Use Jev initial decisions and values-only semantic queries

backend/app.py

@@ -203,7 +203,7 @@ def terminology_asset(asset_path: str):
 
 
 def classify_batch(batch: Batch, review: bool):
-    provider = "openai" if review else os.getenv("PIPELINE_CLASSIFIER_PROVIDER", "openai")
+    provider = "openai" if review else os.getenv("PIPELINE_CLASSIFIER_PROVIDER", "jev")
     if provider not in {"jev", "openai", "anthropic", "google"}:
         raise HTTPException(503, "The classification provider is not configured correctly.")
     model = os.getenv("PIPELINE_REVIEW_MODEL" if review else "PIPELINE_CLASSIFIER_MODEL") or default_model_for(provider)
@@ -244,7 +244,7 @@ def review(batch: Batch):
 @app.post("/api/embed", dependencies=[Depends(authorize), Depends(model_rate_limit)])
 def embed(event: EventRequest):
     """Encode one EDF query; terminology vectors are always precomputed."""
-    query = EventDefinitionForm(name="browser-edf", **event.edf).to_embedding_query()
+    query = EventDefinitionForm(name="browser-edf", **event.edf).to_embedding_query(include_labels=False)
     try:
         response = _get_openai_client().embeddings.create(
             model="text-embedding-3-large", input=query, dimensions=3072, encoding_format="float")

src/data/loaders.py

@@ -97,8 +97,21 @@ class EventDefinitionForm:
         terms = [self.preferred_name.strip(), *(term.strip() for term in self.synonyms)]
         return " ".join(dict.fromkeys(term for term in terms if term))
 
-    def to_embedding_query(self) -> str:
-        """Render the EDF subset used for semantic embedding matching."""
+    def to_embedding_query(self, *, include_labels: bool = True) -> str:
+        """Render retrieval fields; retain labeled defaults for study compatibility."""
+        if not include_labels:
+            values = [self.preferred_name]
+            if self.narrow_definition:
+                values.append(self.narrow_definition)
+            if self.synonyms:
+                values.append(", ".join(self.synonyms))
+            values.extend(self.narrow_decision_rules)
+            values.extend(self.candidate_anchor_codes)
+            if self.morphology:
+                values.append(self.morphology)
+            if self.anatomical_location:
+                values.append(self.anatomical_location)
+            return "\n".join(str(value) for value in values)
         lines = [f"Preferred name: {self.preferred_name}"]
         if self.narrow_definition:
             lines.append(f"Narrow definition: {self.narrow_definition}")