The Goldbach Conjecture and large primes

Mathematicians like exact solutions. Engineers are satisfied with good approximations. Let's see if there's a pattern among the largest primes smaller than a certain even number

10 7

100 97

1000 997

2000 1999 

5000 4999 

10000 9973

100000 99991

The corresponding percents are 70, 97. 99.7, 99.95, 99.98, 99.73, and 99.99.

Thus, we can infer that for even numbers greater than 100,000, the largest prime smaller than that number will be about 99% of that number. 

So, if we want to find large primes, one way would be to multiply a large even number by 0.99 and using the Miller-Rabin primality test. There is another primality test I have worked on that uses the Collatz Conjecture. It works by using the suspected prime as the seed value for the Collatz Conjecture and then checking to see if the first odd result (that is not a multiple of 5) is prime. 

Another way would be to subtract small known primes from the even number and checking if the result is prime. 

In the cases of 10,000 and 100,000, the largest prime smaller than either does not form a Goldbach pair. So actually, what we really want to know is: what is the largest prime smaller than an even number that forms a Goldbach pair?

I suspect many such pairs contain 3, 7, 13, 17, etc as the smaller prime.