diff --git a/packages/api/src/agents/subagentThreads.spec.ts b/packages/api/src/agents/subagentThreads.spec.ts
index 27fd1eb03a..cd78101cd3 100644
--- a/packages/api/src/agents/subagentThreads.spec.ts
+++ b/packages/api/src/agents/subagentThreads.spec.ts
@@ -1224,7 +1224,12 @@ describe('SubagentThreadTaskStore', () => {
         return methods.getMessages(...args);
       }),
     };
-    const options = { leaseTtlMs: 500, leaseHeartbeatMs: 50 };
+    /** The heartbeat fences itself when a renewal commits after the deadline it was
+     * issued against, so the TTL is this test's tolerance for runner jitter, not just
+     * its pace. Keep it well above the scheduling stalls a loaded shard produces: a
+     * half-second lease turns an ordinary pause into a lapse and the prepared run is
+     * refused instead of executed. */
+    const options = { leaseTtlMs: 2_000, leaseHeartbeatMs: 50 };
     const firstWorker = new SubagentThreadTaskStore(slowMethods, options);
     const secondWorker = new SubagentThreadTaskStore(methods, options);
     const config = buildSubagentThreadTaskConfig(firstWorker, { userId, parentConversationId });
@@ -1250,8 +1255,9 @@ describe('SubagentThreadTaskStore', () => {
     const heartbeatIndex =
       heartbeatCall == null ? -1 : intervalSpy.mock.calls.indexOf(heartbeatCall);
     const heartbeat = intervalSpy.mock.results[heartbeatIndex]?.value as NodeJS.Timeout | undefined;
-    const warningIndex = timeoutSpy.mock.calls.length - 1;
-    expect(timeoutSpy.mock.calls[warningIndex]?.[1]).toBe(5_000);
+    const warningCall = [...timeoutSpy.mock.calls].reverse().find(([, delay]) => delay === 5_000);
+    const warningIndex = warningCall == null ? -1 : timeoutSpy.mock.calls.lastIndexOf(warningCall);
+    expect(warningCall?.[1]).toBe(5_000);
     const warning = timeoutSpy.mock.results[warningIndex]?.value as NodeJS.Timeout | undefined;
     expect(heartbeat?.hasRef()).toBe(true);
     expect(warning?.hasRef()).toBe(true);