This article deals with technology transfer from science to agriculture with pearl millet (Pennisetum glaucum (L.)R.Br.) in central Tanzania as example. The major question is which validity recommendations from different types of field experiments have and how geo-information (i.e. soil and landscape position) can lead to more site-specific recommendations. Tied ridging and reduced amounts of placed fertilizer during sowing were tested to increase yields on researcher-managed plots on-station, demonstration plots in villages, and farmer-managed plots on-farm. While on-station trials provided potential yield effects, physical distance to the station and differing conditions led to a higher informational value of village plots that mirror the context of local farmers. The treatments often resulted in significant yield increase. Soil and relief information and distance to settlements (i.e. gradient of management intensity) are key factors for data variability in on-farm trials. Unexplained variability is introduced through leaving degrees of freedom with respect to management to the farmer. Apart from soil and physiographic information, the latter should be part of a detailed data collection procedure in agronomic trials in large numbers addressing Sub-Saharan smallholder farming. Balanced data sets with dispersed trials on crucial soil and relief units are essential for future research.