I thought about that, but this is what I had in mind: take a section (say 100 M) of an undulating road, smooth it out, then put a ridge across it that restores its roughness to its initial value. My feeling is that the latter would be more of a problem (this opinion is colored by the fact that, in my neighborhood, road repair is creating bumps and ridges like this.)

I guess it would depend on how big the ridge you add would have to be. I'm not at all an expert on this, but my thinking is that a ridge of size 2X would have an exponential effect on the travel of the suspension and resulting IRI value when compared to a ridge of size X. So a perfectly smooth road with a ridge of height 2X would have a higher IRI than the same road with 2 ridges of size X.

The wikipedia article has more details on how the measurements are done (there are multiple different ways/instruments used which can have different results) - https://en.wikipedia.org/wiki/International_roughness_index

> The IRI is based on the concept of a 'golden car' whose suspension properties are known. The IRI is calculated by simulating the response of this 'golden car' to the road profile. In the simulation, the simulated vehicle speed is 80 km/h (49.7 mi/h). The properties of the 'golden car' were selected in earlier research[12] to provide high correlation with the ride response of a wide range of automobiles that might be instrumented to measure a slope statistic (m/km).