After developing a predictive market automated trading system using Claude, I discovered some insights: Predicting the 'probability' in the market is not a probability at all. Why do you say that? It is market pricing, not statistical probability. The price reflects the result of the financial game, not the actual possibility of the event occurring. Prices can be distorted by emotions, funding levels, and information gaps. A 60% contract may have a real probability of only 40% or 80%. So what beginners should learn most is not to look at candlesticks, but to look at information - who is buying, why they are buying, and what the information source is. After actual testing, there are still several discoveries: There is basically no room for conventional arbitrage. The market size is already large enough, and the price difference has been quickly smoothed out by the bricklayers. Risk return still corresponds. To earn high returns, one must bear the risk of information asymmetry or time costs. In terms of airdrop potential, Polymarket is not as good as Opinion. The latter is still in its early stages, with greater opportunities. Another painful lesson: Do not misuse open source code. There are many pitfalls, it's better to write it yourself. Ultimately, predicting the market is not a casino, it's an information war. Your information advantage determines your profit limit.