Standardize interfaces to multi-paradigm, multi-mode, and
use discrete interactions,
design for an extended life cycle, and
define observables that will submit to a similarity measure
These four fundamental guidelines flow down into the following more concrete behaviors (principles) that constitute the modeling of functional units.
Every model should be under continual evolution. There is no optimal (finished) model.
Isolated models are nonsense. They only make sense in relation, comparison, and contrast to other models.
Every component of a model should be open to validation (data permitting) based on clearly defined measures. Verification should be limited to the examination of networks of validatable components.
Models must be capable of representing multiple, possiblyincommensurate,perspectives.
Arbitrary Functional Granularity
The composition of any system can be dynamic or open to interpretation. Hence fixed compositional attributes (like hierarchy) are always weak points in any given model. If a model preserves the ability to re-specify or re-interpret its functional granularity, these weak points are hardened.
Experimental Procedure Encapsulation
Capture experimental procedures in an unambiguous way so that different experiments give similarly formatted results and are easily repeatable, preferably automatically repeatable and extensible.
Always run multiple models, in tandem, of the same referent.
Automated Model Generation
Build mechanisms to automate (or at least assist in) the generation of new models.
Treat any computational process as if it were a black box with an impenetrable boundary around it.