diff --git a/python/sglang/srt/disaggregation/prefill.py b/python/sglang/srt/disaggregation/prefill.py
index 585034780..b27c4ec71 100644
--- a/python/sglang/srt/disaggregation/prefill.py
+++ b/python/sglang/srt/disaggregation/prefill.py
@@ -659,18 +659,29 @@ class SchedulerDisaggregationPrefillMixin:
             return
         page_size = self.token_to_kv_pool_allocator.page_size
         for req in batch.reqs:
-            if getattr(req, "disagg_kv_sender", None) is None:
-                continue
-            if getattr(req, "is_chunked", 0) != 0 or req.start_send_idx != 0:
-                continue  # chunked / partially-sent: this forward isn't the full range
-            if len(getattr(req, "prefix_indices", []) or []) != 0:
-                continue  # cached prefix is HiCache-loaded, not forward-written (lever B)
-            end_idx = min(len(req.fill_ids), len(req.origin_input_ids))
-            page_indices = _kv_locs_to_page_indices_cpu(
-                self.req_to_token_pool.req_to_token[req.req_pool_idx, 0:end_idx],
-                page_size,
-            )
-            kv_manager.register_per_layer_transfer(req.bootstrap_room, page_indices)
+            try:
+                if getattr(req, "disagg_kv_sender", None) is None:
+                    continue
+                if getattr(req, "is_chunked", 0) != 0 or getattr(req, "start_send_idx", 0) != 0:
+                    continue  # chunked / partially-sent: this forward isn't the full range
+                prefix = getattr(req, "prefix_indices", None)
+                # NB: prefix_indices is a tensor — never use `or`/truthiness on it.
+                if prefix is not None and len(prefix) != 0:
+                    continue  # cached prefix is HiCache-loaded, not forward-written -> lever B
+                end_idx = min(len(req.fill_ids), len(req.origin_input_ids))
+                page_indices = _kv_locs_to_page_indices_cpu(
+                    self.req_to_token_pool.req_to_token[req.req_pool_idx, 0:end_idx],
+                    page_size,
+                )
+                kv_manager.register_per_layer_transfer(req.bootstrap_room, page_indices)
+            except Exception as e:
+                # Never let lever-A setup crash the scheduler; fall back to the
+                # monolithic post-forward transfer for this request.
+                logger.warning(
+                    "[CP_PER_LAYER_TRANSFER] register skipped for room %s: %r",
+                    getattr(req, "bootstrap_room", None),
+                    e,
+                )
 
     @torch.no_grad()
     def event_loop_normal_disagg_prefill(self: Scheduler) -> None: