Ideally, yes everyone should come from 10s, 11s and 12s when they want to go to A80, C90, D10, N90, etc. because those backgrounds show that you've been able to work a significant volume of traffic and/or complexity. Even then, the difference between some level 11s and 12s is staggering in regards to traffic volume (Example: A80 did 1.2M ops/903K to ATL as a level 12 in 2016 while A90 did 640K ops/409K to BOS as a level 11 in 2016). You should start at a 5-8, then move up to a 8-11, then move on to a 10-12. Your upward mobility will be faster (say level 8 to 11 to 12) based on ability where a slower learner might need to move multiple times to potentially become a level 12 controller (say level 5 to 8 to 10 to 12).
I'll look at the historical success rate (2010-present is most readily available on the weekends) and let you know what the difference in success rates are between level 7s and 8s.
What's a currently accepted better method than historical success rates?!? Unfortunately, this is Minority Report where we can already see whether someone is going to certify or not and not bring in the people who won't make it. The last report I saw, if you came from a level 8 or lower your success rate was less than 24% and if you came from a level 9 or higher it was 64%. THAT is a HUGE difference. Hell, under a thinking other than using historical success rates, we might as well never stop taking tower only controllers because there might, might be one to make it eventually and that one person might feel hurt because they weren't given the opportunity to go from a tower only to a level 12 tracon. Success depends on a lot of stuff (the individual and their drive to be successful is most important), but you have to bring in the people who have the best chance to certify, especially in a critical situation, because we can't just look at someone and know whether they are going to certify or not. If it was that easy, it would make staffing significantly better.