Hazardous products such as pesticides, medicines, novel foods, food additives, go through a regulatory process known as risk assessment. Risk assessment consists in the formulation of an informed opinion about the possible risks linked to the exposure to the substance being examined, on the basis of available studies. Risk assessment is often presented as a straightforward utilization of available evidence, to formulate a final opinion for the regulatory decision-maker. However, in practice, risk assessment involves the forging of knowledge claims amid various, sometimes contending actors and experts, working with massive, uncertain and selected sets of data. This “evidential work” (Demortain and Borraz 2022) is the source of frequent controversies around risk assessment opinions, and of public criticism of the work of risk experts (e.g. van Zwaneberg et al. 2025). Understanding how risk claims are forged in practice, how they emerge in the work of experts that examine data and evidence, to circulate and gain strength or on the contrary trigger controversy, is therefore a central challenge in these domains of risk assessment. The presented work aims to analyse the patterns of judgement and interpretation of the carcinogenicity of pesticides, as emerging in a large corpus of expert assessment reports produced in the European Union about pesticides. We introduce a framework for semantic risk and controversy quantification based on multi-layer semantic network modeling, that allows capturing the types of carcinogenicity, the modes of action but also the evidentiary basis most frequently used by experts, as well as the structural orientations of final regulatory judgements concerning pesticides, as carcinogenic or not.