Quoted By:
Start with the obvious patterns.
The red rectangle and green rectangle are the same pattern, but going in different directions. The vertical pattern continues on column two, but it looks like the horizontal pattern changes.
No external pattern can be determined from a single rectangle. By having green pattern persistence it is possible that the vertical pattern should repeat on column three. If the vertical pattern continues that the black squares in it would be exactly the same as if Row 2 & Row 3 did follow the original red pattern.
So now we test the idea that matching squares cancel each other. The only way for a matching square to be determined is if it fits a pattern, both vertically and horizontally.
Why don't the bottom two squares in the second column cancel out? It is like this specifically to accent that columns have a pattern: an unchanging pattern. Well, that explains it in an ad hoc way. Now what does it actually look like if (2,3) is blank instead of black. No square is a pattern by itself. If both (2,3) and (3,2) were blank it would legitimize a pattern that the black square does disappear under the circle. This would permit two correct answers from the bank: B and F. There are no two correct answers. Both (2,3) and (3,2) can't be blank. This is why one is shaded in. This is why a square hiding under the circle is not valid because it permits two answers to the question.
Next question is why not have both middle and (2,3) blank. From a test making perspective, it makes random B guesses much more likely. Many people wouldn't think to self-persist a pattern for both columns and rows just to cancel their resultant, though it does make sense giving no additional information to go on. So it would make the test simultaneously harder and more people getting the right answer. This is not ideal. I was not able to find an alternate solution to hard exclude it. I think a unique self interference pattern could be defined using only (1,1), (2,1), and (1,2).
