GPT-4.1 Structured CoT Audit: Query, Context, Synthesis

For basic Chain-of-Thought in GPT-4.1, OpenAI recommends placing a step-by-step instruction at the end of the prompt to print document titles and IDs first. To fix systematic eval failures (such as intent misunderstanding or insufficient context analysis), upgrade to explicitly structured sections for Query Analysis, Context Analysis, and Synthesis.

First, think carefully step by step about what documents are needed to answer the query inside these three sections:
1. ### Query Analysis: Define the exact legal question and scope.
2. ### Context Analysis: Print out the TITLE and ID of each relevant document, then format the IDs into a verified list.
3. ### Synthesis: Answer the query using strictly the listed document IDs.