> The answer was no: in more 1000 trials with randomly assigned planet sizes put through a virtual Kepler’s detection scheme, a pattern of similarly-sized planets in the same systems never emerged. This computational experiment did not reproduce what we observe in the Kepler planetary systems. Thus, the regular sizes of planets is a real astrophysical pattern.
This is actually a really cool use of computer simulations that I haven't noticed before (but I'm sure has been done in the past): simulating expected output for a known bias.
I wonder if it would be feasible to do the reverse too for a kind of parameter fitting to fix our model:
1. Create a reasonable approximate model for both Kepler's expected detection bias and for planet size distribution.
2. Constrain parameter space for these models as much as possible, to minimize the search space.
3. Using these models, generate a set different planetary sets for different parameters.
4. Find the outputs with the closest match in statistical "shape" of the actually measured data.
5. generate new input parameters based on these output that lead to set outputs to refine shape until certain level of matching has been found
If done naively this would probably take a lot of computation and result in large a solution space: the more parameters, the higher the conditionality of this space. So the models should be approximations with as few parameters to produce reasonable results, and with constraints on said parameter values based on known physics. This would narrow down both the parameter space to search the possibly valid outputs of this space.
Based on these parameters differ from the output we had expected before, we might then get a direction of where to look for the physics that does explain the distribution.