diff --git a/bioengine/cluster/proxy_actor.py b/bioengine/cluster/proxy_actor.py
index 713e275..1c57242 100644
--- a/bioengine/cluster/proxy_actor.py
+++ b/bioengine/cluster/proxy_actor.py
@@ -342,7 +342,7 @@ def _get_accelerator_type(self, resources: Dict[str, float]) -> Optional[str]:
         """
         for resource_name in resources:
             if resource_name.startswith("accelerator_type:"):
-                return resource_name.lstrip("accelerator_type:")
+                return resource_name.removeprefix("accelerator_type:")
 
     def _get_slurm_job_id(self, resources: Dict[str, float]) -> Optional[str]:
         """Extract the SLURM job ID from a node's resource dictionary.
diff --git a/tests/test_accelerator_type.py b/tests/test_accelerator_type.py
new file mode 100644
index 0000000..b06bcb1
--- /dev/null
+++ b/tests/test_accelerator_type.py
@@ -0,0 +1,26 @@
+"""Unit tests for ``BioEngineProxyActor._get_accelerator_type``.
+
+Exercises the method against plain resource dicts so no Ray cluster is needed.
+"""
+
+from bioengine.cluster.proxy_actor import BioEngineProxyActor
+
+_get_accelerator_type = (
+    BioEngineProxyActor.__ray_metadata__.modified_class._get_accelerator_type
+)
+
+
+def test_reads_the_accelerator_type_resource():
+    resources = {"CPU": 8.0, "GPU": 1.0, "accelerator_type:A40": 1.0}
+    assert _get_accelerator_type(None, resources) == "A40"
+
+
+def test_type_starting_with_a_prefix_character_survives():
+    # str.lstrip("accelerator_type:") strips the character *set*, so a type
+    # whose first characters all appear in the prefix loses them.
+    resources = {"GPU": 1.0, "accelerator_type:tesla": 1.0}
+    assert _get_accelerator_type(None, resources) == "tesla"
+
+
+def test_returns_none_without_an_accelerator_resource():
+    assert _get_accelerator_type(None, {"CPU": 8.0}) is None