I missed this earlier, but it's not a good argument.
It's all about risk-adjusted expected value. There's a distribution of payoffs to spoofing, and it skews far to the right. The conceptual categories of "free" vs "not-free", or "risky" vs "riskless" is the wrong mental framework for evaluating this problem. Yes, it's not risk-free, but why is that relevant?
This is just a restatement of the thing that I was questioning.
Yes, it's real according to the definition that you've put out. The spoof order can be traded against. So? Why does this binary real/not real categorization matter for whether spoofing should or shouldn't be allowed?
An intelligent spoofer isn't going to place large orders directly on the BBO when the fill probability is high. They're going to be doing it when the correlated markets aren't moving against their order, for instance. Or they're going to put it one tick below the BBO and algorithmically pull when a trade hits the level above. So even if they do get filled, which is unlikely if they're smart, it's no big deal since they've made lots of money up until that point and they can get out of the unwanted position without much slippage.
