chapter five

5 Data and information architecture

 

This chapter covers

  • Understanding data architecture as the information-control layer for delegated autonomy
  • Extending lifecycle and cross-cutting controls to artifacts created and used by agents
  • Designing trusted context for specific purposes, decision surfaces, and authority levels
  • Governing retrieval, knowledge graphs, and context graphs
  • Treating policy as structured data and using pre-agent controls
  • Assessing data quality, spotting antipatterns, and applying the implementation playbook

At noon on June 8, 2023, Judge P. Kevin Castel opened a sanctions hearing in Courtroom 11D of the Southern District of New York. Before him sat two lawyers whose filing cited six judicial decisions, complete with case names, judges, quotations, and procedural histories, but no court had ever issued them. ChatGPT had generated the authorities, and the surrounding legal process had carried them into a federal court record in support of a client’s claim. Fourteen days later, the judge sanctioned the lawyers and their firm (Mata v. Avianca). The case matters because it exposes how information acquires authority as it moves through an organization. The model produced plausible text, and people selected it, incorporated it into an argument, approved the filing, and presented it to a court. Each step increased the material’s consequence without establishing the provenance, validity, applicability, and evidential strength that such authority requires.

5.1 The agentic shift in data architecture

5.1.1 Agentic effects on the stages of the traditional data lifecycle

5.1.2 Agentic influence on the horizontal cross-control dimensions

5.2 From data to trusted context

5.2.1 Trusted context: making data fit for authority

5.3 Supplying and interpreting context

5.3.1 Enterprise Retrieval-Augmented Generation: data access for agentic reasoning

5.3.2 Multimodal context: separating input, interpretation, and evidence

5.4 Preserving meaning, situation, and decision lineage

5.4.1 Enterprise knowledge graph: preserving governed meaning

5.4.2 Context graph: representing the live decision episode

5.5 Policy as data and pre-agent context guardrails

5.5.1 Pre-agent guardrails as data-architecture control

5.6 Data quality for agentic use

5.7 Data architecture anti-patterns

5.7.1 Context detached from source state

5.7.2 Derived information without inherited obligations

5.7.3 Relevance mistaken for authority

5.7.4 Policy as prompt

5.7.5 Activity logs without decision evidence

5.8 Implementation playbook: operationalizing the data and information

5.8.1 Phase 1: Implement the authority decision

5.8.2 Phase 2: Extend governance into derived information

5.8.3 Phase 3: Enforce trusted context

5.8.4 Phase 4: Configure and evaluate context supply

5.9 Summary