My bet is that the problem was much worse and more subtle than the other commenter's have noted. They posit a slight imbalance to due to rounding or PRNG chara

My bet is that the problem was much worse and more subtle than the other commenter's have noted. They posit a slight imbalance to due to rounding or PRNG characteristics. I suspect that is it like the problem of choosing values randomly from a sparse hash table. Choosing a random slot and scanning forward to the next empty slot results in a bias where closely spaced entries have a much lower chance of selection than entries preceded by a wider gap. It is easy to make that same mistake when trying to pick a random record number when record numbers have been assigned in chunks (like social security numbers).