Using the original GitHub from autoresearch, I modified it to fit my requirements for a BTC 15min high-frequency self-evolving strategy research (essentially just searching). I set a few research topics: 1. SMC theory / momentum theory / linear deduction 2. Each theory can self-adjust bars (candlesticks) on the corresponding timeframe 3. Year-phase backtest set + locked prediction year set = to prevent overfitting (of course, there are other metrics too) Ran over 100 experimental iterations, pretty interesting.