docs mobile pagination sync device 20261020

Pagination and optional sync: Android device follow-up

Status: LOCAL_DONE Validation: PASS Integration: INTEGRATION_PENDING

The user made the ADB device available for tests on 2026-10-20, authorizing this scoped native follow-up to the source review. This overrides the earlier native deferral for this task only. Source parent: 46bdf737, worktree/branch codex/mobile-pagination-review. Other occupied checkouts are unchanged.

Device safety and candidate

Physical Android API36, arm64, model2311DRK48G. Serial and full ADB inventory remain in temporary local files. Updated only com.anpfuel.app.debug and its app/data test packages using adb install -r; the installed and candidate signer SHA256 matched. No uninstall, package-data clear, user-data seeder, account/consent action, contribution, GPS request or photo capture was part of these tests. Direct instrumentation explicitly passes -e clearPackageData false, avoiding the repository's data-clearing orchestrator configuration. Production package com.anpfuel.app is untouched.

The debug app's existing SQLite header was11 before update and16 afterward. All migrations were registered; the targeted tests independently prove the 11→12→13→14→15 and15→16 segments with retained fixtures. The actual app was not reset or seeded. This observes its schema transition, not an independent census of its private rows.

Candidate product APK SHA256: 5531e0cdad0e6402087b075d6369046b89caae9a0140fcd88e407bddc2727680. The product remains installed. Network test recorded Wi-Fi/mobile-data as1/1, restored them and observed validated connectivity returning. Screen timeout and device-wide locale were not changed. Synthetic light/dark screenshots were visually inspected and stay in /tmp; no personal screen capture or APK is versioned.

Native findings and changes

• The initial data run found invalid fixture UUIDs and missing survey_week parents. After those fixes, the targeted-CNPJ fixture exposed that every synthetic station shared one CNPJ; assigned distinct fixture identities. Corrected the recency fixture's expected id tie order to match the established date/price/id contract. Production SQL and identity validation were not relaxed.
• The new full-list Compose test initially used a plain ComponentActivity with a Hilt-only weekly selector. StationsContent is now internal and accepts a composable weekly-action slot, preserving its existing default. The list test supplies an empty action and exercises actual production list content in isolation.
• Two existing feed tests still expected pre-October14 navigation. Verified source history: contribution already uses Routes.CAPTURE (9494e3118); cards already use the station/fuel page (e2a58f254). Updated exact route assertions to those established contracts, retaining click and value checks.
• The staging smoke now requests twenty rows, validates unique city-bound pages and conditionally continues one page. Removed the invalid assumption that every city fits one100-row response; no larger live traversal or provider load test is performed.
• Added a persistent125-station Room scenario: close/reopen, twenty-row offline windows, no network/GPS calls, origin isolation and public clearing that retains detail rows. Added a5,000-station UI scenario with fewer than40 composed station semantics nodes at sampled positions, reaching the final row. Failed continuation retains the first window and retry appends the next window.

Executed acceptance

Data instrumentation:27 cases, zero failures/skips. Classes:

• DirectoryCacheMigration15To16Test (1)
• DirectoryCacheTransactionTest (7)
• DirectoryCatalogOfflineDeviceTest (1)
• StationPriceDaoPagingTest (4)
• DirectorySyncWorkerTest (7)
• StorageStatsRepositoryIntegrationTest (1; actual fixture import counts2,344 summaries/19,676 station prices)
• V11ToV12DatabaseMigrationTest (3), V12ToV13DatabaseMigrationTest (1), V13ToV14QualityOutboxMigrationTest (1), V14ToV15QualityKeyErasureMigrationTest (1)

App UI instrumentation:8 cases, zero failures/skips: StationsListSummaryTest (1), StationsPaginationDeviceTest (2), CommunityFeedScreenTest (4), StationsScrollFrameDeviceTest (1). Staging HTTPS/validated connectivity:2 cases PASS, with a separate physical loss/return case PASS. Total:38 executed native cases, zero failures/errors/skips. These explicit probes execute rather than skip; they use public reads and an isolated in-memory cache. The synthetic WorkManager registration test in the real app package was intentionally excluded because it overwrites the user's schedule; isolated native worker tests cover schedule parameters and execution guards.

Commands after APK assembly/install:

adb shell am instrument -w -r -e clearPackageData false \
  -e class '<comma-separated data classes above>' \
  com.anpfuel.data.test/androidx.test.runner.AndroidJUnitRunner
adb shell am instrument -w -r -e clearPackageData false \
  -e class com.anpfuel.app.ui.stations.StationsListSummaryTest,com.anpfuel.app.ui.stations.StationsPaginationDeviceTest,com.anpfuel.app.ui.community.CommunityFeedScreenTest,com.anpfuel.app.ui.stations.StationsScrollFrameDeviceTest \
  com.anpfuel.app.debug.test/androidx.test.runner.AndroidJUnitRunner
adb shell am instrument -w -r -e clearPackageData false -e directoryLiveStaging true \
  -e class 'com.anpfuel.app.ui.stations.StagingDirectorySyncDeviceTest#actualValidatedInternetIsObserved,com.anpfuel.app.ui.stations.StagingDirectorySyncDeviceTest#actualVpsReadSurvivesOutageAndReconnect' \
  com.anpfuel.app.debug.test/androidx.test.runner.AndroidJUnitRunner
adb shell am instrument -w -r -e clearPackageData false -e directoryLiveStaging true \
  -e class 'com.anpfuel.app.ui.stations.StagingDirectorySyncDeviceTest#physicalInternetLossAndReturnReachTheConnectivityObserver' \
  com.anpfuel.app.debug.test/androidx.test.runner.AndroidJUnitRunner

Affected app/data test APK builds and app lint PASS; no backend/domain behavior changed during this follow-up, so their previous source-review acceptance is retained instead of rerunning aggregate suites. Final whitespace and tracked/changed-file secret checks PASS. Installed product bytes were independently pulled and matched the candidate SHA256 above.

Frame diagnostic and limits

The first experiment collected FrameMetrics during Compose test-clock scrolling. It produced high numbers (full semantic scan:2,255 frames, p50≈43.3ms/p95≈100.6ms; short injected gestures:34 frames, p50≈177.4ms/p95≈333.9ms). These mixed virtual-clock/automation samples are retained as diagnostic history and do not measure normal user scrolling. The final functional Compose test checks virtualization and continuation only.

StationsScrollFrameDeviceTest separately uses MainActivity, the ordinary Choreographer clock, six actual touchscreen gestures and a synthetic5,000-row catalog. Its state is supplied only to composition; it does not seed storage. FrameMetrics excludes first-draw frames. Final physical diagnostic:828 frames, p50=9.263231ms, p95=16.781043ms. These normal-clock results supersede the unsuitable virtual-clock experiments for this device diagnostic. The test asserts successful input/rendering and records timings; no invented FPS threshold or performance acceptance follows from a passing diagnostic test. This is a debug/instrumented single-device sample, not release profiling or certification.

Only the named native scenarios are accepted. Remaining whole-app flows, a representative profileable/release performance budget and protected integration/revalidation against newer maintained dev remain owed. No production certification, deployment, PR/merge or wiki publication is claimed.

Ephemeral logs: /tmp/mobile-pagination-review-device-{data,ui,staging,network,frame,real-frames}.log, build/install logs with the same prefix. Test sources and this record are the reproducible evidence.