diff --git a/configs/2b_480_640_agibot.yaml b/configs/2b_480_640_agibot.yaml
index b8989c2..6528d78 100644
--- a/configs/2b_480_640_agibot.yaml
+++ b/configs/2b_480_640_agibot.yaml
@@ -84,11 +84,6 @@ dataloader_train:
       - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0902_3239
       - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0903_3237
       - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0903_3248
-      - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228
-      - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228_lang
-      - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228_long
-      - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228_long_6k
-      - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228_test
       - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228_train
       - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228_train_lang_sep
       - /mnt/amlfs-03/shared/datasets/agibot-custom-converted-0829-fullres/agibot.0911_3228_train_lang_sep_max600
diff --git a/tests/test_config_no_test_leakage.py b/tests/test_config_no_test_leakage.py
new file mode 100644
index 0000000..f43b4a0
--- /dev/null
+++ b/tests/test_config_no_test_leakage.py
@@ -0,0 +1,45 @@
+from pathlib import Path
+
+
+def _load_dataset_paths(config_path):
+    """Extract dataset_path entries from a config YAML using only stdlib."""
+    paths = []
+    in_dataset_path = False
+    with open(config_path) as f:
+        for line in f:
+            stripped = line.strip()
+            if not stripped:
+                continue
+            indent = len(line) - len(line.lstrip())
+            if stripped.startswith("dataset_path:"):
+                in_dataset_path = True
+                continue
+            if in_dataset_path:
+                if stripped.startswith("- "):
+                    if indent > 4:
+                        paths.append(stripped[2:].strip())
+                    else:
+                        in_dataset_path = False
+                elif indent <= 4:
+                    in_dataset_path = False
+    return paths
+
+
+def test_no_test_splits_or_parent_overlap_in_configs():
+    config_dir = Path(__file__).parent.parent / "configs"
+    for config_path in config_dir.glob("*.yaml"):
+        basenames = [p.rsplit("/", 1)[-1] for p in _load_dataset_paths(config_path)]
+        base_set = set(basenames)
+
+        test_splits = [b for b in basenames if b.endswith("_test")]
+        assert not test_splits, (
+            f"{config_path}: test splits cannot be used for training: {test_splits}"
+        )
+
+        for b in basenames:
+            if b.endswith("_train") or b.endswith("_test"):
+                base = b.rsplit("_", 1)[0]
+                assert base not in base_set, (
+                    f"{config_path}: base dataset {base} overlaps with its split {b}"
+                )
+