The understanding of an ordinary differential equation has nothing to do with being able to successfully execute a Runge-Kutta method. The latter only gives numerical values for an individual trajectory (or if called multiple times, for many).

But understanding means to know where its fixed points are, how it behaves for large times, how sensitive the solutions are to a change of initial conditions, etc.. Not single numbers or curves but the general pattern of arbitrary solutions.

We are far for such an understanding of QCD except in the very high energy region. Thus people say that QCD is not understood because in the infrared domain (confinement, mass gap, bound states) we have very little grasp on how to obtain properties of arbitrary solutions at a level that conveys more than individual numerical numbers. Lattice QCD is just a black box that spits out a (fairly inaccurate) number for every numerical question we ask, it give not understanding in any sense.

One can say we understand QCD if we can derive from its action the low energy Hamiltonian and the bound state content (mesons, baryons, and perhaps glueballs) to an extent that we can match the meson and baryon data from the Particle Data Group to its particle spectrum.

We wouldn't understand Newtonian gravity if we coudn't do a qualitative analysis of the 3-body problem. There is no analytic solution but still we understand (and can approximate) essentially everything about its behavior, independent of the detailed parameters of the problem.

Lattice calculations in the 3-body problem would correspond to discretizing the dynamics using second-order divided differences, which very poorly resolves the dynamics of a 3-body problem, so a very fine lattice would be required to give good results over a significant time span, and then it would just be for a single system - nothing general.

Perturbation theory around a 2-body problem gives very useful analytic approximations not just for a single system but for all problems in the class, and one can deduce a lot fronm its sutdy, whereas even a better discretization method (like modern symplectic integrators) give just a single trajectory, or if repeated a bunch of trajectories, from which one cannot deduce much about the qualitative behavior.

Understanding always means to be able to derive **qualitative** understanding, not just numbers. (And even the numbers obtained from lattice QCD are not impressive. I haven't seen even a single attempt to compute the full baryon and meson spectrum from QCD. Given that QCD needs no numerical input to define it, the accuracy for basic predictions such as the proton mass (by lattice QCD or by Schwinger-Dyson equations) is perhaps 5 percent, and it will not grow much even if the speed of computers and algorithms increase by a factor of $10^6$ (which is not realistic). No, we need a much better understanding!

We understand QED much better than QCD, because there are many good approximation schemes which give qualitative information about almost everything of interest. But even QED is not completely understood as we don't have a logically satisfying setting for the theory, and things like the nonperturbative existence of the Landau pole are unsettled.