Numerical simulation software for safety assessment rest on hierarchical verification and validation procedures to build confidence in their adequateness for the targeted application. Likewise, data from experiments at vastly different scales inform the underlying physical and chemical models. Pore-scale imaging provides crucial information on the evolution of macroscopic properties. On the other end of the spectrum, experiments in underground research laboratories (URLs) are particularly complex, costly and typically run for years to decades. They provide information on spatial and temporal scales that are not accessible via classical laboratory experiments. However, the interpretation of the data is often difficult due to the complex coupling of processes and the heterogeneity of the geological formations combined with uncertainty in process understanding, structure and parameter values. Thus, numerical simulations based on coupled process simulators are an important tool to interpret the data from these experiments but also design them. In this contribution, we provide a summary of lessons learnt from past modeling of in-situ URL experiments using the open-source code OpenGeoSys and how both hypothesis testing and a systematic integration of uncertainty quantification can help to better understand the data. They include full-scale heater experiments in clay rock, experiments on multi-segmental seals for mine shafts performed at a range of scales, and experiments on radionuclide transport in fractured crystalline rock masses. We then discuss how uncertainty in the coupled flow process thermo-osmosis, that has its origin in surface phenomena on the nano-scale, motivates the conceptualization of a new in-situ experiment guided by extensive numerical simulations validated on previously obtained field data.