The raw inference request body in OpenAI-compatible JSON format.
The job ARN that identifies which model session to route the inference request to.
The trajectory ID for grouping turns into a single rollout. Each turn (prompt and response) is captured for later use.