I received a response in the developers forum which basically said the input data is scaled logarithmically rather than the output being linear which is then scaled logarithmically.
This means it's impossible to answer the question.
However I have asked (and repeated the question here for the sake of anyonewho's interested)...
What is the distribution of DA values across all values? It would be nice to know that "the median DA across all sites in our database is x." That would at least put the numbers in some perspective - and it's perspective I'm trying to get.
Can you also confirm if the "keyword difficulty" is also calculated with logarithmic inputs? And what's the median keyword difficulty?