Knowledge Engineering, Methodology Extraction & Organizational Translation

The Inquiry: Can organizational governance knowledge — authority structures, decision processes, commitment states, constraint boundaries — be extracted from narrative organizational documents into structured, machine-queryable representations using a standardized cross-domain methodology, and do existing traditions provide this capability?

Falsifiable formulation: 1. No existing tradition (enterprise architecture, knowledge engineering, cognitive task analysis, business model ontology, model-driven engineering, structured authoring, or audit methodology) independently provides a complete extraction-to-projection system for organizational governance knowledge. 2. The audit tradition uniquely demonstrates that governance-level organizational understanding CAN be standardized across all industries — a century of practice validates cross-domain applicability. 3. CTA's foundational commitment to domain-specificity does not preclude standardized extraction at the governance level — the resolution lies in the distinction between domain-level cognitive knowledge and governance-level structural knowledge. 4. Formal bidirectional consistency between narrative documents and structured governance representations is impossible for non-bijective transformations, but an asymmetric architecture (structured representation as canonical) resolves this pragmatically. 5. Seven independent traditions converge on the need for structured organizational modeling but each addresses a different part of the problem. The gap is in their integration, not in individual coverage.

The most important synthesis across CTA and audit traditions is the level distinction. CTA operates at the domain level — extracting what experts know and do in their specific domain. The audit tradition operates at the governance level — extracting organizational structure (authority, accountability, constraints, evidence requirements) that is invariant across domains. These levels are not in tension; they address different kinds of organizational knowledge.

CommonKADS provides a structural parallel: its context-level worksheets (OM-1 through AM-1) ARE standardized across domains. It is the Knowledge model's Domain layer that requires customization. This maps to the governance extraction pattern: structural questions about governance (who has authority? what evidence is required? what constraints apply?) work across domains because governance structure is domain-invariant. Deeper probes about domain-specific governance patterns may require domain sensitivity — analogous to CommonKADS requiring domain customization at the Knowledge model level.

The audit tradition validates this with a century of practice: AU-C 315 and ISA 315 require the same organizational understanding dimensions (entity nature, objectives, internal control, risk assessment) across every industry — from manufacturing to healthcare to technology. The standardized framework, customized application pattern is exactly what governance extraction requires.


The convergence table (F16) reveals a systematic pattern: governance-specific constructs are absent across all seven traditions except audit methodology. Authority, evidence, decision provenance, and accountability state are not "nice to have" additions to existing frameworks — they are the governance surface that existing frameworks were not designed to model. EA models organizational structure for architecture. KE models knowledge for expert systems. CTA models cognition for training. Business ontologies model value creation for accounting. None models governance for accountability.

The audit tradition achieves full coverage because auditors must understand governance in order to assess risk. But audit produces risk assessments, not populated governance representations. The gap is not in understanding governance (auditors do this) but in representing governance computationally (no tradition does this). The contribution space is the computational representation, not the understanding methodology.

Bell et al.'s (1997) reframing of auditing as organizational modeling creates a direct bridge to Sprint 8 (World Models). The auditor constructs a systemic model of the client organization — strategy, economic web, business processes, performance measurement — and uses this model to predict expected financial statements, against which actual outcomes are measured. This IS the prediction formalism: `predict(organizational_state, strategy) → expected_outcomes`, with deviation measurement when actual results differ from expectations.

The connection to the Conant-Ashby theorem (RA-008, F5) is immediate: the auditor MUST be a model of the audited organization to regulate effectively (assess risk accurately). Bell et al. arrived at this independently from audit practice. The governance extraction methodology is the tool that populates this model — converting narrative organizational documents into the structured representation the auditor's model requires.

The BX consistency problem (F13, F14) resolves through an architectural decision: the structured governance representation is canonical; narrative documents are derived views. This is not a compromise — it is the architecturally correct choice. Foster's lens theory shows that well-behaved round-trip consistency requires information preservation, which narrative-to-structure transformation cannot provide (rhetorical structure is discarded). Stevens shows even QVT fails for non-bijective cases. The asymmetric lens pattern (canonical source + derived views) is standard in both BX theory and structured authoring (DITA). The consequence: modifying a generated document and parsing it back is not a supported operation. Updates flow through the structured representation, not through document editing.