Fitted
The real sample against a sample drawn from the fitted density, side by side.
The check an analyst asks for first, and the one a test cannot replace. A goodness-of-fit p-value says reject or do not reject; this shows where the fit departs from the data, which is what decides whether the departure matters. A fitted density that misses a small bump in the middle may be perfectly usable as a simulation input, while one that misses the upper tail is not, and no single number distinguishes those two cases.
The picture is honest even where the formal test is not. A two-sample test between these two samples would be invalid, because the drawn sample comes from a density fitted to the real one and so inherits its peculiarities — FitAdequacy records what that costs when measured. Looking at them together carries no such assumption: the eye is comparing two sets of numbers, not computing a null distribution.
Drawn at the same size as the real sample by default, so that the raggedness of the two is comparable. A larger drawn sample produces a smoother picture that flatters the fit, which is why it is not the default.
Parameters
the real observations
the fitted density
how many observations to draw from the fit; the real sample size by default
the stream to draw with, so the figure is reproducible
the stream provider
the dot bin width, defaulting to a hundredth of the range
what the observations measure, for the axis label