Implement Extended Thinking with GPT-5

For the 'R&D Impact Assessment' and 'Recommendation Generation' nodes, implement extended thinking patterns (e.g., Chain-of-Thought, Tree-of-Thought) using GPT-5. The agent should be able to iteratively refine its understanding and arguments, simulating a human expert's thought process. Incorporate an adaptive reasoning budget to control GPT-5's token usage based on the complexity of the current task.