# Thinking Journal: Observability and Cost

Use these prompts after the quiz and code drills.

1. Which trace fields would you require before a RAG feature can enter beta?
2. What is one route where you would pay for a stronger model, and why?
3. What is one route where caching or compression is safer than a model upgrade?
4. Which dashboard slice would reveal a failure hidden by averages?
5. What cost change would you block even if answer quality stayed flat?