recorded. Profile generation now
-passes on the seeded API 36 TV emulator, with fresh sources containing zero
-stale app Home/EPG descriptors and nonzero `feature/live` descriptors. The
-explicit Macrobenchmark selector now runs all eight journeys: 8/8 passed, 0
-skipped, including both 10-iteration release cold-start methods and six
-seeded interaction journeys. This closes the emulator execution gate; matched
-paired performance and physical-device measurements remain open. The app-route
-Live golden was reviewed and regenerated from the current palette, with its
-connected assertion passing.
-All six feature `testDebugUnitTest` tasks pass. Feature lint now passes for all
-six slices, and `:app:lintDebug` now passes after removing stale app-owned
-feature-resource duplicates and fixing the remaining API/Compose findings. The
-lint baseline now contains only the 374 current findings accepted for separate
-remediation.
-Catalog extraction is structurally implemented with passing connected golden
-checks and current-checkout build-isolation samples. The public-M3U seeded
-Home/Live/Movies/Series/Search smoke pass is supplemented by a diagnostic
-temporary Xtream fixture run covering 63-item VOD/series catalogs, VOD/series
-details, episodes, saved filters, and cross-type Search; the semantic evidence
-is recorded in the Catalog validation directory. The checked-in
-`tools/catalog_xtream_fixture.py` now provides a rerunnable development seed,
-and `tools/catalog_connected_validation.py` provides a semantic ADB journey
-for the fixture-backed Home/Movies/Series/detail/favorite/saved/Search path,
-Movies full-library entry/back, Movies `Load more (60/63)` → `Pagination Movie
-63`, Infinite scroll off/on restoration, and Settings-owned Dashboard shelf
- customization cancel/save/reset flow. Series pagination, browse reorder,
- movie download completion, and direct Favorites test-host rendering/Back
- cancellation are covered. Catalog RTL/large-text semantics and a reduced-
- motion connected run also pass. The same Catalog instrumentation suite
- passes with AOSP TalkBack enabled, but the fixture-driven production
- TalkBack traversal loses semantic D-pad focus in the current ADB harness.
- Cast receiver chooser, touch/device variants, and a complete production
- accessibility-service audit remain open.
-A focused Dashboard macrobenchmark rerun passed 1/1 with five warm iterations
-once public-M3U live-channel activity populated a scrollable Home shelf
-(`frameCount` min/median/max 88/91/98; `frameOverrunMs` P50/P90/P95/P99
-63.1/78.9/89.7/106.1). The earlier empty Home fixture emitted no render-thread
-slices; the rerun is diagnostic after-run evidence, not a cache-equivalent
-paired comparison. The paired comparison remains open; the current app lint
-gate itself passes with the existing baseline.
-The paired Dashboard comparison is intentionally deferred until the
-benchmark environment has stable capacity for matched baseline and current
-runs.
-The release-like profile workflow now passes with feature Catalog descriptors.
-System presentation extraction is structurally complete under `:feature:system`:
-Welcome, Downloads, Plugins, controller-free graph registration, resources,
-contracts, and tests are feature-owned while app adapters and plugin/platform
-runtime remain app-owned. Connected acceptance passes 19/19 and paired
-incremental source/test thresholds and the steady-state clean/warm guardrail
-pass. The supported System profile refresh now passes through the generated
-source promotion workflow. Physical-device/touch validation, accessibility
-service traversal, and authorized production-provider/plugin journeys remain
-open; see `COMPOSE_REDUCTION_PHASE5_SYSTEM_REPORT.md`.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_SETTINGS_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE5_SETTINGS_REPORT.md
deleted file mode 100644
index 1cfbfc29..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_SETTINGS_REPORT.md
+++ /dev/null
@@ -1,989 +0,0 @@
-# Compose Reduction Phase 5 — Settings Feature Report
-
-Date: 2026-08-28
-
-## Current status
-
-The settings slice is **in progress**. A compilable, independently checked
-`:feature:settings` module now owns the Settings and parental presentation,
-backup-preview rendering, settings state/models/actions/observers, preference
-mapping, backup/restore and EPG actions, the Hilt `SettingsViewModel`, and the
-related feature tests. `:app` remains the composition root and continues to own
-the platform adapters, route decoding, and Android lifecycle wiring.
-
-This is not Phase 5 completion. App adapter cleanup, dependency-audit work,
-and runtime/manual acceptance are still open. The Settings/parental route
-registration contract is feature-owned. The existing playback and provider
-runtime gates remain open under their respective reports.
-
-## Ownership and boundary evidence
-
-- `:feature:settings` is registered and consumed by `:app` through the feature
- dependency; the app owns root navigation and platform composition while the
- feature owns the complete Settings presentation entry point.
-- The feature boundary allows only `:core:navigation`, `:core:ui`, `:domain`,
- temporary ledgered `:data`, and the audited `:player` dependency.
-- The feature source has no `com.streamvault.app`, `NavController`,
- `NavHostController`, `MainActivity`, `:feature:provider`, or
- `:feature:playback` references.
-- Feature-defined ports cover backup/restore, diagnostics, TV-surface refresh,
- sync, pairing, update checks/downloads, and external settings actions. App adapters remain
- responsible for Android intents, workers, installers, and other platform
- operations.
-- `ProviderBackupPreviewRequest` and `ProviderBackupPreviewContent` remain
- source-compatible; the provider flow can request the settings-owned preview
- without a provider implementation dependency.
-- The nine direct settings `:data` implementation imports are recorded in
- `docs/COMPOSE_REDUCTION_PHASE5_TRANSITIONAL_DEPENDENCIES.md`; the separate
- `AudioCompatibilityMemoryStore` audit is recorded there as a non-`:data`
- dependency.
-- The 2026-08-28 audit enumerated exact Settings source sites for all nine
- `:data` types and the separate `AudioCompatibilityMemoryStore` dependency.
- None has a complete domain-facing replacement, so no dependency was removed
- in this slice; exact locations and replacement readiness are recorded in the
- transitional dependency ledger.
-- Graphify was refreshed after the locale and test ownership moves; the current
- corpus has 15,325 nodes, 29,893 edges, and 378 communities. `SettingsViewModel` remains
- a high-connectivity coordination node while moved presentation nodes now
- resolve under `feature/settings`.
-- The moved shared widget's default overview values match the app catalog
- exactly. The feature now carries all 24 translated locale catalogs for its
- 681-string/five-plural resource set. Existing app translations are copied
- verbatim where present and default English values preserve the app fallback
- for untranslated keys; the app now retains only the 40 shared/app-consumed
- names, while the 646 confirmed feature-only candidates were removed.
-
-## Delivered commits
-
-- `e12f816f` — inventory and baseline evidence
-- `1c0e6ae6` — settings feature module boundary
-- `422c2810` — shared settings presentation contracts
-- `e9a44e24` — settings platform service ports
-- `58da5a74` — settings-owned backup preview rendering
-- `0441d1e6` — parental controls extraction
-- `20a3f3f5` — settings state and actions extraction
-- `3090ad75` — feature lint fixes (network permission and Compose resource lookup)
-- `3a9dda13` — settings ViewModel and operational observer/action ownership
-- `296bb639` — shared settings widgets, EPG source/assignment presentation,
- dialog state, and the parental action model
-- `4c814fe1` — selection dialog primitives, timeout dialog/formatter, and
- top-navigation dialog presentation
-- `39cf0532` ג€” moved Live TV/parental presentation into the feature source set
-- `55642c6a` ג€” feature-owned Live TV/parental strings and enum label formatters
-- `b00eb575` ג€” moved the sync progress overlay and its feature-owned strings
-- `3a1e264a` ג€” moved the app-update label and action-state formatters
-- `19ffbe42` ג€” moved the Internet Speed Test card and feature-owned labels
-- `5a9a5bce` ג€” moved provider sync dialogs and their feature-owned labels
-- `25a04b89` ג€” moved provider selector, catalog-stat, and status-badge widgets
-- `ef1d965d` ג€” moved parental PIN/protection-level dialog orchestration
-- `81135ed5` ג€” moved quality-cap/category-sort dialogs and label formatters
-- `08524a90` ג€” moved timer, A/V offset, and generic text-value dialogs
-- `3f2c490d` — moved the settings navigation rail and shared navigation item
-- `eaff72a5` — moved Privacy presentation behind explicit action callbacks
-- `8ef92853` — moved shared settings rows and Live TV quick-filter presentation
-- `47553c08` — moved resource-free settings primitive formatters
-- `5d60ea1b` — moved mode-selection dialogs and enum/remote label formatters
-- `32e716a6` — moved resource-free language option and locale label formatters
-- `0e496921` — moved subtitle/player option models and label formatters
-- `7692681c` — moved backup selection/management dialogs into the feature
-- `0e309e15` — made backup dialog models and composables public at the boundary
-- `d5ea4004` — moved player preference dialog orchestration into the feature
-- `5aa18484` — made player preference dialog entry point public at the boundary
-- `25c22123` — moved external playback mode dialog presentation into the feature
-- `431a543b` — removed the app-side update-model typealias and updated its test
-- `014e4c84` — moved provider action, warning, and M3U classification cards
-- `d176dbc1` — moved settings preference-dialog orchestration and labels
-- `40cd49df` — moved the settings playback section into the feature
-- `0eb55acf` — moved live-stream-format label formatting to the feature
-- `854fffde` — moved browsing/navigation preference section presentation
-- `3e26db98` — moved combined-M3U profile card and dialog presentation
-- `20244227` — moved provider-management dialog orchestration
-- `dfb4449f` — moved provider settings section, provider card, diagnostics panel,
- catalog helpers, and feature-local time formatting
-- `b67ef1ae` — moved recording settings dashboard/actions cards and shared
- recording presentation primitives
-- `34cf974f` — moved recording metric/status display primitives and retained
- the app-owned time-window adapter for browser orchestration
-- `5407ced4` — moved recording browser detail metrics/actions and the shared
- compact recording action chip
-- `0d8b6046` — moved recording browser display/item helpers and playback URL
- normalization into the settings feature
-- `4d06b695` — moved speed-test summary/value formatting and labels into the
- settings feature, removing the duplicate app formatter file
-- `92333570` — moved recording pattern, retention, concurrency, and padding
- dialogs into the settings feature
-- `32f40371` — moved recording-browser sidebar controls behind an explicit
- television-device capability input
-- `80af3673` — split the recording dashboard/actions section from its
- app-owned browser launcher/navigation adapter
-- `9b72c593` — moved the recording browser dialog renderer into the settings
- feature with explicit time-format and television-device inputs
-- `0f52133d` — moved backup/restore, Drive-backup, and About/update/crash
- presentation into the settings feature while keeping app version/build and
- external-URI adapters in `:app`
-- `27677327` — moved settings dialog orchestration, restore-sync and crash
- report dialogs, and clear-history presentation into the feature
-- `8f47d919` — moved the settings overlay host into the feature and injected
- the app-owned recording-browser launcher as a composable callback
-- `7975802c` — moved settings screen label/state composition and its resource
- formatters into the feature; app build verification is mapped to the neutral
- feature status type
-- `5fe8a42b` — moved the settings content pane renderer into the feature and
- kept the app version label as an explicit composition-root input
-- `ae832ee4` — moved locale-aware backup timestamp formatting into the feature
- while retaining the app-specific `BackupFileBridge` candidate adapter
-- `579a00cc` — moved Settings and parental-control route registration into a
- feature-owned graph contract; `:app` retains route codec and screen callbacks
-- `fba66992` — added feature route-pattern coverage for Settings `backupUri`
- and parental `providerId` compatibility
-
-- `075452c6` — moved the Settings root screen and recording-browser
- orchestration into the feature, wired the app platform host and shell
- destinations, and added Task 8 extraction evidence
-
-Recent continuation commits also completed the locale ownership cleanup and
-dependency audit snapshot (`0d296edc`, `95abe87e`, `222e95e8`).
-
-The design/spec and detailed implementation plan are tracked documentation for
-the ongoing slice:
-
-Locale-resource extraction evidence is recorded in
-`validation/phase5_settings/task8-locale-resources.md`.
-
-- `docs/superpowers/specs/2026-08-27-phase-5-settings-feature-extraction-design.md`
-- `docs/superpowers/plans/2026-08-27-settings-feature-extraction.md`
-
-## Automated verification
-
-The following command completed successfully with exit code 0 after the lint
-fix:
-
-```text
-gradlew.bat :feature:settings:check --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 31s
-```
-
-The focused cross-module command also completed successfully with exit code 0:
-
-```text
-gradlew.bat :feature:settings:verifyFeatureSettingsBoundary \
- :feature:settings:testDebugUnitTest \
- :app:compileDebugKotlin \
- :app:compileDebugUnitTestKotlin \
- --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 5s
-```
-
-These checks cover the fail-closed source boundary, feature debug unit tests,
-app Kotlin compilation, and app unit-test compilation. No runtime or connected
-device claim is implied by these commands.
-
-The shared/EPG presentation batch was rechecked after the move with:
-
-```text
-gradlew.bat :feature:settings:check --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 8s
-
-gradlew.bat :app:compileDebugUnitTestKotlin --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 13s
-```
-
-The app compile initially found the expected cross-module seam (moved symbols
-were still internal or lacked feature imports); those integration points were
-made explicit and the subsequent feature check and app unit-test compile both
-passed. The settings feature still has no app, navigation-controller, or
-activity references.
-
-The follow-up dialog batch also passed:
-
-```text
-gradlew.bat :feature:settings:check --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 54s
-
-gradlew.bat :app:compileDebugKotlin --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 12s
-
-gradlew.bat :app:compileDebugUnitTestKotlin --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 5s
-```
-
-The Live TV/parental presentation batch was then checked as one focused build:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 44s
-```
-
-This run included the feature boundary task, feature debug/release unit tests,
-lint, app Kotlin compilation, and app unit-test compilation. The feature source
-still has no app, navigation-controller, or activity references.
-
-The sync-overlay batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 35s
-```
-
-The check preserved the overlay's cancel-focus behavior and confirmed the
-feature boundary and app integration remain clean.
-
-The app-update formatter batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 25s
-```
-
-The pure formatters now resolve feature resources while the app update adapter
-and launch/install callbacks remain in the composition root.
-
-The Internet Speed Test card batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 29s
-```
-
-The card remains callback-driven; speed-test execution and persistence stay
-outside the presentation module.
-
-The provider-sync dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 22s
-```
-
-The provider type branching, custom selection state, D-pad/click behavior, and
-callbacks remain unchanged; only their presentation ownership moved.
-
-The provider-widget batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 24s
-```
-
-Feature-owned status/catalog labels now back the shared widgets while the
-provider card's orchestration remains in `:app`.
-
-The parental-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 20s
-```
-
-PIN verification, pending protection-level handling, and dismissal callbacks
-remain behavior-compatible; only dialog presentation ownership moved.
-
-The quality-cap/category-sort dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 10s
-```
-
-Selection ordering, selected-state rendering, category descriptions, and
-quality-cap callbacks remain unchanged.
-
-The timer/value-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 15s
-```
-
-Timer preset ordering, offset clamping/reset/save behavior, and trimmed text
-callbacks remain unchanged.
-
-The navigation rail batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 16s
-```
-
-Navigation category order, selected-state styling, focus requester placement,
-and category callbacks remain unchanged.
-
-The Privacy presentation batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 53s
-```
-
-Parental-control dialog state, incognito and Xtream compatibility toggles, and
-clear-history dialog callbacks remain unchanged.
-
-The shared-row and quick-filter presentation batch was verified with the
-focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 18s
-```
-
-Row focus/click behavior and quick-filter add/remove/empty-state behavior remain
-unchanged.
-
-The resource-free formatter batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 4s
-```
-
-Byte/timestamp formatting, playback-speed labels, timeout preset ordering, and
-recording presentation call sites remain unchanged.
-
-The mode-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 3m 12s
-```
-
-Mode option ordering, selected-state rendering, remote shortcut option labels,
-and dismissal/selection callbacks remain unchanged.
-
-The language-formatting batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 31s
-```
-
-Supported-language ordering, system/auto fallback labels, and locale display
-capitalization remain unchanged.
-
-The subtitle/player-formatting batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 41s
-```
-
-Subtitle size/color option ordering, decoder/audio-output/surface labels, and
-fallback labels remain unchanged while these resource-backed formatters now
-resolve feature-owned strings.
-
-The backup-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 2m 14s
-```
-
-Backup item ordering, initial-body focus, D-pad/click behavior, delete
-confirmation, busy-state disabling, and dismissal callbacks remain unchanged;
-candidate discovery and deletion persistence stay in the app composition root.
-
-The player-preference-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 38s
-```
-
-Playback speed, time format, decoder/output/surface, buffer/protocol,
-timeshift, timeout, timer, subtitle, and translation dialog ordering and
-selection callbacks remain unchanged; mutations continue through the existing
-settings ViewModel.
-
-The external-playback-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 45s
-```
-
-Internal/external option ordering, selected-state handling for the legacy
-ask-every-time value, focus/click behavior, and dismissal callbacks remain
-unchanged.
-
-The app-update-model ownership cleanup was verified with:
-
-```text
-gradlew.bat :feature:settings:check :app:testDebugUnitTest --tests "com.streamvault.app.ui.screens.settings.SettingsAppUpdateModelsTest" --no-daemon
-BUILD SUCCESSFUL in 54s
-```
-
-The app test now imports `AppUpdateUiModel` directly from the settings feature;
-the feature remains the sole production owner of that presentation model.
-
-The provider-action-card batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 53s
-```
-
-Provider connect/sync/edit/delete/category-control actions, warning filtering
-and retry actions, M3U classification toggle/refresh behavior, focus styling,
-and busy-state disabling remain unchanged.
-
-The preference-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 2m 19s
-```
-
-Landing-screen availability, guide-category selection, language/audio/subtitle
-options, quality-cap selection, remote shortcut mapping, category-sort state,
-and all player-dialog delegation remain unchanged; state mutations continue
-through `SettingsViewModel`.
-
-The playback-section batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 59s
-```
-
-Playback/live rows, toggle state, live-format option ordering, focus/click
-behavior, and dialog-open callbacks remain unchanged.
-
-The browsing-section batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 57s
-```
-
-Browsing toggle persistence, category-sort and mode row ordering, remote
-shortcut interactions, and all dialog-open callbacks remain unchanged.
-
-The combined-M3U presentation batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 2m 16s
-```
-
-Combined-profile selection, activation, rename/create/member dialogs,
-provider ordering/toggle/remove behavior, selected-state styling, and callback
-sequencing remain unchanged.
-
-The provider-management-dialog batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 2m 8s
-```
-
-Combined-profile create/rename/member flows, provider sync/custom-sync
-selection, delete confirmation/progress, busy-state disabling, and dialog-state
-reset behavior remain unchanged.
-
-The provider-presentation batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 31s
-```
-
-Provider selection/empty states, expiration and onboarding messaging, catalog
-diagnostic counts, diagnostics/database-health panels, M3U controls, warning
-actions, and provider callbacks remain unchanged. `AppTimeFormat` is passed
-through the feature boundary and uses feature-local equivalents of the former
-date/time adapters; no app package dependency was introduced.
-
-The recording-card batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 55s
-```
-
-Recording storage/status cards, action-button focus styling, Wi-Fi-only toggle
-callbacks, output-path summarization, recording source/failure labels, and
-browser/detail call sites remain behavior-compatible; recording launchers and
-persistence remain app-owned.
-
-The recording-display-primitive batch was verified with focused compilation and
-the settings feature check:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 42s
-```
-
-Metric cards, recording title/subtitle fallback, status labels, status colors,
-and browser/sidebar/detail call sites remain unchanged. The browser time-window
-line now uses the feature's date/time formatter with `AppTimeFormat` supplied by
-the app composition layer.
-
-The recording-browser-detail batch was verified with:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 16s
-```
-
-Detail metric cards, play/stop/cancel/skip/delete/retry actions, schedule-toggle
-behavior, compact-chip focus styling, and browser/sidebar call sites remain
-unchanged; recording dialog orchestration and time-window formatting remain in
-the app composition root.
-
-The recording-browser display/helper batch was then verified with:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 33s
-```
-
-The standalone recording item card, playback-URL normalization helper, and
-browser secondary-line formatter now live in `:feature:settings`; the app still
-owns the dialog orchestration, platform time provider, playback launcher, and
-recording persistence callbacks.
-
-The speed-test-formatting batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 39s
-```
-
-Speed-test value/transport/summary formatting now resolves feature resources;
-the app state builder continues to supply its existing context and date-time
-format, while test execution and persistence remain outside presentation.
-
-The recording-settings-dialog batch was covered by the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 31s
-```
-
-Pattern, retention, concurrency, and padding option ordering plus dismissal and
-ViewModel callbacks remain unchanged; the app retains the surrounding screen
-dialog state and recording persistence orchestration.
-
-The recording-browser-sidebar batch was covered by the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 1m 28s
-```
-
-Search filtering, status-chip toggling, D-pad focus, TV keyboard gating, and
-selection callbacks remain unchanged; only the device capability lookup stays
-in the app composition layer.
-
-The recording-section split was verified by the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 44s
-```
-
-The feature now owns the recording dashboard/actions list; the app retains only
-the browser launcher adapter that decodes the current route and invokes the
-player callback.
-
-The recording-browser-renderer batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon
-BUILD SUCCESSFUL in 47s
-```
-
-Dialog rendering, empty-state behavior, picker/detail focus handling, status
-filters, time-window formatting, and recording callbacks remain unchanged; the
-app adapter supplies only platform time/device values and player navigation.
-
-The backup/About presentation batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 34s
-```
-
-Backup export/import, local and USB backup actions, Drive sign-in/push/pull and
-snapshot labels, About/update/crash-report rows, and their focus/callback
-behavior now render from `:feature:settings`. The app supplies the version/build
-label and continues to own URI launching, backup orchestration, and persistence;
-the feature boundary and app Kotlin/unit-test compilation both remain clean.
-
-The settings-dialog orchestration batch was verified with the same focused
-command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 38s
-```
-
-Dialog ordering, restore-sync selection behavior, crash-report scrolling and
-focus handling, clear-history callbacks, and sync-overlay cancellation remain
-unchanged; only the presentation/resource ownership moved to the feature.
-
-The settings-overlay batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 33s
-```
-
-Snackbar, recording-browser visibility, and dialog-overlay ordering remain
-unchanged. The feature now hosts the overlay composition while the app retains
-the recording playback/navigation adapter through the injected callback.
-
-The settings-label-state batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 37s
-```
-
-Settings label derivation, time-format handling, build-verification mapping,
-plural summaries, and existing formatter test expectations remain unchanged;
-the app retains only its platform `MainActivity` lookup and build-verifier
-adapter.
-
-The settings-content-pane batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 34s
-```
-
-Category routing, section ordering, row/dialog callbacks, and feature-owned
-labels remain unchanged; the app now supplies only the version label alongside
-its existing platform and navigation callbacks.
-
-The backup timestamp formatter batch was verified with the same focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 34s
-```
-
-Local, folder, USB, and Drive backup detail timestamps retain the existing
-locale-aware short date/time behavior; only the generic formatter moved into
-the feature.
-
-The feature-route registration batch was verified with the focused command:
-
-```text
-gradlew.bat :feature:settings:check :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 33s
-```
-
-Settings `backupUri` and parental `providerId` arguments retain their existing
-route patterns and defaults; the app continues to decode navigation routes and
-supplies the screen callbacks, while the feature owns destination registration.
-
-The route-pattern compatibility test was verified with:
-
-```text
-gradlew.bat :feature:settings:testDebugUnitTest --tests "com.streamvault.feature.settings.navigation.SettingsRoutePatternsTest" :app:compileDebugKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 29s
-```
-
-The root-screen extraction batch was verified with:
-
-```text
-gradlew.bat :feature:settings:verifyFeatureSettingsBoundary :feature:settings:testDebugUnitTest :feature:settings:lintDebug :feature:settings:compileDebugAndroidTestKotlin :app:compileDebugKotlin :app:compileDebugUnitTestKotlin --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 58s
-```
-
-`SettingsScreen` and the recording-browser orchestration now compile in the
-feature. The app supplies only the `SettingsPlatformHost`, navigation
-destinations, provider callbacks, and close-app callback. Detailed evidence is
-in `validation/phase5_settings/task8-screen-extraction.md`.
-
-Five remaining presentation-model unit tests were also moved into the feature
-test source set. The app-owned update-model test remains in `:app` because it
-exercises app update implementation types.
-
-The final focused verification bundle was re-run after locale consolidation and
-app resource cleanup:
-
-```text
-gradlew.bat :feature:settings:verifyFeatureSettingsBoundary \
- :feature:settings:testDebugUnitTest \
- :feature:settings:lintDebug \
- :feature:settings:compileDebugAndroidTestKotlin \
- :app:compileDebugKotlin \
- :app:testDebugUnitTest \
- --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 25s
-```
-
-This confirms the fail-closed boundary, feature unit tests/lint, connected-test
-source compilation, and app compile/unit-test integration. At the time of this
-earlier run no connected device was available (`adb devices` returned an empty
-device list); the later emulator-backed settings run is recorded below. Manual
-acceptance remains open.
-
-The deterministic unit suites were subsequently rerun fresh. The complete
-`:feature:settings:testDebugUnitTest` suite passed with 34 tests and no
-failures/errors. The app-owned settings seams were run explicitly via
-`AppSettingsAdaptersTest`, `BackupFileBridgeTest`, and
-`SettingsAppUpdateModelsTest`; all 16 tests passed. These results validate the
-feature policy and composition-root adapters but do not replace manual TV
-journeys.
-
-The profile-source check plus beta/release assembly was then re-run together:
-
-```text
-gradlew.bat verifyBaselineProfileSources :app:assembleBeta :app:assembleRelease \
- --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 6m 17s
-```
-
-Release-like profile merge tasks now explicitly depend on the normalized
-profile installer, avoiding Gradle's implicit-output validation failure when
-verification and assembly share one invocation. The dependency is disabled
-only when `generateBaselineProfile` is requested, because the producer must
-package the non-minified release before it copies its newly collected profile.
-
-The seeded-target profile producer was then run on the API 36 TV emulator:
-
-```text
-gradlew.bat :app:generateBaselineProfile --no-daemon \
- --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 12m 37s
-10 profile tests passed; 8 ordinary macrobenchmarks skipped by selector
-```
-
-It generated and installed `baseline-prof.txt` (46,249 rules) and
-`startup-prof.txt` (30,093 rules). The generated source scan has zero
-pre-extraction `com/streamvault/app/ui/screens/settings` descriptors and
-includes feature/settings descriptors. Full evidence is in
-`validation/phase5_settings/profile-validation.md`.
-
-The final neighboring-feature/debug assembly gate was also attempted:
-
-```text
-gradlew.bat :feature:settings:check :feature:provider:check \
- :feature:playback:check :app:testDebugUnitTest :app:assembleDebug \
- --no-daemon --console=plain --warning-mode=none
-BUILD FAILED at :feature:playback:lintDebug
-```
-
-Settings and provider checks had progressed successfully. At the time of this
-neighboring check, the failure was an existing playback lint error in
-`feature/playback/src/androidTest/.../PlayerOverlayGoldenTest.kt:51`
-(`RememberInComposition` for `FocusRequester()`). Playback remained outside the
-Settings slice and was not modified by that Settings change; the lint failure
-was subsequently resolved in the Phase 5 closeout.
-
-With the Android SDK at `E:\androidSdk` and a `Television_1080p(AVD)` API 36
-emulator available, the focused settings connected suite was executed:
-
-```text
-gradlew.bat :feature:settings:connectedDebugAndroidTest \
- --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 11s
-2 tests, 0 failures, 0 errors, 0 skipped
-```
-
-The result XML is
-`feature/settings/build/outputs/androidTest-results/connected/debug/TEST-Television_1080p(AVD) - 16-_feature_settings-.xml`.
-It covers the feature-owned Settings entry point and the provider backup-preview
-title contract. Full manual Settings acceptance remains open.
-
-The neighboring provider and app connected checks were also attempted on the
-same emulator. Provider passed its single test. The app suite ran 26 tests and
-reported five existing failures (download quota timeout lease, Watch Next
-round-trip schema, and three PlayerSmoke focus/track cases). These failures are
-outside `:feature:settings`; they remain open under the neighboring acceptance
-gates and are not being counted as settings failures.
-
-## Expanded behavior-focused connected evidence (2026-08-28)
-
-The Settings connected suite now includes ten new behavior-focused cases plus
-the two existing feature cases. It passed on the API 36
-`Television_1080p(AVD)` emulator with `E:\androidSdk`:
-
-```text
-gradlew.bat :feature:settings:connectedDebugAndroidTest \
- --no-daemon --console=plain --warning-mode=none
-BUILD SUCCESSFUL in 1m 43s
-12 tests, 0 failures, 0 errors, 0 skipped
-```
-
-The result XML is
-`feature/settings/build/outputs/androidTest-results/connected/debug/TEST-Television_1080p(AVD) - 16-_feature_settings-.xml`
-(`tests=12`, `failures=0`, `errors=0`, `skipped=0`, `time=62.167`). Coverage
-now includes section navigation, D-pad focus restoration, dialog open/dismiss
-and Back behavior, parental/PIN state, backup/restore callbacks and state,
-route argument compatibility, accessibility semantics, and RTL navigation.
-The shared `PremiumDialog` host also registers a guarded activity Back callback
-so Back dismisses a Settings dialog instead of finishing the host activity.
-
-Exact failures in the accepted run: none. The captured artifacts are:
-
-The connected suite was rerun after the workstation Gradle path change. A
-first attempt hit the emulator's boot/service race (`emulator-5554` was not
-visible to ADB and installation reported `cmd: Can't find service: package`).
-After the existing emulator reported boot completion and the package service
-was available, the same command passed without clearing or uninstalling app
-data:
-
-```text
-BUILD SUCCESSFUL in 58s
-12 tests, 0 failures, 0 errors, 0 skipped
-```
-
-The refreshed XML at the same path reports `tests=12`, `failures=0`,
-`errors=0`, `skipped=0`, `time=23.957`, timestamp
-`2026-08-28T17:45:10`. The earlier XML timing remains historical evidence;
-the latest rerun is the current connected gate result.
-
-- `validation/phase5_settings/task11_screenshots/settings_navigation_dpad.png`
-- `validation/phase5_settings/task11_screenshots/settings_navigation_rtl.png`
-- `validation/phase5_settings/task11_screenshots/settings_backup_preview.png`
-
-Full runtime evidence, dimensions, hashes, and the exact XML path are in
-`validation/phase5_settings/task11-runtime-validation.md`.
-
-## Production-shell manual journey evidence (2026-08-28)
-
-Safe TV journeys were run on the API 36 `AOSP_TV_on_x86` emulator
-(`emulator-5554`) after installing `:app:installDebug`. The app was launched
-through `com.streamvault.app.debug/com.streamvault.app.MainActivity` and its
-existing data was preserved. Settings entry, section navigation, Playback,
-Privacy, Recording, Backup & Restore, EPG Sources, and About all rendered in
-the production shell. The protection-level dialog opened and platform Back
-dismissed it while leaving `MainActivity` foregrounded. Enter New PIN opened
-without entering a value. Manage local backups and Import Data followed their
-no-fixture safe paths; no restore, external account, credential, USB/Drive, or
-destructive operation was used. EPG Sources correctly reported no configured
-external sources.
-
-The first About → Check now action found a real app-owned seam defect:
-
-```text
-08-28 14:15:31.963 FATAL EXCEPTION: main
-java.lang.StackOverflowError: stack size 8188KB
-at com.streamvault.app.settings.AppSettingsUpdateAdapter.isRemoteVersionNewer(AppSettingsUpdateAdapter.kt:54)
-(the adapter frame repeated recursively)
-```
-
-The adapter's unqualified policy call resolved to its own override. The minimal
-fix aliases the top-level policy import in
-`app/src/main/java/com/streamvault/app/settings/AppSettingsUpdateAdapter.kt`.
-`app/src/test/java/com/streamvault/app/settings/AppSettingsUpdateAdapterTest.kt`
-reproduces the delegation contract and passed after the fix. The patched APK
-was reinstalled without clearing data; the same D-pad route then kept the
-activity alive, advanced `Last checked` to `Aug 28, 2026 2:29 PM`, and left
-`Update status` as `Up to date`. The screen still shows the saved pre-fix crash
-report, which is expected with preserved data. Screenshots and the detailed
-failure/log evidence are recorded in
-`validation/phase5_settings/task11-runtime-validation.md`, including
-`manual_journeys/update-check-post-fix.png`.
-
-The seeded public `iptv-org US validation` provider was synced without
-credentials. The M3U request returned HTTP 200 with a 327,094-byte body and
-1,467 Live channels were staged. The final UI status was `ERROR` only because
-the fixture had no movie entries and no EPG URL:
-
-```text
-Section retry failed ... [MOVIES]: Playlist contains no movie entries
-Section retry failed ... [EPG]: No EPG URL configured for this provider
-```
-
-This is recorded as provider fixture evidence, not attributed to the Settings
-extraction. A focus attempt for Provider Diagnostics opened Add a provider;
-no credentials were entered. Recording storage was inspected without starting
-a recording or player handoff.
-
-The remaining safe TV pass then exercised the Settings shell again from a
-fresh direct launch. Providers, parental controls, and Backup & Restore were
-entered with the D-pad. Opening parental controls showed the existing
-`Unlocked` category state without changing it; Back returned to the Providers
-section with the Settings rail focus restored. The Backup & Restore section
-opened its export and local-backup paths. On this TV the Export Data action
-used the picker-free Downloads path and reported:
-
-```text
-Backup saved locally. On supported TVs it is in Downloads/StreamVault; open it with a file manager.
-```
-
-Manage local backups then listed the generated files without selecting Delete
-or Restore:
-
-```text
-streamvault_backup_20260828_150353_675.json
-streamvault_backup_20260828_150322_143.json
-```
-
-The paired screenshots are under
-`validation/phase5_settings/manual_journeys/`: `remaining-settings-screen.png`,
-`remaining-parental-screen.png`, `remaining-settings-return.png`,
-`remaining-backup-screen.png`, `remaining-export-result.png`, and
-`remaining-local-backups.png`. They are 1920x1080 captures from
-`com.streamvault.app.debug/com.streamvault.app.MainActivity`; SHA-256 hashes
-are recorded in the validation report. No picker location, external account,
-credential, USB/Drive flow, import, restore, or delete action was used.
-
-## Incremental build performance evidence
-
-The required five-pair measurements were captured in isolated detached
-snapshots at pre-extraction ref `e12f816f` and post-extraction ref `cf468d3e`,
-using the same repository wrapper, `E:\androidSdk`, dedicated E: Gradle/temp
-paths, and command flags. Full raw samples and every profile path are in
-`validation/phase5_settings/performance-samples-2026-08-28.md`.
-
-The source-edit commands were `:app:compileDebugKotlin` before extraction and
-`:feature:settings:compileDebugKotlin` after extraction. The five-sample median
-was `16926 ms` before versus `8858 ms` after (47.7% faster), but a one-time
-feature/KSP setup outlier made the five-sample averages `19032 ms` versus
-`19432 ms` and the p95 `34946 ms` versus `60017 ms`. Excluding that first setup
-sample, the four-sample averages were `15053 ms` versus `9286 ms` (38.3%
-faster).
-
-The moved-test commands were `:app:compileDebugUnitTestKotlin` before and
-`:feature:settings:compileDebugUnitTestKotlin` after. Five-sample averages were
-`22852 ms` versus `14915 ms` (34.7% faster), while medians were `11324 ms`
-versus `9889 ms` (12.7% faster) and p95 values were `60069 ms` versus
-`35141 ms` (41.5% faster). These mixed cache/setup states do not support a
-formal claim that the 20% steady-state test target is closed.
-
-The clean guardrail passed in both snapshots: pre `9m27s` (154 actionable;
-123 executed; 31 cached) and post `4m55s` (176 actionable; 119 executed; 57
-cached). Warm no-change runs also passed: pre `10s` and post `12s`, each with
-one executed task and the remainder up to date. The Phase 0 clean reference
-(`2m15.47s`) was recorded under a different cache/storage state, so it is not
-a comparable clean baseline; the observed post clean value is above it. The
-warm post value is below the Phase 0 warm reference (`36.264s`). Accordingly,
-the Settings performance gate remains open pending a comparable clean baseline
-and steadier test-compile comparison; no target closure is claimed.
-
-## Open work and gates
-
-- The resource-usage audit and confirmed-obsolete app locale/default cleanup
- are complete: 646 feature-only candidate entries were removed while 40
- shared/app-consumed names remain. Detailed evidence is in
- `validation/phase5_settings/task10-resource-cleanup.md`.
-- Keep the remaining app-owned update-model test and shared
- navigation/dashboard/platform-label resources in `:app` until their future
- owners are explicit; the feature-only settings resource cleanup is complete.
-- Remove any remaining app-side settings compatibility shims only after their
- platform responsibilities have domain-facing contracts and the app root no
- longer needs them.
-- Complete the DAO/concrete dependency audit and remove each ledger entry only
- after a domain-facing replacement exists.
-- The five paired incremental source/test samples and clean/warm guardrails are
- recorded in `validation/phase5_settings/performance-samples-2026-08-28.md`.
- Steady-state source compilation improved, but setup outliers and the
- non-comparable Phase 0 clean reference keep the formal performance gate open.
-- Safe manual TV journeys for Settings navigation, parental PIN dialogs,
- backup export/local listing and no-fixture import, EPG Sources, update
- checking, and provider sync have evidence. Fixture-dependent restore,
- diagnostics action routing, credentialed provider flows, recording/player
- handoff, and the remaining full-shell accessibility/reduced-motion coverage
- remain open.
-- The focused connected settings result, plus the neighboring provider/app
- result and failure inventory, are recorded in
- `validation/phase5_settings/task11-runtime-validation.md`. The safe
- Settings shell journey is evidenced; fixture-dependent and full-shell gates
- remain open until suitable fixtures and time are available.
-- Do not treat the still-open playback/provider acceptance and performance
- gates as closed by this settings report.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_SYSTEM_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE5_SYSTEM_REPORT.md
deleted file mode 100644
index 757d8f14..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_SYSTEM_REPORT.md
+++ /dev/null
@@ -1,249 +0,0 @@
-# Compose Reduction Phase 5 — System feature extraction report
-
-Date: 2026-09-05
-
-## Outcome
-
-Welcome, Downloads, and Plugins presentation is now owned by `:feature:system`.
-The feature owns its controller-free graph registration, routes, ViewModels,
-rendering, neutral plugin contracts/models, resources, unit tests, connected
-tests, and reviewed goldens. `:app` remains the Android composition root and
-retains startup orchestration, shell composition, platform adapters, plugin
-discovery/IPC/work/provider/playback infrastructure, and Android entry points.
-
-The extraction is structurally complete and the incremental feature-value
-thresholds pass. The clean/warm build-overhead guardrail also passes in
-steady-state. The supported seeded baseline/startup profile workflow now
-completes and the generated sources have been promoted through Gradle. The
-module remains provisional only for the explicitly unavailable physical-device,
-accessibility-service, and authorized production-journey gates.
-
-## Rollback and commit sequence
-
-Rollback SHA: `cddef1526924634819c7fb27d15c8a49b3945a1d`.
-
-The implementation was split into these focused commits:
-
-```text
-e9e89e65 docs(system): record phase 5 extraction baseline
-b189b977 test(system): lock existing presentation behavior
-c9f7b3cd build(system): add isolated feature module
-fac29462 feat(system): define feature contracts and routes
-6ce6517a refactor(system): bridge app startup and plugin services
-541dd966 refactor(system): move welcome presentation
-0c7d911d refactor(system): move downloads presentation
-3ab9239b refactor(system): move plugins presentation
-dcd74f59 refactor(system): register feature-owned navigation
-35ff4f03 test(system): stabilize presentation semantics
-57182dee test(system): verify ownership and resources
-```
-
-Connected acceptance and final documentation are intentionally separate
-follow-up commits after the final evidence review.
-
-## Ownership and size
-
-The Task 0 baseline contained eight app production files and 2,120 physical
-lines: the three screen/state/ViewModel groups plus the two old graph files.
-The authoritative pre-extraction inventory is
-`validation/phase5_system/task0-inventory.md`.
-
-Current feature counts are:
-
-| Area | Files | Physical lines / assets |
-|---|---:|---:|
-| `src/main/java` Kotlin | 13 | 2,459 |
-| `src/test/java` Kotlin | 9 | 700 |
-| boundary fixtures | 12 | 45 |
-| `src/androidTest/java` Kotlin | 7 | 970 |
-| `src/main/res` | 51 | 33 default System keys, 26 qualifiers |
-| reviewed connected goldens | 6 | 1920×1080 PNG each |
-
-The old app screen directories and app `WelcomeGraph.kt`/`SystemGraph.kt` are
-absent. The feature has one `SystemGraph` registration and the app root
-registers it once.
-
-## Boundary and contracts
-
-`:feature:system` has exactly these project dependencies:
-
-```text
-:core:navigation, :core:ui, :domain
-```
-
-The boundary verifier scans production Kotlin/Java and negative fixtures. The
-production result is zero forbidden references for app/data/player/root
-controllers/all sibling features; the fixtures detect app, data, player,
-`MainActivity`, `NavController`, `NavHostController`, and each sibling feature
-token as required. The report is in
-`feature/system/build/reports/feature-system-boundary/report.txt`.
-
-`SystemWelcomePort` exposes `Flow<WelcomeSyncProgress?>` and a
-`WelcomeDevProviderConfig`. `AppSystemWelcomeAdapter` maps the app's
-`SyncProgressBus` and `BuildConfig` values without moving either dependency.
-
-`SystemPluginManagementPort` exposes discovery, provider-source lookup, APK
-install, enable/disable, configuration load/save/refresh, and configuration
-actions. `AppSystemPluginManagementAdapter` delegates directly to the existing
-`StreamVaultPluginManager` and `ProviderSourceRegistry`.
-
-The feature contains no plugin manager, messenger, work coordinator, provider
-ownership, or playback-routing implementation. Those remain under
-`app/src/main/java/com/streamvault/app/plugins` and related app infrastructure.
-
-## Navigation and behavior preservation
-
-The feature owns the exact `welcome`, `downloads`, and `plugins` route
-patterns. `SystemGraph` is controller-free and receives navigation actions,
-startup redirect callbacks, and a scaffold callback from `AppNavHost`. The app
-retains `NavHostController`, `NavHost`, `AppRouteCodec`, startup policy, and
-top-level shell adaptation.
-
-Characterization tests were added before the moves. Feature ViewModel and
-presentation tests preserve loading/error/result handling, download actions,
-plugin configuration JSON/draft behavior, Welcome progress and setup/later
-actions, semantics, focus/test tags, and callback delegation. The connected
-graph test proves one registration for all three routes. Full authorized
-production-provider/plugin journeys were unavailable and are not claimed as
-passed; details are in `validation/phase5_system/task10-device-journeys.md`.
-
-## Resources
-
-The feature owns the exact 33 System string keys from the Task 0 inventory.
-The default and all 25 localized qualifiers were compared (26 total):
-
-```text
-duplicate_system_entries=0
-missing_system_values=0
-value_mismatches=0
-placeholder_mismatches=0
-```
-
-Thirty feature-only app definitions were removed. `app_name`, `nav_downloads`,
-and `settings_cancel` remain app-owned because the manifest, TV input service,
-app shell, Settings, Live, Catalog, and Playback still consume them. The full
-audit is in `validation/phase5_system/task9-resource-cleanup.md`.
-
-## Automated validation
-
-Passing gates:
-
-- `:feature:system:verifyFeatureSystemBoundary`
-- `:feature:system:dependencies`
-- `:feature:system:testDebugUnitTest`
-- `:feature:system:lintDebug`
-- `:feature:system:compileDebugAndroidTestKotlin`
-- `:feature:system:assembleDebug`
-- `:core:navigation:test`
-- `:core:ui:testDebugUnitTest`
-- `:domain:test`
-- `:app:testDebugUnitTest`
-- `:app:assembleDebug`
-- `:app:assembleBeta`
-- `:app:assembleRelease`
-- root `:verifyBaselineProfileSources` (source shape valid)
-
-The final focused matrix completed successfully in 1m29s with 324 actionable
-tasks, including the System boundary, feature lint, connected-test compile,
-feature assembly, app tests, and app debug assembly.
-
-At the System extraction checkpoint, `:app:lintDebug` was red from existing
-app-wide debt (675 errors); the first error was `AppStartupCoordinator.kt:192`,
-`NewApi` on `Trace.beginAsyncSection`. The follow-up lint/resource cleanup below
-records the current repository-wide result.
-
-### Phase 5 lint/resource follow-up (2026-09-05)
-
-The final app resource audit removed 667 stale feature-resource duplicates from
-the app's default and localized string files; feature modules remain the owners
-of those resources. The follow-up also corrected the API-level Trace guard and
-the app shell's Compose resource access and modifier ordering. The full local
-unit/lint sweep passed, including `:app:lintDebug` and all six feature lint and
-unit-test tasks. `:app:lintDebug` reports no new issues; 374 existing findings
-remain filtered by the maintained baseline, with stale historical entries
-removed.
-
-## Connected acceptance and visual review
-
-On the only available device, `Television_1080p(AVD) - 16`, API 36,
-1920×1080:
-
-- six deterministic goldens were recorded, visually inspected, and compared
- pixel-for-pixel;
-- the final System connected suite passed 19/19;
-- the suite covers nine presentation tests, one graph test, six golden cases,
- and three RTL/1.3 font-scale cases;
-- the reduced-motion variant passed with global animation scales set to zero
- and the original scales restored afterward;
-- the retained log evidence reports no app fatal/runtime/resource/launcher
- match and no sensitive-pattern match.
-
-The first 19-test attempt had two transient PixelCopy/window-capture failures;
-the exact same cases passed in isolation. `GoldenCapture.kt` now retries
-capture three times with a 250ms delay, while retaining exact bitmap
-dimensions and pixel comparisons. The rerun passed 19/19.
-
-No phone/tablet profile or TalkBack service was available. Semantics actions
-and connected RTL/large-text/reduced-motion assertions pass, but touch/mouse
-and service-mediated accessibility traversal remain unavailable gates.
-
-The Live-TV long-duration protocol is waived for this slice because no player
-composition, stream preparation, recovery, lifecycle, surfaces, overlays,
-Cast URL rewriting, or plugin playback routing changed.
-
-Evidence: `validation/phase5_system/task10-connected-validation.md`,
-`task10-device-journeys.md`, `task10-screenshots/`, and `task10-logcat.txt`.
-
-## Performance and isolation
-
-The paired warm incremental medians pass the optional-module thresholds:
-
-| Measurement | Before | After | Improvement | Threshold |
-|---|---:|---:|---:|---:|
-| source edit | 20.6311s | 7.3199s | 64.56% faster | ≥25% |
-| unit-test compile | 18.1523s | 7.6457s | 57.88% faster | ≥20% |
-
-All five raw samples per series, cache state, hashes, and build guardrails are
-in `validation/phase5_system/performance-after.md`.
-
-With a reversible System-only edit, the app dry run showed sibling Kotlin
-compile tasks as graph entries marked `SKIPPED`; the actual `:app:assembleDebug`
-run showed sibling Kotlin tasks `UP-TO-DATE` and did not execute sibling Kotlin
-compilation. The System and required app integration/package tasks were the
-only relevant work. Evidence is in
-`validation/phase5_system/system-edit-dry-run.txt` and
-`system-edit-task-output.txt`.
-
-Clean debug, warm debug, Beta, and Release guardrails passed. The paired
-rollback/current clean run is faster in the current checkout, and steady-state
-warm medians are 11.707s rollback versus 11.955s current, a 2.12% slowdown;
-the ≤10% build-overhead gate passes. Full raw samples and cache/task details
-are in `validation/phase5_system/performance-after.md`.
-
-## Profiles, graph, and remaining work
-
-The supported profile command was corrected from the stale plan path
-`:benchmark:pixel2Api36Setup` to the current app-level
-`:app:generateBaselineProfile`. The release benchmark harness now waits for
-UiAutomator idle after launching the target. The startup journey passed 1/1,
-and the seeded critical journey subsequently completed successfully. The
-supported `:app:copyBaselineProfileIntoSrc` workflow then passed its connected
-run with zero failures and refreshed both generated sources. The explicit
-Macrobenchmark selector subsequently completed all 8/8 journeys with zero
-failures and zero skipped tests; both release cold-start methods completed ten
-iterations and all six seeded interaction methods completed five. Each source has
-nonzero `feature/system` descriptors and no old app Welcome/Downloads/Plugins,
-`WelcomeGraph`, or `SystemGraph` descriptors. See
-`validation/phase5_system/profile-validation.md` and
-`validation/phase5_system/macrobenchmark-validation.md` for the command and
-scan/benchmark evidence.
-
-`graphify update .` completed with 16,550 nodes and 32,460 edges. The refreshed
-graph resolves feature `SystemGraph` ownership and an `AppNavHost` to
-`PluginsScreen` path through `AppNavigation`.
-
-Open follow-ups are physical-device/touch validation, TalkBack service
-traversal, and authorized production-provider/plugin journeys. This report does
-not close
-Playback, Provider, Settings, Live, Catalog, accessibility, or repository-wide
-performance gates, and does not claim full Phase 5 modernization completion.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_TRANSITIONAL_DEPENDENCIES.md b/docs/COMPOSE_REDUCTION_PHASE5_TRANSITIONAL_DEPENDENCIES.md
deleted file mode 100644
index c0d4080a..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE5_TRANSITIONAL_DEPENDENCIES.md
+++ /dev/null
@@ -1,267 +0,0 @@
-# Phase 5 Transitional Dependency Ledger
-
-| Feature | Imported implementation | Current consumer | Removal direction | Owner |
-|---|---|---|---|---|
-| playback | data.preferences.PreferencesRepository | PlayerPreferencesCoordinator | domain preference contract | Phase 7 |
-| playback | data.remote.stalker.StalkerPlaybackResolutionException | PlayerContentResolver | player/domain resolution error | Phase 7 |
-| playback | data.remote.xtream.ProviderPlaybackResolver | PlayerContentResolver | domain playback resolver | Phase 7 |
-| playback | data.remote.xtream.XtreamStreamKind | alternate stream support | domain/player stream kind | Phase 7 |
-| playback | data.remote.xtream.XtreamUrlFactory | alternate stream support | domain/player URL factory | Phase 7 |
-| playback | data.security.CredentialDecryptionException | content/provider resolution | domain provider-access error | Phase 7 |
-
-## Task 9 integration status
-
-Task 9 re-scanned the extracted feature after the Player and MultiView moves.
-The six ledger entries remain intentional and are the only direct `:data`
-types imported by feature playback production code:
-
-| Ledger implementation | Current feature files |
-|---|---|
-| `PreferencesRepository` | `MultiViewViewModel`, `PlayerPreferencesCoordinator` |
-| `StalkerPlaybackResolutionException` | `PlayerContentResolver` |
-| `ProviderPlaybackResolver` | `PlayerContentResolver` |
-| `XtreamStreamKind`, `XtreamUrlFactory` | `PlayerAlternateStreamSupport` |
-| `CredentialDecryptionException` | `PlayerContentResolver`, `PlayerProviderCoordinator` |
-
-Accordingly, `:feature:playback` retains its temporary `:data` dependency.
-`:app` also retains `:data`, `:player`, `:domain`, `:core:navigation`,
-`:core:ui`, and `:feature:playback`: its composition root, platform adapters,
-other app features, and `AppNavHost` still consume them. The feature-owned
-`PlaybackGraph` is now registered by `AppNavHost`; the Task 4 graph extraction
-is complete. No project dependency was removed in Task 9 because doing so
-would break an active consumer rather than remove obsolete presentation
-ownership.
-
-The feature is otherwise app-independent: Task 9 source scans found no app
-references to legacy `ui.screens.player` or `ui.screens.multiview` packages,
-and no feature-main references to `com.streamvault.app`, `NavController`, or
-`NavHostController`.
-
-## Provider extraction inventory (Task 1)
-
-The provider slice is not yet moved. The following imports are the planned
-temporary `:data` boundary for `:feature:provider`; each is owned by Phase 7
-and must be removed only after an equivalent domain-facing contract exists.
-
-| Feature | Imported implementation | Current consumer | Removal direction | Owner |
-|---|---|---|---|---|
-| provider | `data.remote.stalker.StalkerParamOverride` | setup form models | domain-owned Stalker setup policy | Phase 7 |
-| provider | `data.remote.stalker.StalkerRequestRule` | setup form models | domain-owned Stalker setup policy | Phase 7 |
-| provider | `data.remote.stalker.StalkerAdvancedOptions` | advanced options model/UI | domain-owned Stalker setup policy | Phase 7 |
-| provider | `data.remote.stalker.StalkerAdvancedOptionsCodec` | provider setup screen | domain-owned Stalker setup policy codec | Phase 7 |
-| provider | `data.remote.stalker.StalkerCompatibilityRegistry` | setup screen/selector | domain-owned Stalker compatibility policy | Phase 7 |
-| provider | `data.util.ProviderInputSanitizer` | provider setup screen | domain-owned provider input validation | Phase 7 |
-| provider | `data.remote.xtream.XtreamAuthenticationException` | setup ViewModel | domain-owned provider-access error | Phase 7 |
-| provider | `data.remote.xtream.XtreamNetworkException` | setup ViewModel | domain-owned provider-access error | Phase 7 |
-| provider | `data.remote.xtream.XtreamParsingException` | setup ViewModel | domain-owned provider-access error | Phase 7 |
-| provider | `data.remote.xtream.XtreamRequestException` | setup ViewModel | domain-owned provider-access error | Phase 7 |
-| provider | `data.remote.xtream.XtreamResponseTooLargeException` | setup ViewModel | domain-owned provider-access error | Phase 7 |
-| provider | `data.security.CredentialDecryptionException` | setup ViewModel | domain-owned provider-access/setup error | Phase 7 |
-
-## Settings extraction inventory (Task 0, 2026-08-27)
-
-The settings slice retains a temporary `:data` dependency while DAO and
-concrete persistence contracts are audited. These imports are presentation
-consumers only; each is a Phase 7 removal candidate after an equivalent
-domain-facing contract exists.
-
-| Feature | Imported implementation | Current consumer | Removal direction | Owner |
-|---|---|---|---|---|
-| settings | `data.local.dao.ProgramDao` | parental controls/settings state | domain-owned program query contract | Phase 7 |
-| settings | `data.local.dao.XtreamIndexJobDao` | diagnostics/settings state | domain-owned indexing status contract | Phase 7 |
-| settings | `data.local.dao.XtreamLiveOnboardingDao` | diagnostics/settings state | domain-owned onboarding status contract | Phase 7 |
-| settings | `data.local.entity.XtreamIndexJobEntity` | diagnostics formatting/state | domain-owned indexing status model | Phase 7 |
-| settings | `data.local.entity.XtreamLiveOnboardingStateEntity` | diagnostics formatting/state | domain-owned onboarding status model | Phase 7 |
-| settings | `data.preferences.DatabaseMaintenanceSnapshot` | database maintenance UI/state | domain-owned maintenance snapshot | Phase 7 |
-| settings | `data.preferences.PreferencesRepository` | settings/preferences state | domain-owned settings preference contract | Phase 7 |
-| settings | `data.sync.ProviderSyncCommands` | sync actions/state | domain-owned provider sync command contract | Phase 7 |
-| settings | `data.sync.SyncRepairSection` | sync repair UI/state | domain-owned sync repair model | Phase 7 |
-
-`player.AudioCompatibilityMemoryStore` is a separate concrete dependency
-audit candidate. It is not included in the nine-entry `:data` ledger above and
-must not become a feature-to-feature implementation dependency.
-
-Settings dependency audit snapshot (2026-08-28): the feature production source
-still imports exactly the nine ledgered `:data` types listed above, plus
-`player.AudioCompatibilityMemoryStore`. The imports are confined to the
-Settings/parental state, observer, action, and formatting orchestration; no
-additional DAO or concrete implementation types were found. Replacing them
-requires domain-facing contracts and remains a Phase 7 task.
-
-### Exact source sites and contract readiness (2026-08-28)
-
-The second audit enumerated the concrete references rather than counting only
-distinct type names:
-
-| Implementation | Exact Settings source sites | Use classification | Replacement readiness |
-|---|---|---|---|
-| `ProgramDao` | `SettingsDerivedStateObservers.kt:36,50`; `SettingsObserverRegistrations.kt:5,161,177`; `SettingsViewModel.kt:16,106` | Provider program-count observation for diagnostics and parental/settings state wiring | No domain query port exists; keep until a provider-scoped program-count contract is introduced |
-| `XtreamIndexJobDao` | `SettingsViewModel.kt:17,115,311` | Xtream indexing-job observation | No domain indexing-status port exists; keep until the job state is modeled in `domain` |
-| `XtreamLiveOnboardingDao` | `SettingsViewModel.kt:18,116,362` | Xtream live-onboarding state observation | No domain onboarding-status port exists; keep until the onboarding state is modeled in `domain` |
-| `XtreamIndexJobEntity` | `SettingsViewModel.kt:19,378` | Diagnostics warning formatting from the persistence entity | No domain indexing-status model exists; keep the entity-to-message conversion in the transitional feature boundary |
-| `XtreamLiveOnboardingStateEntity` | `SettingsOperationalModels.kt:3,92` | Persistence entity to Settings UI-model mapping | No domain onboarding-status model exists; keep the mapper transitional |
-| `DatabaseMaintenanceSnapshot` | `SettingsOperationalModels.kt:4,128` | Database-health snapshot to maintenance UI-model mapping | No domain maintenance snapshot exists; keep the mapper transitional |
-| `PreferencesRepository` | `ParentalControlGroupViewModel.kt:6,48`; `SettingsAppUpdateActions.kt:10,16`; `SettingsDerivedStateObservers.kt:9,94,119,136`; `SettingsGuideDefaultCategoryBindings.kt:4,23`; `SettingsObserverRegistrations.kt:6,28,103,140,163`; `SettingsProviderActions.kt:19,37`; `SettingsStateBindings.kt:11,35`; `SettingsViewModel.kt:20,107` | Preference flows and writes for parental, update, provider, EPG, recording, playback, and presentation state | The repository is still the concrete implementation of a broad preference surface; no complete domain preference contract exists |
-| `ProviderSyncCommands` | `SettingsProviderActions.kt:18,39`; `SettingsSyncActions.kt:6,19`; `SettingsViewModel.kt:21,114` | Provider sync, retry, background EPG, and index command dispatch | No domain sync-command contract covering these commands exists; keep until command ownership is moved |
-| `SyncRepairSection` | `SettingsSyncActions.kt:7,112-114,197-200`; `SettingsViewModel.kt:22` | Sync-warning and selected-section mapping | No domain repair-section model exists; keep until sync repair semantics are domain-owned |
-| `AudioCompatibilityMemoryStore` | `SettingsViewModel.kt:85,125,872` | Clears player compatibility memory when the related preference is changed | Separate concrete `:player` dependency; no feature-to-feature replacement or Settings port exists |
-
-No row is removal-ready in this slice. The audit therefore produces no source
-or Gradle dependency changes; each row remains explicitly owned by Phase 7.
-
-## Catalog extraction inventory (Task 0, 2026-09-02)
-
-The Catalog slice starts with a temporary `:data` dependency for the concrete
-preference and provider-sync implementations already consumed by the app
-presentation. These imports are recorded explicitly so the extraction does
-not turn a presentation move into an untracked data-layer contract.
-
-| Feature | Imported implementation | Current consumer | Removal direction | Owner |
-|---|---|---|---|---|
-| catalog | `data.preferences.PreferencesRepository` | `MoviesViewModel`, `MovieDetailViewModel`, `SeriesViewModel`, `SeriesDetailViewModel`, `VodViewModel`, `FavoritesViewModel`, `DashboardViewModel` | focused domain-facing preference contracts | Phase 7 |
-| catalog | `data.sync.ProviderSyncStateSource` | `DashboardViewModel` | domain-facing provider synchronization-status contract | Phase 7 |
-
-Catalog does not gain a `:player` dependency or any sibling-feature
-implementation dependency. Cast, download, update, plugin, playback
-preparation, Settings-dialog composition, and the app shell remain app-owned
-adapters at the extraction boundary. The exact source inventory and import
-scan are stored under `validation/phase5_catalog/`.
-
-## Live extraction inventory (Task 0/4, 2026-08-28)
-
-The live presentation seam now consumes narrow app-host ports while the
-existing implementations remain in the composition root or Playback feature.
-These are transitional adapters, not new policy owners:
-
-| Feature | Imported implementation | Current consumer | Removal direction | Owner |
-|---|---|---|---|---|
-| live | `data.preferences.PreferencesRepository` | Home live preferences and MultiView status adapter | domain-owned live presentation preference contract | Phase 7 |
-| live | `data.sync.ProviderSyncStateSource` | Home provider synchronization state | domain-owned provider sync state contract | Phase 7 |
-| live | `app.plugins.StreamVaultPluginManager` | `AppLivePreviewStreamPreparer` | host-provided stream preparation contract | Phase 5/7 |
-| live | `app.tvinput.TvInputChannelSyncManager` | `AppLiveSurfaceRefreshAdapter` | host-provided TV-input refresh contract | Phase 5/7 |
-| live | `feature.playback.preview.LivePreviewHandoffManager` | `AppLivePreviewHandoffAdapter` | keep lifecycle/release ownership in Playback | Phase 5 |
-| live | `feature.playback.multiview.MultiViewManager` | `AppLiveMultiViewStatusAdapter` | keep slot ownership in Playback; expose status only | Phase 5 |
-
-The `:feature:live` module itself depends only on `:core:navigation`, `:core:ui`,
-`:domain`, `:data`, and `:player`; its boundary verifier rejects app and
-feature-to-feature imports. Home and Guide ViewModels are being moved behind
-the live boundary incrementally: `HomeViewModel`, the sidebar/preview surface,
-the loading/preview host supporting composables, the live source switcher, Home
-quick-filter panel/chip row, the reusable EPG guide shortcut chip, the shared
-EPG guide-now ticker/provider, EPG search field/row, Home reorder top bar, and
-Home channel row surface, Home channel-results header, Home category sidebar
-header, channel-content state host, channel-list host, and category-list host
-and hidden category/channel dialogs, the category options dialog, the
-quick-filter add dialog, delete-group dialog, rename-group dialog, M3U category
-organizer dialog, M3U series-assignment dialogs, EPG search overlay, guide
-category picker, compact guide-program dialog, guide control rows, EPG preview
-pane, guide toolbar, guide message state, hero badge, timeline header, grid
-rows/cells, grid host, EPG content host, focus reconciliation, Home layout
-metrics, and Home parental-lock policy are now feature-owned, with the options
-overlay and EPG-match dialog also feature-owned
-and the EPG guide category/channel lock policies, category-selection fallback
-policy, and schedule metrics now feature-owned
-while the Home and EPG
-screen/dialog composition remains in `:app` for the remaining app-specific
-dialogs.
-
-The Live-browse remote shortcut dispatcher is no longer app-owned: its handler
-types, Android colour-key mapping, and action dispatch live in
-`feature/live/.../presentation/remote/LiveRemoteShortcutDispatch.kt`. The app
-retains only player-wide shortcut dispatch and adapts Home callbacks to the
-feature handler API. This is a presentation move, not a new transitional
-implementation dependency.
-The source switcher receives localized labels from the app at the current
-composition boundary. Live-owned
-presentation primitives, `HomeUiState`, `HomePreviewUiState`, Guide
-mode/density/reminder presentation models, hidden collection dialogs, the
-category options dialog, the quick-filter add dialog, the delete-group dialog,
-and the rename-group dialog, M3U category organizer dialog, and M3U
-series-assignment dialogs, the EPG search overlay, guide category picker, and
-compact guide-program dialog, guide control rows, EPG preview pane, guide
-toolbar, guide message state, hero badge, stable EPG channel-key helper,
-timeline header, grid rows/cells, grid host, EPG content host, focus
-reconciliation, Home layout metrics, Home parental-lock policy, EPG guide
-category/channel lock policies, category-selection fallback policy, schedule
-metrics, options overlay, and EPG-match dialog are independent of app code;
-screen/resource moves, golden/runtime/performance gates, and legacy app cleanup
-remain open, so no live slice acceptance gate is claimed yet.
-
-Home route composition is also app-independent at the screen boundary: Guide
-uses `LiveRoutePatterns.EPG`, and MultiView is exposed through an injected
-callback. The app graph remains the sole owner of the concrete MultiView
-destination string.
-
-The Home screen consumes `LiveHomeScaffoldContent`; the app graph supplies the
-existing `AppScreenScaffold` adapter with the same top-bar, compact-header,
-and hidden-screen-header configuration. Shell ownership remains in `:app`,
-but the Live presentation no longer imports that app shell directly.
-
-Home also consumes the existing `LiveMultiViewPlannerContent` port and a
-queue-status callback. The app graph owns the Playback `MultiViewViewModel`
-and dialog adapter; no `feature:live` source imports Playback implementation
-packages.
-
-`LiveHomeDialogsHost` now lives in `:feature:live` and consumes typed
-`LiveMultiViewPlannerContent` and `LiveAddToGroupContent` ports. The app graph
-continues to own the concrete Playback planner and Add-to-Group dialog; the
-dialog host itself and its Live strings no longer belong to `:app`.
-
-`LiveHomeScreen` now also lives in `:feature:live`. The app graph invokes it
-with the existing shell, MultiView planner, Add-to-Group dialog, and queue
-adapters, preserving composition-root ownership of platform and route
-navigation. The default Home resource keys were copied into the feature;
-non-default locale parity, behavior tests, golden/runtime checks, and legacy
-resource cleanup remain open.
-
-The Guide surface is now feature-owned as `LiveEpgScreen`,
-`LiveEpgHeroComponents`, `EpgViewModel`, and the deferred
-`LivePlayerTransparentGuideOverlay`. The app graph supplies the typed playback
-callbacks and `LiveEpgScaffoldContent` adapter; the feature owns Guide state,
-preview lifecycle calls, controls, dialogs, hero/grid composition, and default
-EPG strings. EPG behavior tests, non-default locale parity, golden/runtime
-checks, and legacy cleanup remain open.
-
-## System extraction inventory (Task 0, 2026-09-04)
-
-Approved project dependencies: `:core:navigation`, `:core:ui`, `:domain`.
-Direct `:data` imports: none.
-App adapters: `SystemWelcomePort` maps `SyncProgressBus` and app `BuildConfig`;
-`SystemPluginManagementPort` delegates to `StreamVaultPluginManager` and
-`ProviderSourceRegistry`.
-Phase 7 follow-up: keep these ports narrow; do not move plugin IPC, provider
-ownership, playback routing, or Android startup infrastructure into the feature.
-
-The pre-extraction source, import, resource, test, and consumer evidence is in
-`validation/phase5_system/`. The Task 0 note above records the pre-extraction
-state; the current implementation and gate status are recorded below.
-
-## System extraction status (Task 9, 2026-09-04)
-
-The System presentation slice is now feature-owned: Welcome, Downloads,
-Plugins, and their `SystemGraph` live under `:feature:system`; `AppNavHost`
-supplies only navigation actions and the app scaffold adapter. The app supplies
-the two narrow transitional adapters named above:
-`AppSystemWelcomeAdapter` maps BuildConfig and `SyncProgressBus`, while
-`AppSystemPluginManagementAdapter` delegates plugin operations to the existing
-app runtime and provider-source registry.
-
-The feature has zero direct `:data`, `:player`, `:app`, or sibling-feature
-production imports and depends only on `:core:navigation`, `:core:ui`, and
-`:domain`. The plugin manager, messenger, work coordinator, provider registry,
-and playback routing remain app-owned. The 33-string locale audit is recorded
-in `validation/phase5_system/task9-resource-cleanup.md`; 30 feature-only app
-duplicates were removed and the three shared keys remain intentionally
-app-owned.
-
-Task 9 structural, resource, unit-test, compilation, lint, and assembly gates
-are recorded in `validation/phase5_system/task9-structural-verification.md`.
-The follow-up app resource/code cleanup now makes `:app:lintDebug` pass with the
-maintained baseline covering 374 current findings; stale historical entries have
-been removed. Task 10
-connected acceptance and reviewed goldens pass 19/19, the paired incremental
-thresholds pass, and the steady-state clean/warm build-overhead guardrail
-passes. The full System
-execution report is `docs/COMPOSE_REDUCTION_PHASE5_SYSTEM_REPORT.md`; the
-supported general profile refresh, physical-device/accessibility gates, and
-authorized production-provider/plugin journeys remain open follow-ups.
diff --git a/docs/COMPOSE_REDUCTION_PHASE6_BACKUP_VIEWS_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE6_BACKUP_VIEWS_REPORT.md
deleted file mode 100644
index a8d50d15..00000000
--- a/docs/COMPOSE_REDUCTION_PHASE6_BACKUP_VIEWS_REPORT.md
+++ /dev/null
@@ -1,33 +0,0 @@
-# Phase 6 Backup Import Views Experiment Report
-
-## Environment
-
-- Repository `HEAD`: `a223337b` (`feat: add views backup import preview experiment`); benchmark and paired-host changes were working-tree experiment changes on top of this commit.
-- Device/API/resolution: `Television_1080p(AVD) - 16`, API 36, 1920x1080.
-- APK/runner: `com.streamvault.app.debug` debug fixture launched by the Macrobenchmark APK; `StartupMode.WARM`; five iterations per mode.
-- Runner caveat: Macrobenchmark warned that the device is an emulator and the target is debuggable, so results are directional rather than production-device estimates.
-
-## Results
-
-| Mode | Initial display median | Interaction `frameDurationCpuMs` P50 | Interaction `frameOverrunMs` P50 | Functional result |
-|---|---:|---:|---:|---|
-| Compose | 3,319.6 ms | 511.6 ms | 718.5 ms | Passed |
-| Views | 3,531.1 ms | 567.0 ms | 789.0 ms | Passed |
-
-Relative to Compose, Views was approximately 6.4% slower on initial display, 10.8% slower on interaction CPU frame duration, and 9.8% slower on frame overrun.
-
-The focused connected presentation matrix passed for both modes (2 tests), the reviewed Views golden capture passed, and the focused Views controller unit test passed. Macrobenchmark result artifacts were written under `benchmark/build/outputs/connected_android_test_additional_output`.
-
-PSS/memory was not collected because the candidate already failed the primary performance comparison; no memory conclusion is drawn from this report.
-
-## Behavior and visual review
-
-Both implementations supported the paired journey: initial focus, conflict-strategy change, provider toggle, and Back dismissal. The Views candidate rendered the required title, summaries, controls, footer, and focus state in the 1920x1080 capture.
-
-## Decision
-
-Drop the Views candidate. It did not produce the required repeatable performance win and was slower in both primary benchmark dimensions. Keep Compose as the production implementation. The candidate was never wired into the production Settings flow.
-
-## Follow-up
-
-Do not spend more time polishing this Views surface. Use the retained benchmark evidence to choose the next Phase 6 candidate or optimization target. If a future candidate is evaluated, repeat the same control/candidate protocol before visual polish.