Question about global permutation test

[sealx017]

Hello, I noticed that the package compares the actual global I with the permuted global I's to compute the permutation p-value:

    adata.uns['global_stat']['perm']['global_p'] = 1 - (adata.uns['global_I'] \
                         > adata.uns['global_stat']['perm']['global_perm'].T).sum(axis=0) / n_perm

My question is why do you compare only one direction (">"), i.e., greater than the mean under NULL? Global I can be negative, i.e., less than the mean under NULL indicating negative spatial association.

Please let me know if I misunderstood the package and the paper!