It's not a philosophical question. Anyone who does the same research I did gets to the same result. Really the only thing here down to philosophy or opinion more than just a little bit is tires; and even that is more based in concrete reality than some people who don't develop physics would like to think.
Usually, most of the things except perhaps the gear ratios and sometimes the car mass and well known things like the wheelbase are off in most simulation cars. The reason is basically lack of development time, data resources (Like for extremely old cars), and I am afraid sometimes some incompetence potentially.
Due to that reason I basically throw everything out and assume it's all wrong; often it is.
How do you know it's off? Well, have a good understanding of suspension, engine, drivetrain, aerodynamics, tires etc. for one. Decent mathematical prowess and ability to use Excel or something similar also helps.
Without it you won't have any of the means to examine anything intelligently, which includes partly just knowledge, but also some physical/mathematical calculations.
If you don't know what it is that you have, nor what you want, you might still be able to feel the discrepancy as a driver but it will be very difficult to do anything intelligent about it.
Well, to start off, after putting in some general values like the commonly listed mass, approx. fuel amount and position etc, I do suspension geometry first as it's the easiest aspect to do but also one of the most important.
The way you determine if suspension geometry is correct is by finding the basic angles of the assembly, like caster and SAI and putting those in to start. For whatever reason, every KS car I've looked at has wrong caster and SAI. You usually find them from the workshop manual and they're usually truthful, so I don't understand it myself. Error is often 50% or more.
It has nothing to do with tweaking steering feel; you do that by moving the whole assembly forward or back to account for the immobile contact patch in simulators and some other stuff, which can and will mess up feel in some geometries, like BMW's low-trail struts.
Then if you're lucky and there's good orthographic images of the suspension, a laserscan, a CAD model, curves in the workshop manual, curves someone measured, roll center height, arm lengths, *anything*, you start changing the points until you achieve the desired result. For these multilink suspensions it entails quite a bit of knowledge of the virtual pivots of the arms to go about that; you can't just average out the links and put it in.
Worst case scenario there's only images to go by, and you have to rely on your understanding of suspension to estimate.
For example if you know that your trailing arm has a 20deg sweep angle, you can calculate the camber angle change of the wheel for any given arm angle, and that will serve as a guide. You can use it to determine curves after converting arm angle to wheel deflection in your respective geometry.
If you're lucky like me, the OEM will just report the curves in their manual or advertising, like Porsche and Mazda did. They are usually truthful although I'm quite positive Porsche manipulated the droop side of their front geometries, as it is quite awful if real-world measurements by other people are to go by. Also comes out the same in the simulator, so it seems not even OEM data is always trustworthy.
Talking about OEM data, but these simulation developers don't really get anything serious, so it doesn't mean any of these professional made cars are any better than mods inherently. Perhaps a car mass, maybe some gear ratios. Doesn't seem like any curves or suspension geometry was shared to KS respectively, except maybe by Lotus, but I've heard they're quite cooperative.
Engine stuff is pretty much dyno curves and some understanding, differentials are sometimes reported by the OEM or aftermarket, often that's "just knowing" it, or converting a TBR from a torsen into lock % in clutchplate for example.
Aero too; sometimes you get a CD and CL, sometimes not. Sometimes you have CFD, but CFD is inaccurate, so that's not definite either. It's all "guessed" for a large part except perhaps if you get a full aeromap for all orientations based on good data.
Setup ranges; sometimes manual has it, often you need to reverse engineer it or just measure it yourself. For the NSX I found precise min/max angles achievable via adjusting; for most other cars it's largely just down to reverse engineering and logic. Every car will have a bit different tolerance anyway, so...
Car inertia, CoG height, weight distribution and mass; if you're lucky to get the true values, great. Often you get something super general that's not very true, and need to dig deeper. For the FD3S I found inertia, CoG height and weight distribution from Mazda themselves; for most other cars, you don't. Exact mass is also not *that* important as just having +- 1l of fluids in the car will change it; so just try to find the best you can. Often sim cars have loaded mass or dry mass, when it should really be wet mass but with no fuel and only one 75kg driver, for example.
Tires, sometimes you have data for some things, often not. Tires are hard to measure. A lot of it is just stuff "we know" backed up by some data, some of it is pretty arbitrary and tweaked with feel and logic.
How do you know your car is right? You don't. You only know how wrong it is. Simulation hardware and software isn't yet at the point where you can make a clear and radical statement about the accuracy of the simulation.
However physics is physics, and assuming the simulation works as intended (Not all aspects do, but most do) then the proper input results in the proper output. You can somewhat measure output with car telemetry for example, and compare that to the simulation. You can film video and compare those; although that's approaching more subjective areas. You can listen to the extremely unreliable thoughts of people who drive the car in real life.
I like to just look at the numbers, then drive it to see if it makes sense on paper and on-track.
That's the short version.
