In 2005 a prototype Observatory Information System (OIS) was developed for the Illinois River Basin Observatory (IRBO), connected to a federated scientific data network, populated with a representative collection of legacy datasets, and linked to external data streams. The perspective of seven years' time and the disestablishment of the system provide an opportunity to study the system life cycle. We detail best practices for multi-level OIS design for long-term performance, based on a publication-mandatory metadata implementation standard using ISO-19115. These principles balance general users' requirements against the requirements of specific scientific applications, and maximize the system's capacity to deal with legacy and heterogeneous data sources, enhancing long-term sustainability and flexibility for diverse multi-level user groups. These findings are relevant to ongoing developments of networked Scientific Information Systems that are increasingly critical to support and sustain the long-term benefits of modeling and observatory science. © 2014 Elsevier Ltd.