Back then my family kept a dog to guard the yard. My dad trained it to bark only when an unfamiliar person crossed the fence, and not bark when familiar people came and went, no matter what the schedule was. The difficult part wasn’t teaching it to bark—it was teaching it to distinguish, because a dog that barks indiscriminately is more troublesome than not having one, while a dog that stays silent even when a thief slips in is useless.
Today’s crypto wallet is like a dog that only knows how to bark one way: if the key is correct, it stays silent; if the key is wrong, it stays silent too, because there’s no one asking. It can’t tell whether the “stranger” is holding the correct key because the thief managed to steal it, or whether the homeowner is opening the door at midnight due to an urgent matter. With wallets, those two situations look identical.
It’s like retraining that dog: not only to ask who’s holding the key, but also to ask whether this context matches the usual situation before letting a transaction go through smoothly. A policy layer in between learns what “normal” looks like for each wallet and each funding flow—so it knows when to bark.
Self-refutation: but our household dog is trustworthy because my dad trained it to a fixed standard—everyone knows what that standard is: bark at strangers, don’t bark at familiar people. A behavior “policy layer” is much more ambiguous: if that “normal” logic changes based on data, no one can be sure whether tomorrow it will bark or not bark at the exact same action. A smart but unpredictable system could even be more dangerous than a dumb one—yet one that is certainly wrong.
So the real difficulty of @NewtonProtocol isn’t teaching the system to recognize abnormality. It’s keeping that definition of “abnormal” stable, transparent, and something users can look up to see why they were blocked today even though yesterday they did the same thing and it was fine.
I’m still waiting to see that part more clearly in $NEWT , before believing this dog barks the right person and not based on the trainer’s mood.