It could be a measure of each candidate's difference (in diversity points) from the overall cohort, which is an approximation of how much the diversity of the group will change by accepting/rejecting this candidate.

But that seems hard to implement - you'd have to recalculate this for every candidate every time you got a new candidate - so perhaps we can implement an optimization: approximate this by estimating the distribution of the group and measuring differences from some centroid.

HR teams, I'm available for techwashing consultancy.

> HR teams, I'm available for techwashing consultancy.

Nice! Especially if coupled with a secondary business whereby you train applicants on how to increase their "diversity points"

EDIT: I think I have the appropriate "one weird trick": flood the pool with fake applications grouped together but very dissimilar to you.