# Codex Pass 04 — Detection and QA Skeleton

Goal: prepare object detection and QA/QC without requiring heavy model artifacts yet.

Implement:

- model registry schema/service
- detection run API contract
- detection result storage model
- GeoJSON export of detection results
- QA/QC service with IoU, precision, recall, F1 calculations using polygon fixtures

Acceptance:

- QA calculations are tested with simple polygons.
- Detection outputs can be stored and exported.
- Heavy model execution is behind a service interface.