diff --git a/python/sglang/srt/disaggregation/decode.py b/python/sglang/srt/disaggregation/decode.py
index 0ddccdf46..7f2de88cb 100644
--- a/python/sglang/srt/disaggregation/decode.py
+++ b/python/sglang/srt/disaggregation/decode.py
@@ -129,15 +129,17 @@ class DecodeReqToTokenPool:
         return len(self.free_slots)
 
     def alloc(self, reqs: List["Req"]) -> Optional[List[int]]:
-        chunked = [i for i, r in enumerate(reqs) if r.req_pool_idx is not None]
+        # Indices of reqs that already have a req_pool_idx and will reuse
+        # their existing slot (e.g. chunked prefill continuing across chunks).
+        reusing = [i for i, r in enumerate(reqs) if r.req_pool_idx is not None]
         assert (
-            len(chunked) <= 1
+            len(reusing) <= 1
         ), "only one chunked request may reuse req_pool_idx in a batch"
         assert all(
-            reqs[i].is_chunked > 0 or reqs[i].kv_committed_len > 0 for i in chunked
-        ), "request has req_pool_idx but is not chunked"
+            reqs[i].is_chunked > 0 or reqs[i].kv_committed_len > 0 for i in reusing
+        ), "reusing request must be chunked or have committed KV"
 
-        need_size = len(reqs) - len(chunked)
+        need_size = len(reqs) - len(reusing)
         if need_size > len(self.free_slots):
             return None
         select_index = self.free_slots[:need_size]
diff --git a/python/sglang/srt/mem_cache/memory_pool.py b/python/sglang/srt/mem_cache/memory_pool.py
index 10b9d9b49..5312b2edb 100644
--- a/python/sglang/srt/mem_cache/memory_pool.py
+++ b/python/sglang/srt/mem_cache/memory_pool.py
@@ -153,16 +153,18 @@ class ReqToTokenPool:
         return len(self.free_slots)
 
     def alloc(self, reqs: list[Req]) -> Optional[List[int]]:
-        chunked = [i for i, r in enumerate(reqs) if r.req_pool_idx is not None]
+        # Indices of reqs that already have a req_pool_idx and will reuse
+        # their existing slot (e.g. chunked prefill continuing across chunks).
+        reusing = [i for i, r in enumerate(reqs) if r.req_pool_idx is not None]
         if not any(r.is_dllm() for r in reqs):
             assert (
-                len(chunked) <= 1
+                len(reusing) <= 1
             ), "only one chunked request may reuse req_pool_idx in a batch"
         assert all(
-            reqs[i].is_chunked > 0 or reqs[i].kv_committed_len > 0 for i in chunked
-        ), "request has req_pool_idx but is not chunked"
+            reqs[i].is_chunked > 0 or reqs[i].kv_committed_len > 0 for i in reusing
+        ), "reusing request must be chunked or have committed KV"
 
-        need_size = len(reqs) - len(chunked)
+        need_size = len(reqs) - len(reusing)
         if need_size > len(self.free_slots):
             return None
         select_index = self.free_slots[:need_size]