A multi-format supervised fine-tuning (SFT) corpus of reasoning traces covering six
domains: math, coding, reasoning, agentic, tool use, and instruction following.
Each trace is a structured step-by-step solution followed by a final answer, with
explicit intermediate steps and verification.