Some past discussions on sbml-discuss lead to the idea that SBO terms could be used to annotate a model to indicate rules that represent moiety conservation expressions. See for example this thread:
Currently, however, the SBML spec does not mention this idea. I propose that we add some text about it for L2v4. The text would be at least in the section about SBO, and maybe there would be mention or references in other parts of the spec as well (e.g., in the section on rules).