Directly observing the chronology and tempo of adaptation in response to ecological change is rarely possible in natural ecosystems. Sedimentary ancient DNA (sedaDNA) has been shown to be a tractable source of genome-scale data of long-dead organisms^1,2,3 and to thereby potentially provide an understanding of the evolutionary histories of past populations.^4,5 To date, time series of ecosystem biodiversity have been reconstructed from sedaDNA, typically using DNA metabarcoding or shotgun sequence data generated from less than 1 g of sediment.^6,7 Here, we maximize sequence coverage by extracting DNA from ∼50× more sediment per sample than the majority of previous studies^1,2,3 to achieve genotype resolution. From a time series of Late Pleistocene sediments spanning from a marine to freshwater ecosystem, we compare adaptive genotypes reconstructed from the environmental genomes of three-spined stickleback at key time points of this transition. We find a staggered temporal dynamic in which freshwater alleles at known loci of large effect in marine-freshwater divergence of three-spined stickleback (e.g., EDA)⁸ were already established during the brackish phase of the formation of the isolation basin. However, marine alleles were still detected across the majority of marine-freshwater divergence-associated loci, even after the complete isolation of the lake from marine ingression. Our retrospective approach to studying adaptation from environmental genomes of three-spined sticklebacks at the end of the last glacial period complements contemporary experimental approaches^9,10,11 and highlights the untapped potential for retrospective “evolve and resequence” natural experiments using sedaDNA.