If you have experimental data, do compare it against that rather than say another simulation. You never know how far off their values are. With that being said, literature I’ve read through generally agree that K-omega SST and RNG K-esp are quite good in predicting your needed values. For the K-omega SST model in particular, a full resolution solution with a y+ of less than 1 would ensure you obtain highly accurate results. Of course, you will need additional computational power in order to mesh and simulate a grid that fine.
With regards to your Formula SAE simulation, I’m not experienced in it so other will be more equipped to give you tips on this!
Convergence criteria is subjective and relates back to literature concerning what is deemed as converged. This can vary but in general a residual of 1E-4 is deemed as loosely converged, 1E-5 as well converged and 1E-6 as ideal convergence. So based on previous studies that you’ve mentioned if the residuals do not go below 1E-5 as you’ve mentioned then it is not converged. Why is it not converged can be due to your numerics or mesh typically.