but b does not have to be constant, nor does it have to be random. It could be a linear or non-linear function of underlying voting fractions as Mick West uses in his model. There is such a function that will give you a horizontal line (within physically meaningful constraints). In that case, the downslope does have meaning relative to the horizontal line and it does indicate a change of votes relative to that.you will always get a downslope if b is constant; and you will get a downslope trend if b is random. And that's exactly what Dr Shiva did. So getting a downslope means nothing, no lost votes
Matt did not say "Dr Shiva's expectation is unrealistic", he said "This cannot represent a net movement of votes". I only have to find one possible case to disprove that and I've found one so his argument is shaky.