An online prediction request is denied although the endpoint is active. What should be checked first?

1. Whether the training accuracy exceeds 90%
2. Caller authentication and permission to invoke the model deployment
3. Whether the client uses the same charting library as training
4. Whether the model name contains production