Compliance Patterns Specification

Compliance Patterns Specification

What This Domain Is

A collection of example things demonstrating compliance patterns, multi-lens reasoning, and regulatory best practices for domains operating under constraints (GDPR, HIPAA, financial regulations, etc.).

This is not a compliance domain itself. It's a reference library for domain builders creating regulated systems.

Philosophy: Why Examples Work for Compliance

Traditional compliance approaches:

• Write rules and hope the system follows them
• Rely on checklists to prevent violations
• Separate "compliance team" from "engineering team"
• End up with technical debt and audit friction

Better approach: Make compliance verifiable through examples.

LLMs excel at verifiable reasoning—tasks with clear right/wrong answers. Compliance is inherently verifiable:

• ✓ Data classified or not classified
• ✓ Access logged or not logged
• ✓ Data residency UK or non-UK
• ✓ Retention policy followed or violated

By providing both positive examples (compliant patterns) and negative examples (violations with explanations), we transform compliance from abstract rules into verifiable patterns. The LLM learns not just "what to do" but "why this is right and that is wrong."

Contrast Creates Clarity

Showing both the correct pattern AND the violation pattern (with consequences) gives the LLM two concrete reference points. This is why anti-patterns paired with remediation are more effective than rules alone—the LLM can verify its reasoning against both.

Core Principles

Verifiable Compliance: Every compliance decision should be reducible to verifiable facts, not subjective interpretation.

Multi-Lens Reasoning: Every significant decision should be evaluated through domain logic, compliance logic, and audit logic.

Explicit Conflicts: When lenses conflict, surface the conflict for human resolution rather than hiding it in logic.

Pattern-Based Learning: LLMs learn from examples; compliance becomes reinforced through every decision, not imposed as external constraints.