Yes, “amount of melanin content” is a real thing. And it is in large part attributable to genes which have correlations with both other genes (with medically relevant differences) and to geographic locations of ancestors.
So, if by “race” one means “melanin content and geographic location of ancestors”, then that is a real thing. But the way these vary do not come in any particular natural grouping into categories; “race”, in the sense of a categorical variable (not just the concept of ancestry more generally), isn’t real, or, at least, it is less real than the distinction between male and female.
If one has a random real-valued variable which has a continuous distribution, it may be useful to chunk it up into discrete buckets even if the way of chunking ends up being fairly arbitrary. Having a way of chunking it up into intervals being useful doesn’t imply that the arbitrary chunking is actually a natural categorical variable.
So, if by “race” one means “melanin content and geographic location of ancestors”, then that is a real thing. But the way these vary do not come in any particular natural grouping into categories; “race”, in the sense of a categorical variable (not just the concept of ancestry more generally), isn’t real, or, at least, it is less real than the distinction between male and female.
If one has a random real-valued variable which has a continuous distribution, it may be useful to chunk it up into discrete buckets even if the way of chunking ends up being fairly arbitrary. Having a way of chunking it up into intervals being useful doesn’t imply that the arbitrary chunking is actually a natural categorical variable.