we just deployed a fix that was causing a small percentage of requests to our inference service to degrade

should seem improved responsiveness