diff --git a/src/trace_link/kineto_operator.py b/src/trace_link/kineto_operator.py
index 46a45770..7bbf2d05 100644
--- a/src/trace_link/kineto_operator.py
+++ b/src/trace_link/kineto_operator.py
@@ -138,10 +138,19 @@ def is_gpu_op(self) -> bool:
         gpu_categories = {"kernel", "gpu_memcpy"}
         return self.category in gpu_categories
 
-    def is_arrow_op(self) -> bool:
+    def is_ac2g_op(self) -> bool:
         """
         Check if the operator is categorized as 'ac2g', which stands for arrows from CPU to GPU.
 
+        Excerpt from https://pytorch.org/docs/stable/torch.compiler_profiling_torch_compile.html
+        ```
+            Every kernel on the GPU occurs after being launched by code running on the CPU. The profiler can draw
+            connections (i.e. "flows") between the GPU and CPU events to show which CPU event launched a GPU kernel.
+            This is particularly helpful because, with a few exceptions, GPU kernels are launched asynchronously.
+
+            To view a flow connection, click on a GPU kernel and click "ac2g".
+        ````
+
         Returns
             bool: True if the operator is an 'ac2g' type, otherwise False.
         """
diff --git a/src/trace_link/trace_linker.py b/src/trace_link/trace_linker.py
index 0e2372bd..d9dc6c1c 100644
--- a/src/trace_link/trace_linker.py
+++ b/src/trace_link/trace_linker.py
@@ -244,7 +244,7 @@ def construct_kineto_data_structures(self, kineto_ops: List[KinetoOperator]) ->
                 kineto_gpu_ops.append(op)
                 logging.debug(f"Added GPU op: {op.name}")
 
-            elif op.is_arrow_op():
+            elif op.is_ac2g_op():  # arrow from CPU to GPU
                 assert (op.phase == "s") or (op.phase == "f")
                 if op.id is None:
                     error_msg = (