diff --git a/DashAI/back/dataloaders/classes/dashai_dataset.py b/DashAI/back/dataloaders/classes/dashai_dataset.py
index 82d96b2c9..83a12eafd 100644
--- a/DashAI/back/dataloaders/classes/dashai_dataset.py
+++ b/DashAI/back/dataloaders/classes/dashai_dataset.py
@@ -637,7 +637,7 @@ def get_split(self, split_name: str) -> "DashAIDataset":
 
         new_splits = {"split_indices": {split_name: indices}}
         arrow_table = subset.arrow_table  # with_format("arrow")[:] ####Check
-        subset = DashAIDataset(arrow_table, splits=new_splits)
+        subset = DashAIDataset(arrow_table, splits=new_splits, types=self._types)
         return subset
 
     @beartype
diff --git a/DashAI/back/models/scikit_learn/sklearn_like_classifier.py b/DashAI/back/models/scikit_learn/sklearn_like_classifier.py
index a50af4a5b..919f81a57 100644
--- a/DashAI/back/models/scikit_learn/sklearn_like_classifier.py
+++ b/DashAI/back/models/scikit_learn/sklearn_like_classifier.py
@@ -36,10 +36,7 @@ def predict(self, x_pred: "DashAIDataset") -> "ndarray":
         from DashAI.back.dataloaders.classes.dashai_dataset import DashAIDataset
 
         if isinstance(x_pred, DashAIDataset):
-            try:
-                x_prepared = self.prepare_dataset(x_pred, is_fit=False)
-            except ValueError:
-                x_prepared = x_pred
+            x_prepared = self.prepare_dataset(x_pred, is_fit=False)
             x_pred = x_prepared.to_pandas()
         elif isinstance(x_pred, pd.DataFrame):
             pass
diff --git a/DashAI/back/services/scoring_service.py b/DashAI/back/services/scoring_service.py
index afb6ed9a2..b787b5917 100644
--- a/DashAI/back/services/scoring_service.py
+++ b/DashAI/back/services/scoring_service.py
@@ -90,7 +90,12 @@ def _get_all_profiles(
 
         return all_profiles
 
-    def get_available_profiles(self, task_name: Optional[str]) -> List[Dict[str, Any]]:
+    @inject
+    def get_available_profiles(
+        self,
+        task_name: Optional[str],
+        component_registry: "ComponentRegistry" = lambda di: di["component_registry"],
+    ) -> List[Dict[str, Any]]:
         """Get profiles available for a given task.
 
         Fetches profiles from the task's SCORING_PROFILES class attribute
@@ -101,39 +106,49 @@ def get_available_profiles(self, task_name: Optional[str]) -> List[Dict[str, Any
         task_name : Optional[str]
             Task class name (e.g., "TabularClassificationTask") or None.
             If None, returns all profiles from all tasks.
+        component_registry : ComponentRegistry
+            Registry to look up task class (injected).
 
         Returns
         -------
         List[Dict[str, Any]]
             List of profile dicts with keys: id, description, weights.
         """
-        all_profiles = self._get_all_profiles()
+        if task_name:
+            # Access the task class directly to avoid profile-ID collisions
+            # that occur when multiple tasks share the same profile ID keys
+            # (e.g. TabularClassificationTask and ImageClassificationTask both
+            # define "balanced").
+            try:
+                full_task_dict = component_registry[task_name]
+                task_class = full_task_dict.get("class")
+                if task_class and hasattr(task_class, "SCORING_PROFILES"):
+                    return [
+                        {
+                            "id": profile_id,
+                            "description": profile_data["description"],
+                            "weights": profile_data["weights"],
+                        }
+                        for profile_id, profile_data in (
+                            task_class.SCORING_PROFILES.items()
+                        )
+                    ]
+            except KeyError:
+                log.debug(f"Task {task_name} not found in registry")
+            except Exception as e:
+                log.warning(f"Error accessing SCORING_PROFILES for {task_name}: {e}")
+            return []
 
-        if not task_name:
-            # No task specified; return all profiles
-            return [
-                {
-                    "id": profile_id,
-                    "description": profile_data["description"],
-                    "weights": profile_data["weights"],
-                }
-                for profile_id, profile_data in all_profiles.items()
-            ]
-
-        # Filter profiles by task
-        result = []
-        for profile_id, profile_data in all_profiles.items():
-            if profile_data.get("task_name") == task_name:
-                result.append(
-                    {
-                        "id": profile_id,
-                        "description": profile_data["description"],
-                        "weights": profile_data["weights"],
-                    }
-                )
-
-        # If no profiles found for this task, return empty (graceful fallback)
-        return result
+        # No task specified — return one entry per unique profile ID across all tasks.
+        all_profiles = self._get_all_profiles()
+        return [
+            {
+                "id": profile_id,
+                "description": profile_data["description"],
+                "weights": profile_data["weights"],
+            }
+            for profile_id, profile_data in all_profiles.items()
+        ]
 
     def normalize_metric_value(
         self,
diff --git a/DashAI/front/src/components/models/ModelComparisonTable.jsx b/DashAI/front/src/components/models/ModelComparisonTable.jsx
index 78a7d1397..571df06b4 100644
--- a/DashAI/front/src/components/models/ModelComparisonTable.jsx
+++ b/DashAI/front/src/components/models/ModelComparisonTable.jsx
@@ -114,12 +114,20 @@ function ModelComparisonTable({
     fetchProfiles();
   }, [session?.task_name]);
 
+  // Stable string that changes only when a run's status changes.
+  // Used as a dep so the score fetch re-triggers after training completes
+  // without firing on every unrelated re-render of the parent.
+  const runStatusSignature = useMemo(
+    () => initialRuns.map((r) => `${r.id}:${r.status}`).join(","),
+    [initialRuns],
+  );
+
   // ────────────────────────────────────────────────────────────────────────
-  // Fetch scores when profile or runs change
+  // Fetch scores when profile, split, session or any run status changes
   // ────────────────────────────────────────────────────────────────────────
 
   useEffect(() => {
-    if (!runs.length || !selectedProfile || !session?.id) return;
+    if (!initialRuns.length || !selectedProfile || !session?.id) return;
 
     const fetchScores = async () => {
       setLoadingScores(true);
@@ -152,7 +160,7 @@ function ModelComparisonTable({
     };
 
     fetchScores();
-  }, [selectedProfile, metricSplit, session?.id]);
+  }, [selectedProfile, metricSplit, session?.id, runStatusSignature]);
 
   // ────────────────────────────────────────────────────────────────────────
   // Build columns
@@ -297,8 +305,8 @@ function ModelComparisonTable({
         </Tooltip>
       ),
       Cell: ({ row }) => {
-        const { statusCode, id } = row.original;
-        const isRunning = statusCode === 1 || statusCode === 2;
+        const { status, id } = row.original;
+        const isRunning = status === 1 || status === 2;
         if (isRunning) return "-";
 
         const scoreData = scores[id];