If p/q is a close approximation to π, then cos(p), taken in radians, will be close to +1 or -1, because it implies that p is close to an integer multiple of π, and therefore cos(p) is close to the cosine of an integer multiple of π:
p/q =~ π
p =~ qπ
cos(p) =~ cos(qπ)
In turn, the cosine of an integer multiple of π is +1 or -1 because the period of cosine (as a trigonometric function) is 2π radians, with maxima at 0, 2π, 4π, ... and minima at π, 3π, 5π, ...
Could someone elaborate? First, explain the one example from the article (cos(355)) and perhaps show some other oddities?