The model testing environment supports four levels of execution. The simplest level, trial level, corresponds to a single run of a model on a single stimulus, much as a single trial in a standard laboratory experiment involves presentation of a single stimulus and collection of a single response. In trial mode the execution of a model can be stepped through, one cycle at a time, and the changing contents of the functional components can be examined. Block level allows a set of trials to be performed in sequence. Block level may be used to test a model's response to different stimuli, or to test the effects of random elements within a model (such as buffer access or decay) on the processing of a single stimulus. In the later case, where COGENT is used in a monte carlo fashion, the model responses can be presented in the form of a table or a frequency histogram.
Two further levels of execution, subject and experiment, are provided to allow complete "computational" experiments paralleling those of standard laboratory experiments. An experiment consists of a number of "virtual subjects" run on a fixed design. Each virtual subject performs a sequence of blocks, and each block consists of a sequence of trials.
In addition to the above model testing functions, the UNIX version of COGENT includes a scripting interface. This interface allows complex experimental designs to be automated via scripts of execution commands. A script can be developed, for example, to run 20 virtual subjects in each of four experimental conditions, with each virtual subject performing 5 blocks of 30 trials.