When this keyword is present, the data (amplitude and its standard deviation) will be multiplied by the given constant f. The multiplication takes place after all processing of the data is finished (ie, taking differences, squaring, etc). Downscaling the data may be useful when the (extreme) sharpness of the MaxEnt map suggests that the data may be way off the absolute scale (on the high side, ie they must be downscaled). I think that rescaling the data at this stage is totaly unjustified. The only good excuse that I can think of, is in the case of an isomorphous difference Patterson function calculation : if the derivative is non-isomorphous15, then there are good chances that with increasing resolution, the mean fractional isomorphous difference will increase (instead of decreasing). This could fool maxent into believing that there are good-strong data even at high resolution. This, of course is correct, the only problem being that these ``strong'' high resolution data is noise from our point of view16.