Ideally, Mosaic should be able to handle any data whatsoever that is related to molecular simulations. Since no exhaustive list could ever be written down, this requires a modular and extensible design.
Completeness also implies that a Mosaic data file should be as autonomous as possible. All its contents should ideally be interpretable without reference to external data sources.
The completeness criterion is motivated by the principles of Reproducible Research. Reliance on unpublished data, or data in incompletely documented formats, is one of the major causes of non-reproducibility.
Mosaic data files should be readable on any computing platform, present and future. This requires that all file formats must be specified unamiguously down to the byte level.
Mosaic must be able to handle big data sets (a few GB are not untypical nowadays), which requires compact storage and fast I/O. This aspect is currently covered by the HDF5 implementation of Mosaic.
Scientific data has a longer lifetime than most scientific software. The interpretation of Mosaic data should not depend on a complex software stack that may become unusable because of lack of maintenance. It should also be possible to handle Mosaic data in any programming language, current or future, without the requirement to be able to link to any specific other programming language.
This goal is fully achieved with the XML implementation. The HDF5 implementation depends de facto on the HDF5 library, although the file format is documented and partial alternative implementations exist. The Mosaic code library itself is proposed as a convenience, but not required.