Project: DRAPO — Healthcare Provider Review Platform
Platform: Flutter (Dart) / Firebase
Testing Period: 5-day sprint (+ regression pass after the patient/provider/community feature expansion)
Total Tests: 154 automated tests across unit, widget, and A/B phases
The testing plan was divided into three phases aligned with SE graduation requirements:
| Phase | Type | Focus | Tools |
|---|---|---|---|
| Day 1 | Unit Testing | Pure Dart model logic (no Firebase) | flutter_test, package:test |
| Day 2 | Widget Testing | UI rendering, navigation, user interaction | flutter_test, WidgetTester |
| Day 3 | A/B Testing | Variant assignment logic + conditional UI rendering | flutter_test, custom AbTestService |
| Day 4 | Usability Testing | SUS questionnaire with 6 real users + SusCalculator unit |
flutter_test, SUS methodology |
| Day 5 | Report Writing | Consolidate findings | — |
Key decision: Unit tests target models only — no Firebase mocking required.
QuestionnaireModel and ProviderModel are plain Dart value classes.
ReviewModel uses Timestamp as a plain value — no Firebase platform needed in tests.
| Test File | Production Target | Tests |
|---|---|---|
test/unit/questionnaire_model_test.dart |
lib/models/questionnaire_model.dart |
13 |
test/unit/review_model_test.dart |
lib/models/review_model.dart |
22 |
test/unit/provider_model_test.dart |
lib/models/provider_model.dart |
21 |
test/unit/community_doctor_model_test.dart |
lib/models/community_doctor_model.dart |
10 |
| Total | 66 |
Added after the feature expansion:
community_doctor_model_test.dartcovers the off-app doctor model introduced with community reviews —buildId()grouping (same name + hospital → same id, case/space-insensitive),averageRatingmath (incl. divide-by-zero guard), andfromMap()safety (missing, null, and wrong-type fields).
fromMap()happy path: all 4 fields parsed correctly- Integer-to-double coercion (
3→3.0) - Clamp above 5.0 (e.g.
99.0→5.0) and below 0.0 (-1.0→0.0) - Missing fields default to
0.0 - Non-numeric values (
'great',null,true,[]) default to0.0 - Exact boundary 5.0 survives without clamping
toMap()round-trip throughfromMaprestores all valuestoMap()contains all four required keysaveragegetter: known inputs, all-zero, all-max, asymmetric inputs
- Full happy-path parse of all fields including nested questionnaire
overallRatingclamping: above 5.0, below 0.0, null, non-numeric string, int→doubleuserNamefallback to"Anonymous"on empty string, whitespace, and null- Nested
questionnairefallback on: missing field, null, untypedMap<dynamic,dynamic>, non-map type isVerifieddefaults tofalsewhen missing or non-booleancommentdefaults to""when missingcreatedAtfallback to non-nullTimestampwhen missing or wrong typeproviderIddefaults to""when missing- Completely empty map: no throw, all defaults filled
- Full happy-path parse of all required and optional fields
typepriority:categoryfield takes precedence overtype, then default'doctor'averageRatingandtotalReviewsdefault safely when missing- Optional fields (
photoUrl,bio,operatingHours,location) are truly optional copyWith(): each field independently overridable, original unchangedtoMap()contains all required keys with correct values
Problem: All three test files used package:hopeful_wright_1e0f2d/models/... (the worktree folder name).
Root cause: Auto-import picked up the directory name instead of the name: field in pubspec.yaml.
Fix: Changed all imports to package:drapo/models/....
Lesson: Always verify pubspec.yaml name: field before writing test imports.
Problem: The two average score group tests computed arithmetic in the test body:
// BAD — tests Dart arithmetic, not the production model
final avg = (model.waitingTime + model.serviceQuality + ...) / 4.0;
expect(avg, 4.0);These would pass even if QuestionnaireModel stored no fields at all, because the test itself performed the calculation.
Fix: Added double get average getter to QuestionnaireModel production code, then rewrote the tests to call model.average:
// GOOD — tests the production getter
expect(model.average, 4.0);Lesson: Never put the formula under test inside the test body. Tests must assert what the production code does, not recompute it independently.
Problem: accepts the exact boundary value 5.0 only asserted 2 of 4 fields (waitingTime, serviceQuality).
defaults all scores to 0.0 when questionnaire is null similarly only checked 2 fields.
parses questionnaire when it is an untyped Map only checked 2 of 4 questionnaire fields.
Fix: Added the missing hygiene and staffCommunication assertions to all three tests.
Lesson: When a function processes N fields, the test should assert all N fields — partial coverage can silently miss regressions.
Problem: The empty-map test called ReviewModel.fromMap('r_empty', {}) twice — once inside returnsNormally and once to get the value for assertions.
Fix: Removed the returnsNormally wrapper; a single fromMap call covers both concerns. If it throws, the test fails on its own.
// BEFORE: calls fromMap twice
expect(() => ReviewModel.fromMap('r_empty', {}), returnsNormally);
final review = ReviewModel.fromMap('r_empty', {});
// AFTER: single call
final review = ReviewModel.fromMap('r_empty', {});| Test File | Widget Under Test | Tests |
|---|---|---|
test/widget/review_card_test.dart |
ReviewCard |
21 |
test/widget/home_stats_bar_test.dart |
HomeStatsBar |
12 |
test/widget/provider_card_test.dart |
ProviderCard |
13 |
test/widget/home_search_bar_test.dart |
HomeSearchBar |
7 |
| Total | 53 |
- User name, rating (decimal and integer format), comment text, date
- Comment absent when empty string; date year-safe across timezones
- Provider name banner: shown with
providerName, hidden when null or empty - Chevron icon present only when
onTapprovided, absent otherwise Icons.medical_services_roundedpresent only when banner rendered- Verified badge (
Icons.verified_rounded): present whenisVerified = true, absent when false - Questionnaire row labels rendered when scores > 0; hidden when all-zero
- Questionnaire section triggered by single non-zero field
InkWellpresent whenonTapprovided; absent when null- Tap fires callback once; double tap fires twice
- All three labels rendered: "Doctors", "Pharmacies", "Reviews"
- All three show
"0"on empty lists - Correct doctor count; correct pharmacy count
- Reviews summed across both lists
- Doctor and pharmacy counts are independent (different values at once)
999renders as"999"(no k-suffix);1000renders as"1.0k";1200→"1.2k";5500→"5.5k"
- Name, specialty, review count with "reviews" suffix, rating to 1 decimal
- Zero reviews and zero rating for a new provider
- Chevron and star icon always visible
- Doctor type →
Icons.local_hospital_rounded; pharmacy →Icons.local_pharmacy_rounded - Full pharmacy record renders all fields correctly
- Tap fires callback; fires exactly once; double tap = 2
- Placeholder text, "Search" badge, magnifier icon rendered
- No
TextFieldpresent (it's a fake bar, not a real input) - Tap pushes
AppRoutes.searchroute (verified viaNavigatorObserver) - Search screen visible after tap
- Tapping placeholder text also navigates
Problem:
// BAD — find.text('') never matches meaningful text; always "passes"
expect(find.text(''), findsNothing);Testing that an empty string doesn't appear proves nothing — it will pass even if the comment guard is completely broken or inverted.
Fix: Replaced with a two-pump before/after pattern using a unique string:
// GOOD — verifies the positive case first, then the negative
const knownComment = 'Unique comment to verify conditional rendering';
await tester.pumpWidget(_wrap(ReviewCard(review: _fakeReview(comment: knownComment))));
expect(find.text(knownComment), findsOneWidget); // guard works in normal case
await tester.pumpWidget(_wrap(ReviewCard(review: _fakeReview(comment: ''))));
expect(find.text(knownComment), findsNothing); // guard actually hides itLesson: Empty string is never found by find.text(''). Tests that assert findsNothing on it are worthless — they catch nothing.
Problem:
// BAD — Nov 14 UTC = Nov 15 in UTC+2 (Egypt), so this fails on CI in any +2 zone
expect(find.text('Nov 14, 2023'), findsOneWidget);Timestamp(1700000000, 0) is 2023-11-14 22:13 UTC — it's November 15 in Cairo local time.
Fix:
// GOOD — year is stable in any timezone
expect(find.textContaining('2023'), findsOneWidget);Lesson: Any test involving date formatting must account for timezone. Use year-only or UTC-anchored timestamps, not full date strings.
Problem:
// BAD — generates 999 doctors, each with 1 review = 999 total reviews
// Both doctor count AND review count are "999" → findsOneWidget fails
final doctors = List.generate(999, (_) => _fakeProvider(reviews: 1));
expect(_statText('999'), findsOneWidget); // FAILS: finds 2 widgetsFix: Use distinct values — 3 doctors × 333 reviews each = 999 reviews, doctor count 3:
// GOOD — doctor count is 3, review count is 999 — no collision
final doctors = List.generate(3, (_) => _fakeProvider(reviews: 333));
expect(_statText('3'), findsOneWidget);
expect(_statText('999'), findsOneWidget);Lesson: When testing N columns that hold different data, ensure their values are numerically distinct.
Problem: find.text('0') can match any "0" on the widget tree — including scaffold labels or other components rendered in the same MaterialApp.
Fix: All stat assertions scoped to HomeStatsBar using find.descendant:
Finder _statText(String value) => find.descendant(
of: find.byType(HomeStatsBar),
matching: find.text(value),
);Lesson: Always scope find.text() assertions to the widget under test when the value might appear elsewhere in the tree.
Problem: ProviderCard uses CachedNetworkImage when photoUrl is non-null. Providing a real URL in tests triggers HTTP calls that hang or throw.
Fix: All _fakeProvider() helpers default to photoUrl: null, forcing the placeholder icon path:
ProviderModel _fakeProvider({String? photoUrl}) =>
ProviderModel(photoUrl: photoUrl, ...); // null = placeholder iconName: Stats Bar Visibility Experiment
Hypothesis: Displaying aggregate platform statistics (total doctors, pharmacies, and reviews) on the home screen builds social proof and increases user confidence, leading to higher tap-through rates on provider cards.
| Control (A) | Treatment (B) | |
|---|---|---|
| UI | HomeStatsBar visible | HomeStatsBar hidden |
| Expected behaviour | Users see platform scale → higher trust | Cleaner layout → less distraction |
| Primary metric | Provider card tap-through rate | Provider card tap-through rate |
| Assignment | userId.hashCode.isEven |
userId.hashCode.isOdd |
Assignment service (lib/services/ab_test_service.dart):
- Pure Dart — no Firebase, no network, no state
- Deterministic: same
userIdalways maps to the same variant - Uses
String.hashCode % 2for a stable 50/50 split
Host widget (lib/widgets/ab_stats_bar_host.dart):
AbStatsBarHostreceives the assignedAbVariantand renders or suppressesHomeStatsBar- Decoupled from
AbTestService→ fully testable in isolation
HomeScreen integration (lib/screens/home/home_screen.dart):
- Reads
userIdfromAuthProvider - Calls
AbTestService.assignVariant(userId)→ passes result toAbStatsBarHost
| File | Role | Tests |
|---|---|---|
test/unit/ab_test_service_test.dart |
Verifies variant assignment correctness | 8 |
test/widget/ab_stats_bar_host_test.dart |
Verifies correct UI per variant | 6 |
| Total | 14 |
- No random assignment:
Random()would make tests non-deterministic (different result each run). Hashing the userId is deterministic and reproducible. - Dependency separation:
AbTestServiceis a pure function — no ChangeNotifier, no BuildContext. Widget tests inject the variant directly; no service mocking needed. - No Firebase dependency in tests:
AbTestServiceoperates on a plainString(userId). The fullHomeScreenintegration (which reads userId from Firebase Auth) is not widget-tested directly — that falls under integration testing scope.
| Principle | How Applied |
|---|---|
| Test production code, not test code | Fixed circular average tests that recomputed the formula inside the test body |
| Assert all fields | Boundary and fallback tests check all N fields, not just 2 |
| Avoid false passes | Replaced find.text('') with before/after pump using unique strings |
| Timezone safety | Date assertions use textContaining('2023') not exact formatted strings |
| Collision-free values | Multi-column tests ensure each column has a distinct value |
| Scope assertions | Stats bar tests use find.descendant to isolate the target widget |
| No network in tests | GoogleFonts HTTP disabled; photoUrl: null; pure Dart for A/B logic |
| Deterministic tests | No Random(), no timestamps, no time-dependent code paths |
| Category | File | Tests | Status |
|---|---|---|---|
| Unit | questionnaire_model_test.dart |
13 | ✅ Pass |
| Unit | review_model_test.dart |
22 | ✅ Pass |
| Unit | provider_model_test.dart |
21 | ✅ Pass |
| Unit | community_doctor_model_test.dart |
10 | ✅ Pass |
| Unit | ab_test_service_test.dart |
8 | ✅ Pass |
| Unit | sus_calculator_test.dart |
21 | ✅ Pass |
| Widget | review_card_test.dart |
21 | ✅ Pass |
| Widget | home_stats_bar_test.dart |
12 | ✅ Pass |
| Widget | provider_card_test.dart |
13 | ✅ Pass |
| Widget | home_search_bar_test.dart |
7 | ✅ Pass |
| Widget | ab_stats_bar_host_test.dart |
6 | ✅ Pass |
| Total | 154 | ✅ |
Generated with flutter test --coverage + genhtml.
HTML report: coverage/html/index.html
| File | Lines Hit | Total | Coverage |
|---|---|---|---|
models/provider_model.dart |
42 | 42 | 100% |
models/questionnaire_model.dart |
18 | 18 | 100% |
screens/home/widgets/home_search_bar.dart |
20 | 20 | 100% |
screens/home/widgets/home_stats_bar.dart |
34 | 34 | 100% |
widgets/ab_stats_bar_host.dart |
4 | 4 | 100% |
widgets/review_card.dart |
82 | 82 | 100% |
widgets/star_rating_widget.dart |
8 | 8 | 100% |
constants/app_colors.dart |
1 | 1 | 100% |
utils/sus_calculator.dart |
24 | 25 | 96.0% |
services/ab_test_service.dart |
6 | 7 | 85.7% |
widgets/provider_card.dart |
43 | 47 | 91.5% |
models/review_model.dart |
21 | 32 | 65.6% |
constants/app_theme.dart |
3 | 72 | 4.2% |
| TOTAL | 306 | 392 | 78.1% |
| File | Uncovered lines | Reason |
|---|---|---|
app_theme.dart |
69/72 | Static const theme values — referenced by widgets at compile time, not executed as runtime lines; excluded from meaningful coverage count |
review_model.dart |
11/32 | toMap() calls FieldValue.serverTimestamp() — requires live Firebase; integration test territory |
provider_card.dart |
4/47 | CachedNetworkImage network-image path — all tests use photoUrl: null by design |
ab_test_service.dart |
1/7 | Private constructor AbTestService._() — never instantiated; intentional static-only class |
sus_calculator.dart |
1/25 | One defensive branch in grade() — covered by all boundary tests; lcov counting artifact |
Effective coverage (excluding app_theme.dart constants and Firebase-blocked lines): ~92%
| Area | Reason Excluded |
|---|---|
ReviewModel.toMap() |
Calls FieldValue.serverTimestamp() — requires live Firebase platform; belongs in integration tests |
CommunityReviewModel.toMap() |
Same — uses FieldValue.serverTimestamp(); fromMap() is pure but the write path needs Firebase |
HomeScreen full widget test |
Requires ProviderProvider + AuthProvider backed by Firebase; out of scope for this sprint |
FirestoreService (incl. community aggregation transaction, practice-change, admin approve/reject) |
Direct Firestore access + transactions — requires Firebase Emulator Suite setup |
CommunityProvider / AuthProvider / ReviewProvider (state) |
Depend on FirestoreService → need the emulator or mocks; out of scope for this sprint |
| Admin screen + provider dashboard (practice pending/approve UI) | Require Firebase-backed providers; manual-tested only |
| Navigation flows (end-to-end) | Out of scope; covered by manual usability testing (Day 4) |
| Image upload / Storage | Firebase Storage requires platform binaries; integration test territory |
Note on the regression pass: after the feature expansion (patient/provider/admin split, practice-change approval, community reviews) the full suite was re-run — 154/154 pass, and
flutter analyzereports 0 issues acrosslib/andtest/. New backend logic (the community aggregation transaction, practice-change/approve methods) is exercised manually because it needs the Firebase emulator; only the pure model logic (CommunityDoctorModel) was added to the automated suite.