From Text to Graph: Reimagining Government Policy for the AI Era

The Limitations of Text-Based Policy

For decades, government policy documents have been structured as dense narratives, endless paragraphs, and carefully cited memos. The Federal Acquisition Regulation (FAR) exemplifies this approach—organized into 8 subchapters and 53 parts that prioritize administrative categories over practical workflow. This structure creates a significant cognitive burden for users who must navigate complex text to find relevant information for their specific needs.

The problem has become so acute that on April 15, 2025, the President issued an Executive Order on "Restoring Common Sense to Federal Procurement," which noted that "the FAR has swelled to more than 2,000 pages of regulations, evolving into an excessive and overcomplicated regulatory framework and resulting in an onerous bureaucracy." The order called for significant reforms to ensure the FAR "contains only provisions that are required by statute or that are otherwise necessary to support simplicity and usability."

This text-based approach has served us adequately in an era dominated by human interpretation. However, as we move toward AI-augmented governance, these traditional formats are becoming increasingly inadequate. Large Language Models (LLMs) don't reason over unstructured prose the way humans do—they perform better, faster, and more accurately when provided with structured knowledge in the form of graphs, schemas, and explicit relationships.

Beyond Static Documents: The Need for Dynamic Knowledge

Government policies like the FAR are still encoding 21st-century missions into 20th-century formats. While documents worked for a while, they're holding us back from truly leveraging AI capabilities across policy domains—from acquisition to healthcare regulations, environmental rules to tax code.

Consider how reorganizing policy content by user need rather than by administrative categories could transform usability. In acquisition, this would mean organizing by "what you're buying" rather than by document section. This approach follows the natural thought process of practitioners who begin with "I need to accomplish X" rather than "I need to understand Part Y of a regulation."

Yet even this improved organization still relies on static documents. A more transformative approach involves moving to a knowledge graph representation of government regulations. This approach, known as GraphRAG (Graph Retrieval-Augmented Generation), combines:

• Natural language accessibility from Retrieval-Augmented Generation
• Structured compliance enforcement through knowledge graphs
• Dynamic, context-aware responses
• Clear traceability to authoritative sources

Thinking in Graphs vs. Static Documents

The transition from text-based policy to knowledge graphs requires a fundamental shift in how we conceptualize and communicate regulatory information across government:

From: Text-Based Thinking

• Long narratives with implicit relationships
• Information buried within dense paragraphs
• Cross-references requiring manual navigation
• Rules described in prose that must be interpreted

To: Graph-Based Thinking

• Explicit entities and relationships
• Clear connections between requirements and constraints
• Navigable pathways through regulatory requirements
• Rules as structured logic that can be computationally processed

This shift means training policy professionals across government domains to think not just about "what the rule says" but about:

• What entities are involved (people, organizations, thresholds)
• What constraints apply to these entities
• How these entities and constraints connect to each other
• Under what conditions specific requirements apply

Balancing Benefits and Risks

Potential Benefits Across Government

1. Improved Accessibility: Citizens and practitioners can quickly locate precisely the information relevant to their specific scenario
2. Enhanced Compliance: Clear relationship mapping reduces the risk of overlooking applicable requirements
3. Dynamic Guidance: Policy systems can adapt to specific contexts, presenting only relevant information
4. Faster Updates: Modular knowledge nodes can be updated individually as policies change
5. AI-Ready Structure: Enables more accurate AI assistance across government processes

Significant Risks and Challenges

1. Implementation Complexity: Knowledge graph development requires specialized expertise in both domain-specific policy and knowledge engineering
2. Technical Capabilities Gap: The government workforce currently lacks sufficient personnel with the skills needed for knowledge graph development and maintenance
3. Legal and Compliance Concerns: Questions may arise about whether restructured, AI-assisted systems constitute substantive changes requiring formal rulemaking
4. Resource Intensity: Initial development requires significant investment in both technology and human capital
5. Resistance to Change: Policy professionals comfortable with traditional document structures may resist learning new systems
6. Data Quality and Maintenance: Each policy update requires graph restructuring, potentially creating maintenance challenges
7. System Security and Integrity: Unauthorized modifications could lead to incorrect guidance
8. Over-Reliance on Technology: Users might defer too readily to system recommendations rather than applying professional judgment

Required Workforce Competencies

The transition to knowledge-driven policy systems necessitates new competencies across the government workforce:

1. Interdisciplinary Policy Experts: Professionals who can decompose complex regulatory requirements into discrete, logical components
2. Knowledge Engineers & Data Architects: Specialists in ontology development, knowledge graph design, and metadata schema management
3. Legal & Regulatory Compliance Specialists: Experts who can validate the legal sufficiency of knowledge nodes
4. AI & Language Model Specialists: Professionals skilled in RAG system development and evaluation
5. User Experience & Interface Designers: Specialists who can create intuitive interfaces for various user personas
6. System Administrators & DevSecOps Engineers: Technical staff to maintain infrastructure and ensure security

Conclusion

The evolution from document-based policies to knowledge graph frameworks represents a significant modernization opportunity across government. The FAR is just one example—similar transformations could revolutionize how we manage tax regulations, healthcare policies, environmental rules, and countless other domains of government activity.

This transformation requires us to think differently about how we encode policy. We need to move from paragraphs to relationships, from implicit connections to explicit ones, and from static documents to dynamic knowledge structures. While the technical and organizational challenges are substantial, the potential benefits for government effectiveness and citizen experience make this a worthy pursuit.

If we want to reform government policy for the AI age, we need to train professionals to map meaning, not just write more of it. Let's train them to think in relationships, not just paragraphs. That's the real key to unlocking AI-guided governance across all domains of public administration.

Example: FAR-Based Knowledge Graph for Commercial Service Acquisition over SAT but under $7.5M (FAR 13.5)

🧠 Context

Let’s say we want to model FAR 13.5 – Simplified Procedures for Certain Commercial Products and Commercial Services into a graph, using the scenario where a buyer wants to purchase a commercial service valued at $500,000 — above the Simplified Acquisition Threshold (SAT), but below $7.5 million, with no special contingency conditions.

🕸️ GRAPH STRUCTURE

🔸 NODES (Entities/Concepts)

🔸 EDGES (Relationships)

⚙️ HOW A GraphRAG WOULD WORK IN PRACTICE

Step 1: User Query

“Can I use simplified acquisition procedures to buy a $500,000 commercial service?”

Step 2: GraphRAG Traversal

• Detects “commercial service” → connects to FAR 2.101 definition
• Identifies $500K value → above SAT but under $7.5M cap
• Connects to FAR 13.5 — applicable because item is commercial and within cap
• FAR 13.5 requires compliance with FAR Part 12
• FAR Part 12 triggers:

Step 3: Output

Yes, you may use simplified acquisition procedures under FAR 13.5. Your $500,000 commercial service acquisition:

Step 4: Traceability

Each decision point is backed by specific graph nodes:

• Value check from N009 → N002/N003
• Product type from N004
• FAR 13.5 application from N001
• Clause requirement from N005
• Market research from N006

📈 BENEFITS IN REAL TERMS

• Policy changes? Update thresholds (N002/N003) and everything stays connected.
• Audit trail? You can show exactly why FAR 13.5 was applied.
• LLM integration? Pre-structured logic makes answer generation accurate.
• Training? Easier to onboard staff on when and how simplified procedures apply.

🧠 TL;DR

FAR 13.5 allows simplified acquisition over SAT for commercial buys up to $7.5M.

GraphRAG lets you encode this logic with nodes and edges:

• Nodes: Value, thresholds, product type, clauses, procedures
• Edges: Conditions, triggers, requirements

→ No interpretation guesswork, just traceable logic.