The averages are over different questions
A criterion with no facts behind it is not scored here, and it is not filled in with a default either. So a bookmaker whose lobby nobody has counted is not scored on the size of its lobby, and its total is an average over everything else.
That is the right way to score. It also means two totals in the same column can be averages over different sets of questions, and printing them to one decimal place implies a comparison the arithmetic does not support.
The perverse result it produces
Run the model over this list and the book that comes last is the only one whose record can answer every question the model asks. Its lobby has been counted, so it is scored on its lobby. The others are not scored on theirs at all.
Publishing that order would reward the bookmakers that are hardest to read and penalise the one that is easiest, and it would do it under a heading that says the opposite. That is not a ranking of Ghanaian bookmakers. It is a ranking of how much anybody managed to find out about them.
What replaces it
Membership, and the two facts that decided it, printed on every row. A reader can check the rule against the roster in about a minute, which is more than any decimal place has ever let anybody do.
Where a record cannot answer part of the model, its own page says which part and what share of the model that is.