Circulating tumor (ctDNA) can be used for sensitive detection of minimal residual disease (MRD). However, the probability of detecting ctDNA at low tumor burden is limited by the number of mutations analyzed and available plasma volume. Here we applied a tumor-informed WGS approach for ctDNA-based MRD detection (91% sensitivity, 92% specificity) and treatment response evaluation in 916 longitudinally collected plasma samples from 112 patients with localized muscle-invasive bladder cancer. We show that WGS-based ctDNA detection is prognostic of patient outcomes with a median lead time of 131 days over radiographic imaging. We performed genomic characterization of post-treatment plasma samples to study tumor evolution and observed acquisition of the platinum therapy-associated mutational signatures and copy number variations not present in the primary tumors. Our results support the use of WGS for ultra-sensitive ctDNA detection, and highlight how tracking of tumor evolution using WGS of plasma samples opens opportunities to refine precision oncology.