We have a production service used by customers and a separate development environment where we run tests and experiments.
Our concern is that development experiments or high-volume tests could impact production through shared organization-level rate limits or billing/usage limits.
Project-level budgets and alerts seem to be soft limits, not hard spending limits.
What is the recommended best practice to create hard separation between production and development usage?
Should we use separate OpenAI organizations for production and development, or is there another recommended way to enforce hard limits per environment/project?