If a job class uses removeDuplicates=true and root job A1 exists and A2 is enqueued, then A2 enqueue becomes a no-op. At the same time, the A1 job is considered as "superceded by a new job" and is no-oped upon being run. A similar problem can happen with the sub-jobs from these root jobs if they also use removeDuplicates=true.
Maybe jobs with root timestamps should flip the "oldest job wins" logic to "newest job wins" for per-job deduplication. A simpler option is to just not update the "latest root job timestamp" registry if no job was actually inserted (due to removeDuplicates). The later option retains the normal removeDuplicates behavior.