Some courts publish decisions as a single document that contains all of the opinions (majority, concurrences, dissents) that comprise the decision. Others issue the opinions in different writings. The challenge is to contrive a model that is equally useful in either situation.
The more difficult case seems to be the one in which the decision has been broken into multiple files. But in fact it is possible to have a representation that would work for both (roughed in here using element and attribute names that are generic):
<!-- other useful elements here, eg. identifiers for the decision -->
Majority opinion, not in this file</writing>
<writing idref="whatever">A part of the opinion contained in
this file and carrying an id attribute</writing>
Under this system, any part of the decision that is found within the "current" file is identified with an idref attribute that presumably points to a container element within the file that has the corresponding id attribute. Parts that are held externally are pointed to by URI attributes. Most decision TOCs will thus consist of writings all of which have idref attributes (everything is bundled in the current file), or of writings only one of which has an idref attribute (the current file) and the rest of which carry URIs (the related files). (By "current" file, we mean either the file in which this metadata is found (eg. if it is part of a metadata header), or the item identified by this metadata (eg. if it is part of a metadata record issued by an OAI server).
The actual text contained in the element is more or less arbitrary, and can be thought of as the "title" of the writing being described. See "Elements vs. attributes".