Three Dimensions of Design Development

Neil M. GoldmanAAAI 1983

Abstract

Traditional formal specifications often present a finished, monolithic product that obscures the design intent behind complex systems. This paper introduces a framework for specification development based on three "dimensions of evolution": structural granularity, temporal granularity, and coverage. By treating the development process as a series of intuitive refinements from a simple base model, designers can bridge the "comprehensibility gap," making formal methods more accessible and documentation more meaningful.

Introduction: The Comprehensibility Gap

In software engineering and formal methods, there is a persistent disconnect between how a system is explained and how it is formally specified. When a developer explains a complex system to a colleague, they rarely start with the final, detailed code. Instead, they begin with a simplified abstraction and gradually layer on detail.

However, traditional formal specifications are typically written as "flat" sequences of text. This approach hides the design intent, making it nearly impossible for a reader to discern why certain constraints exist or how the system reached its current state. This paper argues that we must formalize the process of development itself, ensuring that the evolution of a design becomes an integral part of its documentation.

The Three Dimensions of Evolution

The core of the proposed framework is the decomposition of design changes into three orthogonal dimensions. These dimensions allow a designer to start with a "crude approximation" and move toward a high-fidelity model through structured refinements.

1. Structural Granularity

Structural granularity refers to the "resolution" of the system state. At a low level of resolution, a model might only record the most basic outcomes. For example, in a specification for a baseball game, a low-resolution state might only track the final score. As structural granularity increases, the state expands to include more detail, such as the position of every player on the field or the specific count of balls and strikes for the current batter.

2. Temporal Granularity

Temporal granularity deals with the "tick rate" or the time slices of the process. It defines the atomic units of action within the system. In the baseball analogy, a coarse temporal model might treat an entire inning as a single transition. A more refined model would break this down into individual at-bats, and a high-resolution model would treat every individual pitch as a distinct state transition.

3. Coverage

Coverage refers to the range of possible behaviors permitted by the specification. A base specification might only cover "normal" or expected behaviors—such as a standard nine-inning game. Refining coverage involves expanding the specification to include edge cases, exceptions, and rarer events, such as extra innings, rainouts, or player ejections.

Methodology: Evolution as Explanation

Using the Gist specification language, the author demonstrates how these dimensions function in practice. The methodology suggests that the most effective way to document a system is to record the steps taken to design it. In this view, a "language of change" is often more readable and informative than the final, static code.

One of the most significant insights from this approach is the interdependence of dimensions. A designer often discovers that they cannot achieve a goal in one dimension without first refining another. For instance, to accurately "contract coverage" (such as ruling out tie games), a designer must first refine the temporal granularity—moving from nine-inning blocks to half-inning blocks—to allow for the logic required to handle extra innings.

Furthermore, this approach allows for the emergence of complex rules as theorems rather than axioms. If the designer chooses the correct level of structural and temporal granularity, many complex system behaviors should emerge naturally from simpler, underlying rules, rather than being hard-coded as arbitrary constraints.

Results and Conceptual Shift

The baseball example successfully demonstrates that a complex specification can be decomposed into a sequence of meaningful, domain-specific changes. By moving from an atomic "nine-inning game" model to a "player-at-bat" model, the designer can naturally incorporate rules about scoring limits and game termination. These rules would feel arbitrary or "absurd" if forced into a coarser model, but they feel logical when introduced at the appropriate level of detail.

This represents a fundamental conceptual shift: moving away from viewing specifications as static text and toward viewing them as a sequence of functions mapping one process to a more refined version of that process.

Limitations and Future Outlook

While the framework provides a robust conceptual foundation, the paper notes several areas for future development:

  • Formalizing the Change Language: At the time of publication, the evolutionary steps were described in natural language. A fully realized version of this methodology would require a formal, machine-readable "language of change."
  • Tooling Support: Future development environments should support these structured modifications directly, acting as more than just text editors by tracking the dimensions of evolution.
  • Scalability: While intuitive for the baseball example, further research is needed to determine how this three-dimensional approach scales to massive, multi-million-line software projects where dimensions may overlap in increasingly complex ways.

By focusing on the how and why of design, rather than just the what, this three-dimensional approach offers a path toward more transparent and maintainable complex systems.

Summary updated

BibTeX
@inproceedings{goldman1983three,
  author = {Neil M. Goldman},
  title = {{Three Dimensions of Design Development}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1983},
}

Loading related content...

Figure