What does the function FindGraphPartitions do?

The documentation doesn't explain this function clearly. I can see that this is the key:

FindGraphPartition[g,k] gives a partition of vertices into k approximately equal-size parts.

FindGraphPartition finds a partition of vertices such that the number of edges having endpoints in different parts is minimized.

But that's not a precise definition. What is "approximately equal-size"? What is the quantity being optimized here? If that's not easy to explain, then what algorithm does it use?

I have asked support about this more than once and received no useful response. I know others have asked too and they also received no response.

A function like this is not useful except as a toy because I don't even know what it does (not only how it does it).

The same can be said about FindGraphCommunities too. The description is extremely vague. In that case however, I am familiar enough with the topic that I can guess what the different methods actually do. However, this is only a guess (even if it's a confident one), so no one can use the result form these functions in a peer-reviewed publication where reproducibility is required. The fix would be so easy: just put in references for each community detection method. E.g. Method -> "Centrality" is probably based on this paper. Why does't Wolfram do it then?

I know that many people complain that Mathematica is not open source and we can't know how it computes e.g. integrals (and oh the horror, it even gives wrong results sometimes!). I hope it's clear enough that what I am talking about here is something entirely different and much more serious. I don't have to know how to compute an integral to know what an integral is. But what FindGraphCommunity does is defined by the algorithm (i.e. how it does it) and without a proper description the function is simply useless. In the case of FindGraphPartitions I literally cannot even figure out what it actually does!