Mostly right, though it depends enormously on which market you are looking at. A deep question with real size behind the price almost certainly knows things the model does not. A thin question with fifty dollars of lifetime volume is closer to noise than to a forecast, and there a decent model can genuinely be the better estimate. The disagreement is only informative in proportion to the liquidity standing behind the price. Without that qualifier the rule cuts the wrong way about half the time.

and the thin ones are exactly where the model looks smartest. selection bias all the way down

补充一点，两边的口径经常也对不上。模型算的是某个事件在某个时间窗口内发生的概率，而合约条款里可能还叠了别的限定条件。这时候价差里有一部分其实是条款差异，不是观点分歧。看到大价差先去核对结算规则，比直接下注理性得多。

对 先看条款 再看深度 最后才轮到观点