Bayes factor for direational hypothesis

Has any thoughts or experience on the suitability of using arviz.plot_bf() for a directional hypothesis. It works for a non-directional model eg H1 is not equal to a ref_value but I think it fails for a directional hypothesis say: H1 >ref_val. The arviz web site does not specify that it shouldn’t be used in the directional case. I think It fails because when obtaining the ordinate at the ref value ,(to apply the Savage Dickey method) the kde approximation at the ref_val is rounded at the truncated half distribution for a directional model eg a half normal or half Cauchy begining at the ref_val,representing the alternative model. Any help on clarifying the correct use of the arviz function?