You have understood the primitive algorithm. You're right, it's not efficient. However, it is generalized as the thesis develops and removes all impossible values in the set of test subjects (you have to read the whole paper to understand this). Still, the fully developed algorithm is not all that efficient as a prime tester. The idea, however, is not to render a computationally efficient prime test so much as to stimulate and promote the idea that if natural number theory could be placed on some geometric palette, the key to primes might unfold.
