Yes, as you can see from the VDBE code, that is using 2 columns for the index scan and doing no subsequent "culling" of the candidates obtained from the index scan. This is different from your original example where the VDBE code shows that only a ONE column index was being used for the index scan and that TWO constraints were being used to cull the candidates from the index scan. This would imply that although you thought you had a usable index on all three columns, that you did not and that in fact the only usable index was a single column index. Since we cannot see your schema which likely does not contain an index that is more useful than the one chosen, we are not able to help.