Random and expensive relay behavior

Add some debug nodes to your flow, set to send the output to the console. Give the nodes names so they can be easily identified in the output. Then when it goes wrong you can look at the log and work out where it is failing.