I will have to dig into some books, since it is a long time ago.
however, on a pure certification data collection side :
On a V1 cut takeoff, a theodolith is (was) used to measure the angle from the reference point to 400ft, a simple arithmetics would give the gradient; now, if my memory is correct in FAR25 we needed 7 valid points between 2 inflexions of any performance curve, therefore at least 7 measurements, during the life of the test program we were gathering much more data, in order to give "sensible" information, these data were crunched using statisical methods in particular the RMS methodology for Gauss distributions; leading to a sample of around 70% of the mean value for each point.
This being said, the manufacturer can publish any number within the 7 points in the Gauss distribution since he is covered by the RMS buffer; this is why some aircrafts are "flying the book" better than others.
Today, as far as I can see the gradient published in tab data is the one at 400 ft AGL, again this is the RMS gradient, leading to a conservative publication, this is why when asking the manufacturer a special set of data for specific conditions , you usually end up with performance numbers you would have not imagine.