btc-reasoning-traces-gpt5.2

Summary

A 2,020-row dataset of LLM reasoning traces for BTC/USD trading decisions on the 1-hour timeframe, where each row captures account state (exposure, position, PnL, leverage), 5-minute and 1-hour candle market data, a trading action (0=short, 1=flat, 2=long), and the model's chain-of-thought reasoning analyzing trends and price action. It is designed for reinforcement learning–based trading agents and is part of the TorchTrade project.

Keywords

trading bitcoin btc reasoning-traces llm reinforcement-learning finance torchtrade

Topics

Finance / Reinforcement Learning

Research notes

• Generated with GPT-5.2 (per dataset name). Each row includes system_prompt, user_prompt, market_data_5min, market_data_1h, account_state, action, and thinking fields.