Various strategies for fault identification exist - e.g. based on formal analysis of code or on testing - of which each focuses on certain identification aspects and fault types. This paper characterises the strengths and weaknesses of methods – in theory and practice - focusing on application-independent identification strategies, and it suggests strategies to maximise the number of detected faults while minimising the related effort. Fault activation conditions are discussed in detail, resulting in an extended scope on stimulation needs. In particular, the contribution of automation in raising the activation probabilities is investigated. Various examples of fault activation mechanisms and statistics on fault types vs. identification methods are provided as observed in practice. An interesting result is the identification of application-dependent test cases by application-independent test strategies.

Certification is based on compliance of the code of the code generator with given standards. Such compliance never can guarantee correctness of the whole chain through transformation down to the environment for execution, though the belief is that certification implies well-formed code at a reduced fault rate.

The approach presented here goes a direction different from manual certification.. It is guided by the idea of automated proof: each time code is generated from a model the properties of the code when being executed in its environment are compared with the properties specified in the model. This allows to conclude on the correctness of the whole chain for every application and related generated code.