rank model variants on average precision, and say when they are not comparable

The workbench ranks model variants by a stored F1, each measured at that
variant's own confidence threshold. A conservatively calibrated detector then
looks worse than a liberal one without detecting anything differently: the
number says as much about the threshold as about the model. POST
/detection/runs/compare ranks on average precision instead, which describes the
whole ranking a model produced, and keeps each run's own-threshold F1 visible
next to it so the difference between the two readings is auditable.

Comparability comes before the ranking. Runs over different source rasters,
scored against different references, without a proven inference footprint, or
covering a different evaluated population are not alternatives to one another,
and no metric makes them so. The report names which of those applies and still
returns the numbers — they are simply not a ranking.

Each run is scored through the same QA path the workbench uses, so a comparison
and the persisted quality checks cannot drift apart.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

backend/app/api/routes/detection.py

@@ -10,6 +10,8 @@ from app.schemas import (
     AnalysisQaResponse,
     DetectionListResponse,
     DetectionModelsResponse,
+    DetectionComparisonRequest,
+    DetectionComparisonResponse,
     DetectionQaRequest,
     DetectionRead,
     DetectionRunListResponse,
@@ -22,6 +24,7 @@ from app.schemas import (
     ModelAssetListResponse,
     YoloPreflightResponse,
 )
+from app.services.detection_comparison_service import DetectionComparisonService
 from app.services.detection_service import DetectionService
 from app.services.model_asset_catalog_service import ModelAssetCatalogService
 from app.services.model_registry_service import ModelRegistryService
@@ -232,6 +235,27 @@ def get_dataset_detection_geojson(
     )
 
 
+@router.post("/runs/compare", response_model=Envelope[DetectionComparisonResponse])
+def compare_detection_runs(payload: DetectionComparisonRequest, db: Session = Depends(get_db)) -> dict:
+    """Rank several runs against one reference on average precision.
+
+    The workbench ranks model variants by a stored F1 measured at each
+    variant's own confidence threshold, which orders the thresholds as much as
+    the models. Average precision describes the whole ranking a model produced.
+    Comparability is reported first: runs over different rasters, different
+    references or different inference coverage are not alternatives.
+    """
+
+    return envelope(
+        DetectionComparisonService.compare_runs(
+            db,
+            analysis_run_ids=payload.analysis_run_ids,
+            reference_dataset_id=payload.reference_dataset_id,
+            iou_threshold=payload.iou_threshold,
+        )
+    )
+
+
 @router.post(
     "/runs/{analysis_run_id}/qa/reference",
     response_model=Envelope[AnalysisQaResponse],