I think this is a straw man version of Utilitarianism.

The right way to do Utilitarianism is to take probabilities into account - in which case we can predict that the mother who deliberately tried to harm her child will eventually succeed (and is therefore a threat), and the woman whose child falls by accident will likely learn to keep the window closed in the future (and is therefore NOT a threat).

You mention further down that Utilitarianism's reliance on prediction is a weakness. But every decision any human ever makes is always based on prediction. Just because the stakes may be higher in some scenarios that others, doesn’t mean there is any other option. Prediction is still our best and only option. So there is nothing wrong with Utilitarianism in this regard.

It seems to me that you are more interested in judging people as good or evil, rather than judging which actions will bring more good or evil (e.g. happiness or sadness) into the world. But it seems to me like judging people as good or evil is a pointless endeavor. The most you should do in terms of judging people is in determining whether they are a threat or not - not whether they are "evil" or not.