§2.1 The Problem
Governance intent — what should be, what cannot happen, what must be recorded, who decides — is expressed in human language. Human language is rich, ambiguous, and context-dependent. Machine-processable structure is precise, deterministic, and context-free. Any governance system that aspires to be both humanly meaningful and computationally enforceable must bridge this gap.
Most governance frameworks do not attempt the bridge. They operate entirely in human language (policy documents, procedures, standards) and rely on human interpretation to enforce. This works at human scale. It does not work when the governed system includes autonomous agents, real-time decision loops, or distributed computational processes that operate faster than human review cycles.
The Computational Governance Grammar (CGG) is the infrastructure that bridges governance language and machine-processable structure. It is not the governance content itself — it is the classification regime and structural enforcement mechanism by which governance content becomes computationally enforceable.
§2.2 The Three Classes
The CGG classifies every governed artifact into one of three structural classes based on how the governance system processes it:
Directive records (OBJ in the DLP substrate) declare direction. They state what is being optimized, what constraints bind, who holds authority, how conflicts resolve, and when the directive activates and terminates. The governance system processes both their governance layer (the objective structure) and their operational layer (the execution state). Charter records, organizational objectives, research priorities, and operational plans are directive records. A directive record answers: what should the world look like if this record is satisfied?
Descriptive records (NOBJ in the DLP substrate) capture execution against an already-declared directive. They use the same structural grammar as directive records but do not carry the governance layer. They are processed structurally by the governance system — their content drives routing, classification, and learning — but they do not commit the system to direction. Session reports, evidence records, observations, and suspended ideas are descriptive records. A descriptive record answers: what actually happened under this directive?
Hosted records (WP in the DLP substrate) are artifacts whose internals follow a domain-external standard rather than the governance grammar. The governance system manages the artifact's perimeter (identity, authority, lineage, classification, cross-references) but does not parse the artifact's body. Research analyses, audit workpapers, tools, diagrams, and instruments are hosted records. A hosted record answers: what domain-specific work product was produced, and who is responsible for it?
§2.3 Why Three Classes
The directive/descriptive distinction prevents over-standardization. Forcing governance-layer structure on every artifact would be cost-prohibitive and would constrain exploration. Descriptive records can be captured in pre-vocabulary form. Organizational learning emerges bottom-up from descriptive accretion.
The descriptive/hosted distinction prevents over-processing. Forcing every artifact to be grammar-mapped would distort domain work. A philosophical analysis cannot be coherently structured around a "capacity" primitive because that is not how philosophical analysis is structured. Hosted records carry whatever internal structure their analytical or evidentiary purpose demands; the governance system hosts them with a thin perimeter so they remain traceable and authority-bound without their internals being constrained by governance grammar.
The combination keeps governance disciplined where discipline matters and open where openness matters. Directives are tightly governed. Descriptive records are structurally processed but directionally uncommitted. Hosted records are perimeter-governed but internally free.
§2.4 The Universal Grammar
Every directive and descriptive record shares a common structural grammar. In the DLP substrate, this grammar consists of nineteen primitives across five tiers:
- Tier 1 — Structural skeleton: intent, evidence, authority, work, constraint, decision, account, commitment, capacity. Always required.
- Tier 2 — Deployment: identifier, entity, context, namespace. Required for any deployment.
- Tier 3 — Temporal continuity: orientation, learning, activation. Conditional on governed operation.
- Tier 4 — AI-native extensions: interpretation, environment interface. Conditional on AI participation.
- Tier 5 — Temporal governance: cycle. Conditional on temporal legibility.
The universal grammar is what makes a governed record a governed record. It is what the conformance test architecture (§3.6) tests against — all nineteen primitives as structural test axes, organized into four conformance categories. Without it, governance is policy prose — valuable but structurally unenforceable.
This framework is specified against the DLP substrate. To adapt it to a different substrate, the structural requirement is not "use these nineteen primitives" but "have a universal grammar that all governed records share, such that governance tests can be applied uniformly" — the adopter maps their grammar to the nineteen primitives so the matrix's tests carry over. The ten ingredients (Part 4) specify what that grammar must be able to express regardless of its specific form.
§2.5 The Bridge
The CGG bridges human and machine language through three mechanisms:
Classification. Every artifact is classified into one of three classes. The classification determines which governance surfaces apply. This is the first bridge: a human decision about what kind of artifact this is becomes a machine-enforceable constraint on how it is processed.
Grammar mapping. Directive and descriptive records map their content to the universal grammar. Each primitive in the grammar has a defined operational meaning. This is the second bridge: human prose about intent, evidence, authority, and constraints becomes machine-parseable structure.
Perimeter governance. Hosted records carry a thin perimeter of governance-managed fields (identity, authority, lineage, gate, cross-references) even though their internals are domain-formatted. This is the third bridge: artifacts that cannot be grammar-mapped are still traceable, authority-bound, and cross-referenceable within the governance system.
Together, these three mechanisms ensure that no governed artifact is invisible to the governance system, while respecting that different artifacts require different structural treatment.