Our team's alpha development of the CoinKarma trading system is probably less than 10%. You heard right, less than 10%. I personally developed over a dozen strategies, but only two or three actually performed well. Finding patterns and identifying high-probability trading opportunities from massive amounts of data is incredibly difficult; turning it into a quantitative strategy is even harder. The bottleneck of manually developing strategies is that only one hypothesis can be tested at a time, and the search space is easily limited by one's own intuition and biases. Thinking of an idea, writing a strategy, backtesting, failing, thinking again—this cycle is too slow. So we're doing something new: letting AI do this. The entry point is outliers. When our indicators show extreme values or rare combinations, it indicates that the market is approaching some extreme state, and these moments are more likely to be entry opportunities. We have AI systematically scan all historical outlier moments, see what happened afterward, and find patterns that the human brain cannot exhaustively search. Previously, humans looked at the data and did post-mortem analysis. The system now directly accesses data from the past few years to automatically identify high-probability, high-expectation-value trading opportunities. How exactly does it do this? We use Claude Code in conjunction with a plugin called Ralph Loop. The process is as follows: Claude Code first performs an automated search to find high-probability and valuable indicator pattern combinations; once found, these are submitted to Codex for review using GPT 5.2; feedback is generated after the review and sent back to Claude Code for further iteration until the quality meets standards. This process can run continuously for four to five hours without interruption. This project started at the end of last year, but it coincided with the largest liquidation event in history on October 11th. I believed that the data patterns and price movements after such a massive crash were very valuable for research, so I decided to wait three months to collect data. Now, almost three months have passed, and the market has experienced multiple crashes, rebounds, and market corrections. This data is extremely valuable to us. This project is called CoinKarma Insight, and its ultimate goal is to enable the system to automatically identify patterns, allowing users to place orders automatically and eliminating the need for human review and manual order placement. Others vibe coding, we vibe trading.