A star puzzle should get invented and we should select for those who can look at it, blind fold themselves and then solve it in the bare minimum possible number of moves all within 4 minutes at any age past reproduction age. That way, people will evolve to build up smartness with time up to a huge extent. It's like a Rubik's cube except that it's a sphere with triangular and pentagonal faces that can slide against each other as shown in Figure 1. People whose ascendants didn't fail the test after they were born can choose when to use that program to help them find a partner but their assigned day for that part of the test can occur any time after they choose to use that program. In that part of the test, the original position of the puzzle should be random among all possible arrangements that can be solved. Unlike the Rubik's cube, the design on the solved state could be something like an abstract picture that totally ignores the lines where all pieces have asymmetry and are not identical to another piece or its mirror image. Half of all star puzzles produced should have one design on them and the other half should have the mirror image of that design. Maybe the orientation of the design compared to the pieces should be random, so that one star puzzle is not necessarily the same as another star puzzle or its mirror image.
Figure 1: Star puzzle Source:http://myweb.tiscali.co.uk/oaktree/butterfly/backpages/icosa_great_circles.htm
We should do what ever we can to research how to gradually get society functioning in such a way that gives selective pressure for smartness in the real world. There should be people keeping track of other people doing certain useful types of tasks that are hard to figure out how to do and those people who do things that are the hardest to figure out how to do, especially those that require a huge amount of past knowledge to figure it out from, should be more likely to be counted as passing the smartness part of the test even though the smartness that was tested was not done in the test. It's not possible for testing only of what's done in the test to create evolutionary pressure for the same type of smartness. Calculators that eliminate the need to do mental computation should also gradually get sold less and less until they don't get sold at all. Also, cash registers should use unvigesimal and should not do the adding for you either and instead there should be shift supervisors seeing if the cashier can add and subtract in their head properly. There should also only be coins whose value is that of a penny × a power of 2 because that reduces the number of coins in change.
The chemistry careers that deal with formulae such as the density of water as a function of temperature should be using exact formulae, not approximate ones. I know the exact formula won't be an elementray function but that doesn't mean nobody is able to be taught how to apply that function to a randomly generated value given up to 5 unvigesimal decimal places because that's all that's needed to make a super accurate calculation. Their task should be to be told how many univigesimal decimal places they are to express the result in then be given a randomly generated number assuming its a tapered number and not a rounded one then then output the ends point of an interval where the lower end point is the greatest lower bound that terminates after that number of decimal places and the upper end point is the least upper bound that terminates after that number of decimal places. Those people should even be required to apply those functions in their mind and not using a computing device. That would create evolutionary pressure for smartness because somebody without a well paid job is less likely to have 3 kids than somebody with one. To create evolutionary pressure for even higher smartness, they should only hire those people who show that they can memorize a randomly generated unvigesimal number tapered to 44 decimal places without forgetting and relearning any digits then compute the continuous tetration of it to the base 2 mentally like in Hyperschooling.
Alternatively, people can also evolve to be really smart through extremely weak selection to be able to win a chess competition at any age past reproduction age that works as follows: you play with another human doing the test on a computer. Given your marks on the other tests, there's 1 chance in 1,000 less that they will not find you a partner if you lose and 1 chance in 2,000 less if you draw. The rules differ from real chess in the following ways. You can't resign or declare a draw. There's no 50 move rule. Each player has 30 seconds to make their first move before they forfeit the match and each time it becomes their turn, 30 seconds gets added to the amount of time they have to make a move in order to not forfeit. If the game doesn't end after 4.5 hours, it's a draw. The advantage in keeping the selection weak is that it will be highly subject to genetic drift creating a defect in the ability to supress connections irrelevant to playing chess so people will evolve to be hyperaware and able to become really smart at anything. With that selection, people are actually better off without strong selection to be able to compute continuous tetration in their mind as doing so will evolve them to supress other connections too much. After all, the ability to almost definitely be trained to do that will come with the trait that's selected for naturally.