Yep, but it seems plausible these would be outweighed by the value she will create for others, assuming she eventually gets a job, pays taxes, etc. Assuming you think humanity is net positive value, absent some particular reason to think the child will be negative it seems reasonable to assume she will be positive.

I mostly agree, however there have to be scenarios in which another person is a net negative, if we're to avoid the repugnant conclusion. Moreover, in the case of abortion we're considering an undesired child which might mean psychological damage both to the child and the unwilling parent. To take an extreme example, consider pregnancy from rape.

Btw, I think temporal discount is necessary but it's not exponential as naively thought but linear wrt time of Big Bang. I can explain why if it's interesting.

According to UDT we have to compute the expectation value of the utility function over the whole "a priori" Tegmark IV multiverse (without conditioning on observations). A natural way to model this is considering the Solomonoff measure on the space of infinite sequences of bits (each such sequence is a "universe") . Thus the expectation value is an integral over all such sequences. It is natural to require the utility function to be bounded in order for the integral to converge (avoiding Pascal's mugger). Since each universe enters the integral together with its time translated versions, the resulting time asymptotics is 2^{-Kolmogorov complexity of t} which decays only slightly faster than 1/t. This result doesn't depend on the details of the time discount in the "bare" utility function: it is a universality result. See also this.