# Testing And Debugging

## Baseline Commands

```bash
flutter analyze
flutter test
dart run build_runner build --delete-conflicting-outputs
```

Focused suites:

```bash
flutter test test/features/share_intake
flutter test test/features/ai_digest
flutter test test/features/local
flutter test test/features/sync
flutter test test/core/llm
flutter test test/integrations/backend
```

## Manual Smoke Tests

Fake backend:

1. Launch with `USE_FAKE_BACKEND=true`.
2. Sign in with fake credentials.
3. Add a person.
4. Add a moment.
5. Add an idea.
6. Add a reminder.
7. Open Settings and verify flags.

Share intake:

1. Settings -> `Simulate Share Capture`.
2. Paste a sender-prefixed text share.
3. Confirm review sheet behavior.
4. Verify profile attachment or Share Inbox routing.
5. Open profile and check facts/messages/signals.

AI digest:

1. Configure provider in Settings.
2. Run LLM Debug if needed.
3. Enable weekly digest.
4. Run Private Digest Now.
5. Open AI Review.
6. Accept and dismiss drafts.

Notifications:

1. Request notification access from Settings.
2. Create or update reminders.
3. Verify scheduler tests still pass.
4. Test real delivery on device where possible.

## Debug Surfaces

Settings includes:

- LLM Debug
- LLM diagnostics export
- Run Private Digest Now
- Open AI Review
- Request Notification Access
- Simulate Share Capture
- Open Share Inbox
- Share Diagnostics

Sync view includes:

- pending changes
- rejected changes
- repair actions
- retry/requeue tools

## Common Failures

### Share intent delivered nothing

Check:

- `ENABLE_WHATSAPP_SHARE_INTAKE`
- iOS extension target build
- App Group ID matches Runner and extension targets
- `ReceiveSharingIntent.instance.getInitialMedia()`
- Share Diagnostics in Settings

### LLM digest creates bad suggestions

Check:

- `AnonymizedLlmContextBuilder` payload
- sensitive/unreviewed facts filtering
- prior suggestions included
- source age and `shared_message_datetime`
- parser response tests

### Background digest does not run

This can be normal. iOS background execution is best effort. Verify manual run
first, then check Workmanager setup, background modes, charging/network policy,
and run-state backoff.

### Full widget tests fail due layout hit targets

Some app-flow tests are sensitive to viewport/layout. Prefer focused tests for
feature work, then investigate full-suite failures separately with screenshots
or widget tree dumps.

## Test Data Rules

- Avoid real names, messages, phone numbers, or URLs in fixtures.
- Use synthetic names and pseudonymous tokens.
- Add regression tests for parser edge cases.
- Add privacy assertions when prompt payloads change.