Inference requests remain constant, but total cost rises. Prompts now include more retrieved content and answers are longer. Which telemetry should be examined?

1. Input and output token consumption and model/deployment cost dimensions
2. Only the count of application routes
3. Only storage filenames
4. Only developer headcount