# Test Strategy

GeoIntel is data-heavy. Tests must focus on contracts, geospatial correctness, and regression safety.

## Test layers

### 1. Unit tests

Backend:

- Geometry validation.
- Area calculation.
- Dataset metadata extraction wrappers.
- Analysis formulas.
- QA/QC metrics.

Frontend:

- API client methods.
- Basic component rendering.
- Empty/loading/error states.

### 2. Integration tests

- FastAPI endpoint + database session.
- Project CRUD.
- Area creation with valid GeoJSON.
- Dataset metadata refresh with fixture files.
- Vector clipping using small fixture polygons.
- QA/QC using fixture detection/reference polygons.

### 3. Smoke tests

Minimum commands:

```bash
# backend
pytest

# frontend
npm run build
npm run lint
```

### 4. Geospatial correctness tests

Use small deterministic fixtures:

- One AOI polygon.
- Three building polygons.
- Two detection polygons.
- Known intersection results.
- Expected precision/recall/F1.

No huge geospatial datasets in the repo.

## Fixtures

Place lightweight fixtures in:

```text
tests/fixtures/
  geojson/
  rasters/
  vectors/
```

Large datasets are not committed. Document download/setup in `docs/FIXTURE_STRATEGY.md`.

## Required V1 acceptance tests

- Create project.
- Create area from GeoJSON.
- Upload or register vector fixture.
- Extract vector metadata.
- Calculate building stats.
- Create fake detection fixture through service/test-only path.
- Run QA/QC against reference polygons.
- Export QA output as GeoJSON or JSON.

## Regression rules

- Do not remove existing tests to make a pass green.
- When an endpoint changes, update API contract and frontend client together.
- If a feature is stubbed because dependency is unavailable, it must return an explicit `not_configured` state, not silent success.