# Codex Phase 5 Prompt — Detection Pipeline Interface

Implement the object detection pipeline interface without pretending that AI works when models are absent.

## Goals

- Detection service abstraction.
- YOLO provider interface.
- `not_configured` behavior when YOLO is disabled or model missing.
- Test fixture provider for deterministic detection polygons in tests only.
- Detection result persistence.
- GeoJSON export of detections.

## Requirements

- Do not hardcode fake production detections.
- Store confidence, class, geometry and model metadata.
- Add API endpoint according to `docs/API_CONTRACTS.md`.
- Add frontend state that shows configured/not configured.

## Done when

- Detection endpoint exists.
- It returns `not_configured` by default unless explicitly enabled.
- Tests can use fixture provider.
- Persisted detections can be exported.