The assessments described in the previous sections involve the use of multiple PA models and, in some instances, chains of models. The appropriateness of a model depends on its subjective representation of the real-world system and its intended use. Suitability for the intended purpose can be evaluated through a well-planned Code and Calculation Verification and Model Validation (V&V) programme, considering qualitative and quantitative performance measures.

In the probabilistic workflow described previously, uncertainty in the ranges of input parameters is propagated directly to uncertainty in model output and potentially further (e.g., if parameter uncertainty results in situations where a performance target may not be met) to the development of safety scenarios and to the analysis of radiological consequences. Uncertainty is, however, also associated with model abstraction, or conceptualisation, which inherently introduces inaccuracy and additional uncertainty by simplifying real-world systems. Assessing such uncertainties within the PA workflow is integral to building confidence in the evaluation of post-closure performance and safety for the deep geological repository.

Chapter 5 of NTB 24‑22 Rev. 1 (Nagra 2024u) provides a detailed discussion of model abstraction in the context of PA, including:

• geometric abstractions,

• the simplified representation of phenomena and processes, and

• homogenisation and the scaling of parameters.

Nagra’s efforts in code-benchmarking and PA model validation are detailed in Section 5.7 of NTB 24‑22 Rev. 1 (Nagra 2024u). In general, it is shown that the impact of abstractions is either small or errors on the side of conservatism; see, however, the discussion of the characterisation of the geological barrier below.