Species distribution models (SDMs) are central to forecasting how species will respond to climate change, yet their accuracy when extrapolated to climate conditions far outside present-day variability is rarely tested directly. Here, we evaluate hindcast SDMs against a rare empirical benchmark: eggshell remains of Mute Swan Cyngus olor from an archaeological site in eastern Jordan, identified through palaeoproteomics and directly radiocarbon-dated to the Last Glacial Maximum (LGM) c. 26000 years ago. The Mute Swan’s current breeding range lies far to the north of the site, making its presence in glacial-age Jordan strong evidence of a substantial breeding-range shift relative to today. We use this evidence to test whether ensemble SDMs, built from standard bioclimatic predictors alone,