Samuel Pawel and Leonhard Held's contribution to the Discussion of 'Safe Testing' by Grünwald, de Heide, and Koolen.

The article critically examines the concept of "safe" inference tests, particularly those based on e-values, highlighting that despite their theoretical advantages, these methods are not inherently free from practical and cognitive challenges faced in scientific research. It notes that issues such as preference for classical study designs, difficulties in implementation, and common interpretative biases like "nullism" and "dichotomania" persist regardless of the inferential framework used. The authors emphasize that e-values, interpreted as Bayes factors quantifying relative predictive performance, require careful contextual application and cannot guarantee validity without correct model specification and bias adjustment. To improve interpretation, they propose graphical displays of e-values as functions of tested parameter values to encourage gradual, quantitative inference rather than dichotomous decision-making. Overall, the article underscores the need for cautious use and nuanced understanding of "safe" tests within the broader scientific process.