A scientist's toolbox is multiple and diverse; it contains experiments, computer simulations, arguments, models.... One of these tools, thought experiments (TEs), have become a hot topic in philosophy, especially due to Kuhn's (1964) puzzling question: “How [...] relying exclusively upon familiar data, can a thought experiment lead to new knowledge or to new understanding of nature?”. During these last three decades several epistemic accounts of TEs emerged, which results in wide divergences in the literature as to what TEs are, how they function, how they justify their conclusions, .... In addition, most accounts rely on a-historically analysed case studies, which also yields disagreements about the conclusions we can derive from some TEs, thus in divergences pertaining to their epistemic function(s). In this talk, I will first survey the recent history of the debate on TEs. Second, I formulate a novel account of TEs: TEs share a common general structure which characterises them as inconsistency revealers and eliminators. Third, I defend this account with illustrations from the history of physics. Finally, I describe 4 ways TEs could fail, which will allow me to further formulate this account along the way. Finally, I conclude with a brief comparison with several accounts of TEs found in the literature.