Just because it turned out poorly doesn’t make it a bad rule.
That kinda depends on the point of view.
If you take the frequentist approach and think about limits as n goes to infinity, sure, a single data point will tell you very little about the goodness of the rule.
But if it’s you, personally you, who is looking at the business end of a gun, the rule indeed turned out to be very very bad. I think the quote resonates quite well with this.
Besides, consider this. Let’s imagine a rule which works fine 99% of the time, but in 1% of the cases it leaves you dead. And let’s say you get to apply this rule once a week. Is it a good rule? Nope, it’s a very bad rule. Specifically, your chances of being alive at the end of the year are only 0.99^52 = about 60%, not great. Being alive after ten years? About half a percent.
That kinda depends on the point of view.
If you take the frequentist approach and think about limits as n goes to infinity, sure, a single data point will tell you very little about the goodness of the rule.
But if it’s you, personally you, who is looking at the business end of a gun, the rule indeed turned out to be very very bad. I think the quote resonates quite well with this.
Besides, consider this. Let’s imagine a rule which works fine 99% of the time, but in 1% of the cases it leaves you dead. And let’s say you get to apply this rule once a week. Is it a good rule? Nope, it’s a very bad rule. Specifically, your chances of being alive at the end of the year are only 0.99^52 = about 60%, not great. Being alive after ten years? About half a percent.