Spectral "Compensation"?

Curious what direction you will go with then.

I do find the results of whatever core algorithm (ITU BS 1770?) is being used in fluid.loudness~ quite effective.

And/or if whatever you do come up with will back propagate to descriptors~ (descriptors~ funny business with loudness)

Indeed. @jamesbradbury has also kindly agreed to a brain picking session too, so that will no doubt be fruitful.