This paper studies the stability of latent preferences measured in conjoint experiments. It investigates the amount of over-time change in the support for ideal presidential candidates and the degree to which they are subject to real-world campaign effects. We present four sets of original pre-registered hypotheses that draw on the literature on conjoints, attitudinal stability, and vote choice. They are tested using data from a repeated conjoint experiment that was embedded in a general five-wave panel survey conducted in the context of the 2023 Czech presidential election. The results show a striking stability of latent preferences and negligible campaign effects. Latent and real-world preferences do not increasingly align as the campaign progresses, and changes in preferences for real-world candidates do not have a significant bearing on respondents' behavior in the conjoint experiment. These findings cary important implications for the conduct of survey experiments and the meaning of what they measure.