The best option is almost entirely random. That's the closest you're going to get to mimicking the NFL.
Look at a guy like Randy Moss. Bunch of good years with the Vikings, then pure CRAP in Oakland...then breaks records with the Patriots. How the heck do you represent that in any type of progression system that isn't hugely random?
Look at some RBs that ride the bench for 2 or 3 seasons, then the starter gets traded and they have a breakout season when they finally start. Again, how would you represent that in Madden? You'd have to have progression while the guy isn't even playing. And why not? Maybe the team traded the starter because they saw that this guy was possibly better.
Instead of worrying about potential ratings, progression is somewhat randomized and players have a consistency rating that affects it more. A player with a lower consistency rating might bounce around from year to year, while a guy with a higher consistency rating would hit a peak and stay there longer. There's players that come into the league and dominate almost immediately... and there's guys that are unknown for years.
Also, you could have a "value" rating tied into the stats that would affect player contract requests and trade interest/value, without changing physical ratings. That would help represent "system players" and the overhyped aspect of some players, and maybe help shut up the "I can run for 2500 yards with my 68OVR RB" guys.