diff --git a/reasoning_gym/algorithmic/jugs.py b/reasoning_gym/algorithmic/jugs.py
index 43e97620..b1458279 100644
--- a/reasoning_gym/algorithmic/jugs.py
+++ b/reasoning_gym/algorithmic/jugs.py
@@ -232,9 +232,9 @@ class JugsConfig:
 
     def validate(self):
         """Validate configuration parameters"""
-        assert self.num_jugs > 2, "edge_probability must be gt 2"
-        assert self.difficulty > 0, "edge_probability must be gt 0"
-        assert self.difficulty < 200, "edge_probability must be gt 200"
+        assert self.num_jugs > 2, "num_jugs must be gt 2"
+        assert self.difficulty > 0, "difficulty must be gt 0"
+        assert self.difficulty < 200, "difficulty must be lt 200"
 
 
 class JugsDataset(ProceduralDataset):
diff --git a/tests/test_jugs.py b/tests/test_jugs.py
index dc5f0476..a8bb4615 100644
--- a/tests/test_jugs.py
+++ b/tests/test_jugs.py
@@ -7,7 +7,7 @@ from reasoning_gym.algorithmic.jugs import JugsConfig, JugsDataset
 
 def test_jugs():
     """Test basic properties and solution of generated items"""
-    config = JugsConfig(seed=42, size=10, num_jugs=3, difficulty=20)
+    config = JugsConfig(seed=42, size=1000, num_jugs=3, difficulty=5)
     dataset = JugsDataset(config)
 
     # easy
@@ -20,3 +20,31 @@ def test_jugs():
         # Test the scoring
         assert dataset.score_answer(answer=json.dumps(item["metadata"]["possible_answer"]), entry=item) == 1.0
         assert dataset.score_answer(answer=None, entry=item) == 0.0
+
+    config = JugsConfig(seed=42, size=1, num_jugs=3, difficulty=50)
+    dataset = JugsDataset(config)
+
+    # med
+    for item in dataset:
+        assert isinstance(item, dict)
+        assert "question" in item
+        assert "answer" in item
+        assert "metadata" in item
+
+        # Test the scoring
+        assert dataset.score_answer(answer=json.dumps(item["metadata"]["possible_answer"]), entry=item) == 1.0
+        assert dataset.score_answer(answer=None, entry=item) == 0.0
+
+    config = JugsConfig(seed=42, size=1, num_jugs=3, difficulty=99)
+    dataset = JugsDataset(config)
+
+    # hard
+    for item in dataset:
+        assert isinstance(item, dict)
+        assert "question" in item
+        assert "answer" in item
+        assert "metadata" in item
+
+        # Test the scoring
+        assert dataset.score_answer(answer=json.dumps(item["metadata"]["possible_answer"]), entry=item) == 1.0
+        assert dataset.score_answer(answer=None, entry=item) == 0.0