Algorithms and models for automated analysis of data are developed with a great deal of promises in mind, about precision, completeness, up-to-datedness, optimisation and anticipatory capacity gained by organisations employing these technologies. Many of these promises and expectations will, of course, be denied in practice, and algorithms may even produce consequences in decision that are adverse to the values which are meant to drive the regulatory decision in the first place. This paper outlines the conditions under which the promises and effects of privately developed algorithmic tools used in public action, may be evaluated.