Else comparisons not penalizing enough

[nyc-de]

Hi,

I've been using Splink for deduplication on a couple of projects now and one thing that has continued to trouble me is that lack of penalty in the match weights in the "else" result. I have attached a screenshot of my model which illustrates my issue.

I would like some of the other comparisons to have a penalty more along the line of what you see in the "custom_first_last" comparison. Until now I have achieved this by meddling with the m and u weights in the model settings json file.

The data I am working with is person data with the usual address, phone and some geo location fields. This dataset is only 50k and from what I can tell there aren't many duplicates.

My question is is there something I can do while training the model to achieve my desired result?