Metrics foster trust in governing bodies, but their uncertainty can elicit an opposite sentiment of mistrust. In chemicals governance, most of the conversations concerning computational models revolve around their uncertainty, and the extent to which simulations of safe doses can be transposed in regulatory decisions. To understand the source of this mistrust in models, research in science and technology studies on policy modeling, particularly research that looks at models as an interface between science and policy, must be extended to consider the private production of predictions. Looking at the full set of actors involved in predictive regulatory knowledge – companies, regulatory agencies, modelers working with one or the other – and their concurrent articulations of uncertainty, it appears that regulators audit physiologically based pharmacokinetic models (PBPK, a key class of models used to compute safe chemical doses), because the chemical industry initially introduced them to challenge its methods of risk assessment. Regulators and their modelers established model auditing, to be able to negotiate the predictive claims of companies and their consultants. At the end of the day, neither companies nor regulators appear to dominate the production of predictive knowledge. It is the product of the shifting distribution of expertise in the regulatory space, and of the outcomes of the recurrent trials of credibility that this distribution enables.