I don't think I worded my question right. Since the NET rankings adjust results according to opponent strength, there has to be some preconceived AKA "preseason" ratings in the database (even if NET rankings aren't released to the public). There has to be something telling the computer that a 10 pt win over Purdue is better than a 10 pt win over WIU.
This isn't necessarily true. RPI for instance just looks at a teams win percentage, average opponents winning percentage, and their opponents winning percentage. Basically, if you win against teams who win against teams who also win, you'll be ranked very highly. But if you win against teams who win against teams who lose, you will be ranked lower.
NET says they take into account "game results, strength of schedule, game location, scoring margin, net offensive and defensive efficiency, and the quality of wins and losses," so they add additional layers of analytics to RPI.
For an example of how RPI works, lets take the lowest ranked "good team (I used 3 losses as a cutoff)" in the NET rankings, Quinnipiac, currently 7-3, ranked at 217. They have wins over Coast Guard (not included in NET rankings, Division 3, 6-8), Central Connecticut (3-6), West Point (1-9), Albany (5-5), Stonehill (1-11), Niagra (2-6), Navy (2-6), Holy Cross (2-9), with losses to UMass (5-4), Canisius (5-5), Yale (5-5). So they beat a lot of bad teams, lost to mediocre teams (in terms of W/L). RPI would then take into account the winning percentage of every opponent these teams played before spitting out a ranking. So you don't need any data going into the season, it build its dataset during the season. It takes a few weeks because you need enough common opponents for the data to work/make any sense.
Disclaimer: Not a statistician at all (but I do basic stats as part of my job), so anyone please correct me if I'm wrong about anything.