I wouldn’t even be thinking of blame, but of how to prevent a similar error in future. Would a test have flushed it out? Some progress or thread assertions? Better documentation? A different API?