and/or the addition of supplemental data or metadata derived from scientific research, user feedback, and other knowledge-building processes. For example, calibration and validation studies typically lead to successively higher-quality versions of a given data set via improvements in processing algorithms or error-handling procedures. Of course, these repeated improvements tend to exacerbate the challenge of increasing data archive volumes, particularly with multiple reprocessed data sets. Chapter 5 contains guidelines that can help data managers deal with this issue, noting, for instance, that demand will typically be far greater for the most recent version of each data set and so it may not be necessary to provide immediate access to older versions. It is vital that data stewards be engaged in these decisions.
As with most aspects of data management discussed in this report, the assessment and improvement function of data stewardship is most effective when it occurs on a regular basis and under a flexible but systematic set of rules and requirements. These rules should be advertised both to users and to data providers, who should in turn be given a chance to provide input to the process. Similarly, these rules should explicitly take into account the estimated costs and likely benefits of improvement efforts.