DNA sequencing library preparation is a crucial step for next-generation sequencing of cell free DNA of plasma for cancer diagnostics. We present SRSLY, a robust single-stranded DNA library preparation method, that generates libraries with unique molecule identifiers (UMIs), with advantages over traditional double-stranded DNA preparations. Using 5 ng of plasma-derived cfDNA from acute myeloid leukemia (AML), myelodysplastic syndrome (MDS), and gastrointestinal cancer cases, we prepared both SRSLY and double-stranded DNA (dsDNA) sequencing libraries with UMIs. The libraries were enriched using a Twist Biosciences custom hybridization probe set for an 800 kilobase cancer mutation panel. We sequenced to a panel depth of 1000x to 2000x, after correcting UMI sequences and removing duplicate reads. With comparable fold enrichment and on-target percentages, we observe complexities of 11-21 million unique molecules with SRSLY. This is approximately four times higher than dsDNA complexity of 2.8-4.9 million unique molecules. Using duplicate reads from the same unique molecule, we perform error correction on template read bases and are able to call low-variant allele frequency mutations. In particular, we show an increase in KRAS p.G12D variant allele fraction during progression from MDS to AML. SRSLY shows increased recovery of small fragments over dsDNA preparations and retains the native lengths and termini of all fragments. Apart from improved library complexity, we show that MDS and AML samples can be distinguished from healthy samples by their disrupted genomic profile of short to long fragments, which has been previously demonstrated in solid tumors but not for MDS or AML.