Years ago, I did broadcast consulting engineering.

An FCC ground system efficiency proof involves measuring field strength at points along imaginary lines extending out from an antenna. The line extends radially in a straight line, and is referred to as a "radial". Although this process might be called "measuring radials", "doing radials", or "taking radials", it actually is the imaginary line being referred to as a radial. It is not a physical wire radial!

To do a proof, field strength is measured outward along a straight line at planned intervals. The distance is outside the Fresnel region, and outside the nearfield of the antenna, where signals decay at normal predictable rates. This is generally over 1 wavelength, and extends outwards to many miles. The resulting readings, at multiple points, are recorded.

The FCC publishes reference material in the form of graphs. This is covered in section 73.184 of FCC rules. A sample is shown here:

These graphs are available from the FCC upon specific request for various frequency ranges. The little numbers labeling each graph line are soil conductivity.

The SLOPE of the measurements along a radial is matched to the closest line slope in the graph. This then "allegedly" tells us average ground conductivity along each radial. 

From conductivity, we predict field strength normalized to some reference distance, such as one kilometer or one mile. For example, perfect ground would be the flat downward slope labeled "100mV/m AT 1 km" Note that line is above the graph line 5000, which represents 5000 mS/m  

This graph covers soil conductivities from 0.1 mS/m to infinite (perfect). For example, it contains 0.1, 0.5, 1, 1.5, etc. All values are millisiemens per meter.

Once we find and match the slope, we can predict the expected field strength at any distance. If we match the predicted field strength for a certain applied power, the ground system is considered good.

Naturally, field strength readings wobble all around with distance. As such, engineers often get "creative" in selecting a mean value to match a curve. I'd guess there is a few dB latitude in matching a curve, depending on scattering in readings. They should be choosing mean values, or taking more "points" to establish a mean value, but I've seen people not do that and just pick readings that make the system look good.

If ground conductivity is underestimated, antenna efficiency is overestimated.

Note the greater divergence in field strength with increasing loss when distances are large. Because Hams do not normally work extended groundwave, soil isn't drastically important for Ham applications.