Governance Disciplines and Conformance Architecture

§3.0 Overview

This part defines the governance test surface for world model ingredients. It is organized in two layers:

Layer 1 — Five governance disciplines (§3.1-§3.5). Five cross-cutting disciplines that apply to every ingredient as structural tests. These are conceptual categories, not the primary test axis of the governance matrix. Each discipline maps to a COSO IC 2013 component, providing a bridge to existing internal control vocabulary. The five disciplines are Authority Mapping, Decision Boundary Specification, Constraint Propagation, Evidence Obligations, and Accountability Resolution.

Layer 2 — Conformance test architecture (§3.6). The nineteen DLP primitives — the same universal grammar that all governed records share (§2.4) — are the primary test axes for governance conformance. They organize into four conformance categories: Structural (cross-cutting invariants), Behavioral (actor and decision boundary constraints), Temporal (ordering and lifecycle), and Mechanical (per-primitive structural tests). This four-category framework is the native conformance architecture; the five governance disciplines are a COSO-bridged grouping overlay on it.

Why two layers? The five governance disciplines answer questions practitioners already ask ("who authorized this?", "where is the decision?", "what evidence?"). The nineteen primitives provide exhaustive structural coverage — no primitive left untested, no governance surface omitted. The disciplines are the conceptual on-ramp; the primitives are the structural test surface. Matrix D (Part 5) uses all nineteen primitives as its primary test axis because governance completeness requires testing every ingredient against every structural dimension, not just five.

§3.1 Authority Mapping

The question: Who is permitted to decide what, under what conditions, and by whose delegation?

Without authority mapping, every other governance artifact is unanchored. You cannot audit a decision if you cannot establish whether the actor had standing to make it. You cannot resolve a conflict if you cannot determine whose authority prevails. You cannot delegate if the delegation has no traceable source.

Authority mapping distinguishes between inherent authority (authority that exists by virtue of position) and delegated authority (authority that is explicitly granted, scoped, and revocable). Authority operates through a three-layer behavioral-contract architecture: actor type (what the actor is — constraining eligibility), role envelope (what the actor may do in a given role — narrowing the ceiling to a specific surface), and per-interaction contract (what the actor will do in this specific interaction — the most granular authority specification). Each layer can only narrow, never expand, the layer above.

Authority axioms: (1) Every governed record has exactly one authority holder. (2) Authority chains terminate at an apex constituting authority. (3) Delegation is revocable by the granting authority. (4) Computational actors cannot hold accountable or governance-level authority roles — their authority is always delegated, never inherent. (5) Accountability does not pool and cannot be summoned: the accountable holder of any decision resolves to exactly one located entity, and accountability cannot be aggregated across entities, transferred to a shared or on-demand capability, or held "by the system." Where a non-accountable capability is invoked under an entity's authority, that delegation is scoped to the single invocation and does not persist beyond it (ephemeral delegation).

Distinctive design choice: Model providers are not actors in the governance system. They are constraints — structural conditions under which actors operate.

§3.2 Decision Boundary Specification

The question: Where exactly in a process is a choice being made and an alternative path was available?

Governance is not about governing outcomes — it is about governing decisions. Every governance failure traces back to a decision that was either unnamed (nobody knew it was a choice), unowned (nobody knew who was choosing), or unrecorded (nobody can reconstruct what was chosen and why). A governed decision surface names the decision (what is being chosen between), the authority (who holds decision rights), the alternatives (what paths were available), and the record (what artifact captures what was chosen and why).

Decision boundaries exist at three scales: process-level (lifecycle gates, handoff mechanisms, disclosure gates), record-level (every governed record carries lifecycle fields that are machine-readable decision boundaries), and classification-level (assigning an artifact to a class determines which governance surfaces apply — classification is itself a structural decision). Decision provenance enforcement is the structural rule that every resolved question must produce a named decision artifact.

Reducibility of a decision. A decision is delegable when its requirements can be satisfied by demonstrated competence — a capability (human or computational) can perform it, and where accountability is required it may draft the decision for an accountable holder to confirm. A decision is irreducible when it requires an accountable signature: accountability cannot be evidenced by competence — it is held, never demonstrated — so by construction no capability, however competent, can satisfy it. This is the decision-boundary form of the authority axiom that computational actors hold only delegated authority (§3.1) and of the closure property that accountability cannot pool or be summoned (§3.1, §4.3). The diagnostic for why a decision demands an accountable signature — existential downside, genuinely ambiguous inputs, a values-laden trade-off, or external accountability — routes it to an accountable holder and locks it against full delegation.

§3.3 Constraint Propagation

The question: When two requirements interact, which takes precedence? When an exception is granted, where does it stop?

Governance systems layer requirements from multiple sources. When these requirements conflict, the governance system must have a composition rule or the conflict is resolved by whoever happens to be in the room — which is not governance, it is narrative power.

Constraint propagation specifies how requirements, tolerances, and prohibitions flow across boundaries through six named patterns: tighten-only inheritance (child constraints narrow but never loosen parent constraints), halt-and-escalate on conflict (conflicts route to the nearest authority with jurisdiction over both sources), explicit composition (interaction rules between constraint sources must be stated), negation propagation (commitments to what will not be done are traceable), lateral inheritance with conflict resolution (cross-class conflicts resolve through declared weights), and effective constraint set (the total constraint on any record is the union of inherited, class-membership, and locally declared constraints with precedence rules).

§3.4 Evidence Obligations

The question: What must be recorded, by whom, at what granularity, and what constitutes sufficient attestation?

The evidence obligation discipline specifies five structural commitments: the re-performance standard (documentation sufficient for an independent reviewer to re-do the analysis), the finding structure (condition, criteria, cause, effect, recommendation — five elements, not optional), the three follow-up obligations (implementation tracking, effect examination, assumption validation), the position/evidence separation (binding claims and their evidentiary basis are structurally distinct record classes), and evidence quality criteria (independence, relevance, validity, reliability — structural criteria, not aspirational principles).

Evidence pipeline (operational): For organizations running periodic governance assessments, the evidence pipeline flows: control objectives, controls inventory, testing, findings, corrective actions, tracking. Each step produces named evidence artifacts.

§3.5 Accountability Resolution

The question: When something goes wrong — or when an audit finds a gap — how is responsibility resolved and through what process?

Accountability resolution consolidates six structural elements: finding lifecycle (open, in progress, addressed, verified, closed — findings cannot be silently aged out), escalation path (halt-and-escalate is the default; routing is structural, not ad hoc), disposition tracking (cycle-over-cycle history of every corrective action), resolution provenance (every resolved question produces a named decision artifact), self-referential closure (the governance system must be able to examine itself), and governance-health measurement (indicators that track whether governance mechanisms are functioning, not just existing).

The commitment-breach protocol — what happens when a stated negation commitment is found to have been violated — remains the least specified element. The structural elements above provide the process infrastructure; the specific protocol requires further authoring as the framework matures.

§3.6 Conformance Test Architecture — The Nineteen Primitives

The five governance disciplines (§3.1-§3.5) are the conceptual vocabulary. The nineteen DLP primitives are the structural test surface. Matrix D (Part 5) uses these nineteen primitives as its test axis because structural completeness requires testing every ingredient against every primitive, not just the five that map most naturally to COSO.

The nineteen primitives organize into four conformance categories:

Structural conformance (S). Cross-cutting invariants that test whether the governance substrate itself is well-formed. Six test families:

Behavioral conformance (B). Actor and decision boundary constraints. Eight test families:

Temporal conformance (T). Ordering and lifecycle constraints. Four test families:

Mechanical conformance (M). Per-primitive structural tests. One test per Tier 1 primitive, verifying that the primitive's structural requirements are met at the implementation level:

Tier 2-5 primitives (identifier, entity, context, namespace, orientation, learning, activation, interpretation, environment-interface, cycle) are tested through the structural, behavioral, and temporal conformance categories rather than standalone mechanical tests. Their governance surfaces are embedded in the S, B, and T families above.

COSO overlay. The five governance disciplines (§3.1-§3.5) group sixteen of these nineteen primitives for COSO-bridge purposes (intent, work, and capacity have no COSO-component home and are tested directly through Mechanical conformance):

This overlay preserves the COSO bridge without constraining the test surface to five groupings. Sixteen of the nineteen primitives map to exactly one discipline grouping; intent, work, and capacity carry no COSO-component home and are tested directly through Mechanical conformance (M-intent, M-work, M-capacity) rather than a discipline. The grouping is navigational, not structural — the authoritative test surface remains the full nineteen-primitive matrix.