[openai-blog] A shared playbook for trustworthy third party evaluations