Highlight reconstruction is all about not "exposing properly," (or is it?) so that one, two or all three channels get clipped in the raw file. The idea is to purposely do this to see how the application deals with sensor saturation (so as to avoid the magenta highlights problem when reducing exposure to recover highlights) and how it infers the missing color channel data from intact data to reconstruct highlights. When you are shooting a high contrast scene, knowing how far you can let the highlights go before your raw converter cannot reconstruct them is critical to setting the best exposure for the scene.
I recently switched from shooting Canon for years, with all of the 5D cameras, to Fujifilm. For each camera I used, I would shoot a sequence of raw files with increasing exposure to probe how to spot meter for the camera and raw conversions. I would set the reference metering on the brightest highlights for which I wanted to retain detail (say, fluffy white clouds, or a white terry cloth towel in direct illumination) and set that spot meter reading as my initial exposure. I would then increment shutter speed to increase exposure in 1/3 stops for about +5 EV total bias from the initial meter reading. I then examined each file in Raw Digger to see what bias value gave me clipping in the green channel (typically, this channel reaches clipping first in daylight). The red and blue were typically 1 to 1.5 stops under the green value, providing plenty of data for reconstructing the missing green channel - especially if the clipped highlights were neutral. Depending on the raw converter, I could spot meter the highlight and add 3 to 4 stops of exposure bias and still have intact highlights with highlight reconstruction added in raw conversion.
I suppose if you wanted to test a bunch of converters to see how they all handle highlights and highlight reconstruction, then you could shoot a few different scenes with 0) the exposure set so that no channels clip - that is the hottest channel is exposed so that it is right next to the right edge of the raw histogram; 1) an exposure that clips the green; 2) a slightly higher exposure that clips both green and blue; and 3) a slightly higher exposure that clips all three channels. For each of these images per test scene, you could then run them through the raw converter you are testing and see how it responds. I would suggest that you test scenes that have both neutral highlights (like clouds) as well as colored highlights (like a towel or some other bright, textured colored surface - this may skew the order in which the channels clip).
As far as I know, Raw Therapee is the only relatively popular raw converter that explicitly exposes several different highlight reconstruction strategies directly to the user.
Kirk
