Keyzick - look at the comment above the 83.3% chance of having a 20% chance of achieving the "good" column. The "stand alone" comment.
I'll restate that a different way:
Columns duplicate, 16.7% chance of a 100% chance to get a "good" column. Maximum chance to achieve 16.7%
Columns do not duplicate, 83.3% chance to have a 20% chance to get a "good" column. Maximum chance to achieve 20%
The 16.7/83.3 chance split determines whether or not the column duplicates, not the outcome after that determination has been made. Essentially, the non duplicate column lineup is a bet against a duplicate roll.