Recent analysis of public policy is based on elaborate political-economic models. A major advantage of these useful models is that they have micro-foundations. Their main shortcomings are that they are usually based on drastic simplifications regarding the institutional setting and, more importantly, that the results they yield are not robust to the assumptions regarding the institutional setting, the policy space or the agents' preferences. An alternative reduced-form modeling is the public-policy lobbying contest. This model can be applied to public-policy determination in representative democracies. However, its main drawback is that the notion of contest success function (CSF), a crucial component of the contest model, does not have micro-foundations and, therefore, the random behavior of the government seems ad-hoc. In the present paper we propose a partial micro-foundation for the public-policy contest. The possible rationalization of random government behavior is illustrated in the case of the most commonly studied CSFs: the function associated with the all-pay auction and Tullock's lottery logit functions. We also clarify how stake asymmetry, lobbying-skill asymmetry and return to lobbying effort determine the relative desirability, from the government's point of view, of these CSFs.