The % error is simply a rough indicator and is actually much less important than the graphical fit that you can verify by inspection of the RESAP output. One bad data point can make a perfect fit look terrible when you look at the % error figure, even if only one data point is off a curve that matches all the other data points. On the other hand, a low % error can get you into trouble if you are not paying attention to the degree to which RESAP is extrapolating deep layer soil resistivities to very high or very low values that are unsupported by the data or simply (especially when you choose the Levenberg-Marquardt Method instead of the default Steepest-Descent Method) yields extremely high or extremely low soil resistivity layer resistivity values (like 50,000 or 0.1 ohm-m) to give a beautiful mathematical fit, but unrealistic soil resistivities.
The % error value is simply the square root of the mean value of the squares of the differences between each measured data point and the corresponding computed apparent resistivity from the soil model RESAP has come up with… and about as enlightening as my definition. The bottom line is that the % error value usually simply confirms what your eyes can see: the computed apparent resistivity curve does or does not fit the measured data. But your eye can see infinitely more detail: are there one or two data points that are interfering with the curve-fitting process and should be eliminated in order to obtain a better fit? Does RESAP extend the computed curve in a reasonable manner on the right-hand side of the graph or is it climbing or dropping to much higher or lower levels than the measurements can justify? At what pin spacings is the computed curve not matching the measured data: close to the surface, which may have significant variation from one measurement traverse to another and little influence on grid performance? At greater, yet still short spacings, corresponding to soil resistivity at grid or rod depth, which will have an important impact on touch and step voltages? At large pin spacings, which could have a greater impact on GPR, ground impedance and transferred potentials?