Pipelines could contain the context for the extractor and loader, then schedules would be lighter and transform would be easier to run (needs the context of the pipeline)