colab data mismatch

When I ran the colab, I found the input data did not fully match the variants recommended in previous rounds of model prediction. For example, the input data for round 2 do not cover all the top 12 variants predicted in round 1.  Is the model on the colab page different from the released version, and  therefore in need of additional tuning?

Thanks for helping!