I voted realistic as I would ultimately want it to become a realistic sim, but I think it should start off as an almost novelty physics sandbox doing things very differently which would be fun to mess around in without really being a "game" as such, but with potential to grow into one as the project progresses. Might sound odd but I'll try to explain.
BeamNG.drive for a long time was basically just a sandbox for the soft-body car physics, allowing you to crash and wreck cars in weird and wonderful ways in a blank level with a few obstacles in it. No real objectives but still a lot of fun to launch the cars around and see the damage physics. Gradually it's grown into a genuinely very good driving game with many modes and features which even simracing enthusiasts are starting to take seriously. All built on top of an incredible physics engine which simulates every component of the car, aerodynamics, even air density/temperature, brake temperatures and loads more... The physics engine is so good that it allows helicopters and planes to be made which work in a realistic way when it wasn't even designed for them.
I would prefer something really wonky initially that is trying to do something truly new and ambitious with a great foundation to grow over something just copying the same old gameplay/physics football games have used forever.
Perhaps the most relevant example would be
Goofball Goals - this truly is a wonky physics sandbox for comedy purposes, but I am serious when I say this is the kind of base I would love to see a football sim based on. Two of the major complaints with football games are that the ball is not a seperate entity/true physics object and theres no proper footplanting. As stupid as Goofball Goals looks, it does both of these. The players move because they are physically simulated rather than animated and push with their feet, and the ball is literally just a simple physics object the players interact with.
Goofball Goals looks ridiculous because the characters only have a few behaviours/actions, and they are not very robust. Imagine that kind of silly looking physics sandbox but with the characters given tons more behaviours for locomotion, ball interaction etc using something like
DeepMotion Motion Brain.
I don't know whether this is open to indie devs, what the pricing would be or whether it's viable but the ceiling for how realistic a game on such foundations could become is incredibly high and could become more feasible as the technology progresses. Ubisoft have an even more impressive solution like this in-house, and maybe in a few years it'll become standard and accessible to indie devs (Unreal Engine 6 feature?

)
The standard animated style of football game, I'm not really interested in any more. If I wanted to play one of those then I doubt an indie project could surpass FIFA 16, PES 2013, PES5 etc.