The urgent need to retrofit the existing building stock is a well-known theme, abundantly dealt with in the scientific literature. Thermal simulations have a key role, assessing the effective energy performance of buildings. At the same time, researchers frequently outlined significant discrepancies between real and simulated outputs, questioning the credibility of the obtained results. In this context, the present contribution introduces a comprehensive framework for the building energy simulations, ensuring the necessary model reliability. The overall approach is structured into six phases: building audit; initial model definition; initial model validation; sensitivity analysis; model calibration; and calibrated model validation. The validated model is the design premise to propose well-balanced retrofitting scenarios, improving the energy performances, enhancing the internal comfort, and preserving the architectural values. As operative case-study, Chauderon suspended building in Lausanne (1969-1974), by the Atelier AAA and Jean Prouvé, has been selected. Today, the complex is well-preserved in its original materiality and represents an iconic example of the modern aesthetics, with an expressed need for retrofitting. A focus on the architectural, engineering and technical features that made of Chauderon a unicum in the Swiss panorama allowed to identify the intrinsic value of this object, assessing the pertinence of its preservation. The final aim of this study is to minimize the risk of inappropriate interventions. The presented workflow is thus intended as a practical guideline, promoting reduced energy demand; higher thermal comfort; heritage preservation; architectural authenticity. A new way for reconciling the aspects of energy saving and architectural preservation is outlined, reconducting these two aprioristically opposed issues as two complementary parts of the same rehabilitation agenda. For high-quality post-World War II building stock, the use of a calibrated and validated model is finally justified and recommended over the simplified modelling approaches still commonly employed today.