Should there be a subheading someplace for the extra steps that are required in the life cycle if significant inference is involved, e.g. with large scale DL/OWL based terminologies/ontologies such as SNOMED-CT.
* The use of a reasoner makes development more like developing rather like developing with an compiled language. You only know what you have after your run the reasoner, but you have to understand the results in terms of what was asserted. The consequences of assertions are often unintended - "bugs".
* Finding unintended inferences can be hard
* Tracing unintended inferences to the source in the assertions can be hard.
We have practical experience with SNOMED CT - see our KCAP2011 paper  and other papers on my website 
Professor of Medical Informatics
School of Computer Science
University of Manchester
Manchester M13 9PL, UK
TEL +44 (0) 161 275 6149/6188
FAX +44 (0) 161 275 6204