A developer is diagramming the Copilot Chat request lifecycle. Which sequence correctly distinguishes preparing the request from generating and presenting a response?

1. Run inference on an empty request, then add repository context only while formatting the result.
2. Format a completed response, retrain the model on it, then collect the user request.
3. Collect the user request, execute every suggested command, then ask the model what output to display.
4. Combine the request with relevant context, run model inference, then format the generated response for the user.