用 Jev 工程分层管理预测市场策略研究

hammertime @hammertime_one原推发布于 2026.10.11

在 X 上打开原推

A trader placing $60 bets and a hedge fund managing $10 billion can be trying to exploit the exact same thing: a price that doesn't match the probability of an outcome. The difference is usually in the research. And that's exactly what I'd use Horizon for: testing whether a trading idea actually holds up against historical data before putting money behind it. Take a short-term BTC Up/Down market. The market prices the Up outcome at $0.58. Your model estimates a 64% probability. That's a theoretical edge of six percentage points. A trading bot can buy the outcome, but the existence of that gap doesn't automatically make it a good trade. Your model might be wrong. The order book could be too thin. The probability could change while your order is being filled. And the six-point edge might disappear after execution costs. This is where I'd spend my time with Claude. Not asking it to predict whether Bitcoin will go up. I'd have it investigate how the probability model behaves across different market conditions. Does momentum matter more with 10 minutes left? Does volatility make the model overconfident? How often does a six-point theoretical edge produce a profitable trade after fees? Then use JEV engineering to manage the research process. Cheap calculations handle the repetitive probability updates. The expensive model gets involved when the results are ambiguous or a hypothesis needs to be redesigned. Horizon fits into this workflow at the research stage, helping test the historical behavior of a strategy before deciding whether it's worth trading. A $60 trade doesn't need a $10 billion research department. But it still needs a reason to exist beyond "the chart looks bullish."

这条是引用着另一条写的：x.com/hammertime_one/status/2108916760673964533