In general, to see what algorithm is producing better results, you need to either (A) have a dataset with some known differences or (B) perform an independent validation of some of the discordant results produced by the various packages.

I have a template for comparing those algorithms here with relevant description here.

One important limitation with the demo dataset is the small number of validated sites, but you can use the templates to get an idea of what the results would look like for your own data.