Here’s the cumulative distribution of chi^2 from Pengfei’s paper. Red line is a fixed RAR, basically equivalent to MOND. The other lines let g+ vary with Gaussian or flat priors. One gains nothing meaningful by letting g+ vary. 
Note that this distribution does not look right for chi^2. There are too many “bad” fits (high chi^2). That’s true. But it doesn’t help to vary g+ or anything else. Dark halo fits have a CDF with the same shape. One can’t best-fit one’s way into a sensible CDF. So what’s up??
Here is one of the individual galaxy fits. Traditional rotation curve top left, fit parameter right, a(R) bottom left, and on the RAR bottom right. The latter is what we’re fitting to. The eye says this is a good fit: the line goes through the data. But read the value of chi^2.
For UGC 11455, chi^2 = 6.5. Formally, that’s a terrible fit, contrary to what the eye can see. Looking closely, it is obvious that most of the “bad” fit stems from the innermost points. So the interpretation depends on how much we believe these points, and the error bars thereon.
Even if those are both bang on what they should be, there are unquantified systematics. The necessary assumption in the fits is that every point represents purely circular motion. That is a good approximation in most cases, but it certainly isn’t true everywhere all the time.
But if you don’t look at the data and don’t know how orbits in galaxies work, or appreciate both the strengths and weaknesses of astronomical data, then you can still learn to run a black box MCMC code, put it all through the thresher, and rediscover: garbage in, garbage out.
