Represented by the first diagram is the closed-loop, data-flow pattern that both Ejs and Ips use although Ejs set output as the priority process while Ips prioritize input. Data from the senses is acquired by processes that are band limited to either S or N-type information. Before this information is rationalized, it's first compared to references already residing in memory. This comparator function at the trailing edge of input processing is in itself comprised of algorithms that consume brain resources and time; it restricts data-throughput by narrowing focus, which facilitates all processes being online at all times. Rationalization, which is limited to either T or F-type processing, operates on this restricted data set and produces information for memory and or as controls for physical activity. Via memory management processes, there's a feedback loop where rationalization can alter or add to the set of references that is used by the comparator function in order to refine and or narrow input filtering for more efficient future rationalizations.
closed-loop.jpg
Represented by the diagram below is the open-loop structure that Eps and Ijs share; Eps set input as the priority while Ijs prioritize output. Raw data is acquired by similar front-end processes to those of closed-loop but instead of it being compared, it's dumped directly into memory. With no comparator eating up resources, more capacity is available to acquire and or analyze information but there is a price to pay. Input and output operate autonomously from one another with the dominant process determining when the subordinate can operate and on what will be its area of focus; the subordinate can be overridden at any time, the control of which is represented by the dotted line. Unlike the first configuration, one process is actually off-line when the other is functioning, which has inherent drawbacks even though switching from one process to the other can occur rapidly. Rationalization processes actually have to access memory directly in order to retrieve the information that was dumped there by the input processes; input and output are functionally disengaged from one another.
open-loop.jpg
Now every type possesses both these configurations; there's little difference between the two except for the comparator function, which does significantly affect how processing is done. Also, there are no real differences between extroverted and introverted preferences (N, S, F and T); rather, a perception of extroverted and introverted orientations are created simply from the two processing configurations. Although all these processes reside in memory, those that directly draw information from memory are defined as internalized (introverted?) and contain the mechanisms for cognitive stability.
As well, when a type switches from primary to its secondary configuration (dual-like), it's switching the comparator in or out and accessing the lesser used data filters and processing algorithms (N to S and or F to T, or vice versa), which accompany the other configuration and usually reside in the overlap regions between S or N-type information and processes.
As a final note, each information element represents a grouping of like-behaviours that are observations of cognitive functioning, but in no way represents an actual function. In order to properly model function, distinct independent variables have to be defined, but the elements are too granular and haven't enough homogeneity to facilitate modelling. As a minimum for example, homogeneity would require that the definition of S be restricted to information of an independent, self-contained or quantitative nature while N would be information of a relative, connective or qualitative nature. Similar redefinitions would be required for absolute (T) and associative (F) rationalization processes, and only then could Socionics be related to a hard-science model.