diff --git a/temporal-sdk/src/test/java/io/temporal/internal/worker/AsyncPollerTest.java b/temporal-sdk/src/test/java/io/temporal/internal/worker/AsyncPollerTest.java
index 246ea4b6df..fdcd6ce2ec 100644
--- a/temporal-sdk/src/test/java/io/temporal/internal/worker/AsyncPollerTest.java
+++ b/temporal-sdk/src/test/java/io/temporal/internal/worker/AsyncPollerTest.java
@@ -379,14 +379,17 @@ public SlotSupplierFuture reserveSlot(SlotReserveContext<SlotInfo> context)
     poller.resumePolling();
     assertFalse(poller.isSuspended());
     pollLatch.await();
-    assertEventually(
-        Duration.ofSeconds(5),
-        () -> {
-          assertEquals(0, executor.processed.get());
-          assertEquals(1, slotSupplierInner.reservedCount.get());
-          assertEquals(0, slotSupplier.getUsedSlots().size());
-        });
-    // Suspend polling again, this will not affect the already issued poll request
+    assertEquals(0, executor.processed.get());
+    assertEquals(1, slotSupplierInner.reservedCount.get());
+    assertEquals(0, slotSupplier.getUsedSlots().size());
+    // Wait for iter 2 of the poll loop to invoke reserveSlot. Completion of that reservation
+    // requires iter 1's slot to be released (via completePoll below) since the supplier has
+    // only one slot — but the *call* to reserveSlot happens as soon as iter 2 passes the
+    // suspend-latch check. Waiting on that call here removes a race where the suspendPolling
+    // below could otherwise strand iter 2 at the suspend latch before it reserves.
+    assertEventually(Duration.ofSeconds(5), () -> assertEquals(2, reserveCalls.get()));
+    // Suspend polling again, this will not affect the already issued poll request or the
+    // iter-2 reserveSlot call above (which is now waiting for the slot to free up).
     poller.suspendPolling();
     completePoll.get().apply();
     assertEventually(