Skip to main content

1. Introduction

1.1 Purpose and Scope​

CADL (Contract Architecture Description Language) is a domain-specific language (DSL) for formally specifying, verifying, and deploying institutional designs in System of Systems (SoS). This document presents the CADL language specification (Version 0.3, Draft), covering language objectives, requirements, syntax and semantics definitions, and toolchain design.

This document consolidates the specification for CADL implementation based on a survey of SoS research and SoS architecture design studies conducted at the ERTL.

Chapters 1 and 2 describe the goals of the language; what the reference implementation provides today is listed in Chapter 10 and Appendix A, §A.12.

1.2 Background: Challenges of Institutional Design in SoS​

An SoS is a system configuration in which multiple independently operated systems cooperate to achieve capabilities that no single system can achieve alone [Maier, 1998]. [ISO/IEC/IEEE 21841:2019] classifies SoS into four types: Directed, Acknowledged, Collaborative, and Virtual.

Traditional SoS research has focused on control/coordination algorithms and technical optimization, but governance and institutional design are now recognized as upper-layer design variables that structurally determine overall system behavior. However, current SoS institutional design relies on natural language contracts and agreements, presenting the following challenges.

  • Difficulty detecting ambiguity and contradictions: Natural language institutional descriptions cannot be formally verified for consistency.

  • Lack of consistency guarantees for institutional updates: Institutional updates accompanying system addition/removal are manual, with no engineering means for consistency checking.

  • Gap between design and implementation: There are no means to verify that institutional design intent is accurately reflected in implementation.

  • Absence of institutional parameterization: No framework exists that treats governance structures as quantitatively manipulable variables.

  • Lack of three-layer integrated description: No language can uniformly describe institutions (authority structures), protocols (coordination procedures), and algorithms (computational methods).

  • Difficulty of non-expert participation: Existing formal methods and architecture description languages presume advanced expertise, making it difficult for students, enterprise practitioners, municipal officials, and citizens to actively participate in institutional design.

In addition to the above challenges, a fundamental motivation for CADL is the need to treat SoS institutional design as a computational object. The following three objectives serve as the direct motivation for designing the CADL language.

Objective 1: Making SoS Institutional Design Computationally Tractable​

In real-world SoS, many institutions exist as human tacit knowledge, customs, and rules of thumb, often without any documentation. For example, household chore-sharing rules, implicit agreements in local communities, and informal inter-organizational cooperation exist only in participants' minds, making external observation and analysis extremely difficult.

CADL's primary objective is to model and formalize real-world SoS institutions, including such tacit rules and customs, making them manipulable as computational objects. When institutions exist as computational data, the following operations become possible.

  • Copy and reuse: Effective institutional designs from one context can be copied and adapted to another. For example, a household's chore-sharing rules can be transplanted to another household and adjusted for family composition.

  • Integration and composition: Multiple partial institutional descriptions can be integrated to compose larger SoS institutions. For example, delivery robot routing institutions and data sharing institutions can be merged into a delivery data coordination institution.

  • Version control and diff comparison: Track institutional change history, revert to any point in time, visualize changes as diffs, and identify the scope of impact.

  • Search and classification: Search repositories of accumulated institutional descriptions for similar patterns and recommend reuse candidates.

  • Tacit knowledge elicitation: Through human interviews and AI analysis of observational data, implicit rules are formalized as CADL descriptions. This creates a common foundation for sharing, discussing, and improving institutions.

Objective 2: Computing Institutional Violations and Contradictions​

If institutions are formally described, it becomes possible to computationally verify consistency and detect violations at runtime. Currently, contradictions between institutions (e.g., one contract mandates information sharing while another prohibits it for privacy protection) can only be checked manually, and oversights become inevitable as SoS scale increases.

CADL's second objective is to automate institutional contradiction detection and violation monitoring through computational power. This encompasses the following two aspects.

  • AI-integrated design support: AI such as LLMs receives CADL descriptions as input and automatically generates contradiction detection, correction proposals, and institutional improvement suggestions. When asked in natural language "Are these rules contradictory?", AI responds based on CADL's formal semantics. In the future, a workflow will be realized where CADL descriptions are automatically generated from natural language requests, formally verified, and deployed.

  • Human decision support: Visualize institutional contradictions and potential problems to enable informed decisions by human institution designers. Quantitatively present contradiction severity and impact scope, supporting prioritization of corrections. Simulate institutional changes (what-if analysis) to understand impacts in advance.

Objective 3: Automatic Conversion from Institutional Design to Control Logic, Optimization, and Code Generation​

Once institutional design is described in a formal language and verification is complete, the natural next step is to automatically generate control logic and executable code from institutional descriptions. Currently, there is a large gap between institutional design (policy documents) and implementation (code), with no guarantee that design intent is correctly reflected in implementation.

CADL's third objective is to provide a conversion pipeline from institutional descriptions to executable artifacts. Specifically, the following conversions are envisioned.

  • Automatic conversion from institutions to control logic: Automatically generate each actor's control logic (state machines, event handlers, coordination protocol implementations) from CADL contract and protocol definitions. For example, generate skeleton code for DISPATCHER assignment algorithms and ROBOT route-following logic from a robot delivery system's institutional description.

  • Institutional parameter optimization: Automatically tune institutional parameter values (alpha, beta, lambda, etc.) through simulation or mathematical optimization. Explore optimal parameter combinations for changing environmental conditions and automatically construct regime maps.

  • Runtime code generation: Automatically generate contract monitoring code (assume/guarantee satisfaction checks), protocol execution engines, and metrics collection code in target languages (Python, TypeScript, Solidity, etc.).

  • Integration with assurance cases: Automatically incorporate institutional verification results as evidence in assurance cases (safety arguments) and export to IEC 62853 agreement description databases.

Cross-Cutting Feature: Institutional Description Language Anyone Can Understand​

As a cross-cutting feature spanning all three objectives, CADL emphasizes that "anyone from students to enterprise practitioners, municipal officials, and local residents can understand the intent of institutional descriptions and participate in institutional design." SoS institutional design is a process involving diverse stakeholders beyond just engineers, and if the description language can only be read by a small number of specialists, democratization of institutional design cannot be achieved.

To achieve this universal readability, CADL adopts the following approaches.

  • YAML-based declarative syntax: Adopts indent-based descriptions that allow intuitive structural understanding even without programming experience, prioritizing readable syntax over JSON/XML.

  • Vocabulary close to natural language: Uses vocabulary that directly reflects institutional concepts such as authority, responsibility, and incentive, while minimizing exposure to formal method notation.

  • Graduated description levels: Three levels are provided: (1) Overview level (descriptions close to natural language, entry point for AI auto-generation), (2) Design level (parameter and constraint definitions), (3) Verification level (formal assume-guarantee descriptions), enabling participation according to user expertise.

  • Visualization: Automatically generate diagrams (authority structure diagrams, protocol sequence diagrams, regime maps) from CADL descriptions, enabling non-technical users to visually grasp institutional overviews.

  • AI-assisted barrier reduction: Through LLM-mediated CADL description generation, modification, and explanation in natural language, provide an environment where "institutions can be described by communicating intentions in natural language without knowing CADL syntax."

Integrating these three objectives with the cross-cutting feature of universal readability, CADL's vision is the establishment of an engineering cycle for institutional design: "formalize real-world institutions in a form everyone can understand, verify and improve them through computation, and automatically convert them into executable form." This can be positioned as extending the model-driven development (MDD) approach from software engineering to the SoS institutional design domain, while simultaneously broadening participation from specialists to society as a whole.

1.3 Glossary​

This document uses many technical terms. The Glossary defines them by topic — basic concepts, institutional design, the institutional parameters α, β and λ, institutional dynamics, verification and code generation, and the SoS types and related standards — with plain explanations and, where it helps, concrete examples.