Here in the early days of what some are calling the age of component-based software engineering, we are awash in stories where the architect thought he or she could plug two components together with a connector, only to find out that the component didn't implement the right protocol, or was otherwise badly matched with the expectations of that connector. This is why we prescribe writing a justification where the match-up is less than obvious. For a thoughtful treatment of element mismatch, see [Garlan+ 95].
It is tempting to treat architecture as an assembly of components, but there are great conceptual advantages to be gained from elevating connectors to a first-class architectural status as well. Mary Shaw makes an eloquent argument for doing so in [Shaw 96b]. Shaw and Garlan [ShawGarlan 96] treat software architecture in terms of components and connectors and address concerns such as constructing systems as assemblies of components. For a more thorough discussion of connector mechanisms see [Shaw+ 96]. Allen and Garlan lay out the semantic foundations for connectors as first-class entities [AllenGarlan 97].