The IPCC assessment report relies partly on CMIP (Coupled Model Intercomparison Project) models, which project a warming rate of the hottest days averaged over midlatitude land to be 1.5 times that of the global- and annual-mean. However, observations spanning the last four decades indicate a slightly higher warming rate, double that of the global- and annual-mean. This raises concerns about potential underprediction by the CMIP models used by the IPCC. Zhang and Boos (2023) introduced a theoretical upper bound for surface-air temperature (T_s) beneath atmospheric anticyclones, dependent on the 500-hPa temperature (T₅₀₀). Here we examine how the hottest days are positioned in the T_s-T₅₀₀ phase space described by this theoretical framework. Our analysis shows that the hottest days simulated by CMIP models tend to drift further away from the upper bound under anthropogenic forcing, whereas ERA5 reanalysis exhibits a closer alignment with the slope of the upper bound. To explain this model-reanalysis discrepancy, we show that the warming trends and the surface air specific humidity trends of the hottest days, for each degree C of free-tropospheric (T₅₀₀) warming, are anti-correlated among models and ERA5. In particular, ERA5 lies on the high-warming-and-low-moistening end of the model distribution. These findings identify a potential model bias in the warming trend of the hottest days and underscore the significance of accurate land hydrology representation in CMIP models. Overall, this study provides a quantitative framework that integrates midlatitude blocking dynamics, atmospheric lapse rates, and land evapotranspiration to understand and constrain midlatitude extreme temperatures.