The extent to which evolution can rescue a species from extinction, or facilitate range expansion, depends critically on the rate, duration, and geographical extent of the evolutionary response to natural selection. While field experiments have demonstrated that adaptive evolution can occur quickly, our understanding of the duration and geographical extent of contemporary evolution in natural systems remains limited. This is particularly true for species with large geographical ranges and for timescales that lie between 'long-term' field experiments and the fossil record. Here, we introduce the Virtual Common Garden (VCG) to estimate genetic differences among phenotypes observed in natural history collections. Reconstructing 150 years of evolution in Lythrum salicaria (purple loosestrife) as it invaded across North America, we analyze phenology measurements of 3,429 herbarium records, reconstruct growing conditions from more than 12 million local temperature records, and validate predictions across three common gardens spanning 10 degrees of latitude. We find that phenology evolves rapidly and repeatedly along parallel climatic gradients during the first century of evolution. However, the rate of microevolution stalls thereafter, recapitulating macroevolutionary stasis observed in the fossil record. Our study demonstrates why preserved specimens are a critical resource for understanding limits to evolution in natural. Our results show predictability of evolution emerging at a continental scale across 15 decades of rapid, adaptive evolution.