The accuracy of nso decreases after disrupting the order of samples in the source and target domains

[puppnn]

Hello, I am very interested in the nso method you proposed, but in the process of reproduction, I found that if I shuffle the samples of the source and target domains, or reorder the samples of either domain:
ids = randperm(size(Xs,1)); Xs=Xs(ids,:); Ys = Ys(ids,:); idt = randperm(size(Xt,1)); Xt=Xt(idt,:); Yt = Yt(idt,:);
This will lead to a significant drop in the accuracy of the nso method, or even failure. Theoretically, the data order of the source and target domains cannot affect the results of domain adaptation.
I found that the accuracy of the nso method is almost 100% when the label order of the target domain is consistent with the source domain label order.
This problem has been bothering me for a while, I don't know what is the reason for this, I hope your reply, thank you！