You have a large corpus of similar workload patterns. Simply using a 750GB trace or many samples from a cluster of machines doing the same work does not make it a meaningful variety. It would be reasonable to have also tested with the traces that your comparison algorithms used, which shows your implementations are valid and how they differ on those authors terms. Skipping that may come across as cherry-picking because I find that QDLP under performs by -20% in zigzag, -15% in DS1, -33% in blockchain mining. Your algorithm is good in specific cases, like many others are, but there are many real workloads where it is not usable.