Each ingredient is defined, its absence condition is stated, and implementation guidance is provided. The per-primitive governance testing for each ingredient is in Part 5 (the governance matrix), not here. This part answers what and why; Part 5 answers how it is governed.
§4.1 Purpose
Primary primitive: Intent. Charter record: the apex objective.
Purpose is teleology — what should be, the directional bias that makes action evaluable. Without purpose, no action can be assessed as better or worse, every risk is equally concerning because there is no objective against which to measure severity, and audit findings have no criteria.
A governed world model's purpose takes the form of an apex objective — a single statement from which all other objectives inherit. The apex objective must satisfy the four-fold translation test (as loss, as reward, as utility, as likelihood — a statement that fails any reading is collapsing meaning across audiences). The apex objective must also carry a negation clause — what the purpose explicitly excludes — because drift is detectable only when you have recorded what the system committed to not doing.
§4.2 Distinctions
Primary primitives: Identifier, Namespace. Charter record: ontology.
Distinctions is the capacity to draw a boundary between this and not-this. It is prior to classification — it is the act that produces the things classification operates on. Without distinctions, identity collapses, namespace conflicts are invisible, and every other ingredient loses its referential foundation.
A governed world model implements distinctions through identifiers (stable addresses that persist across content revisions, deterministic resolution) and namespaces (scoped vocabularies where terms have defined meanings, inherited from parent with tighten-only override rules). The operative terms of any governance commitment must be namespace-resolved; an unresolved operative term governs nothing.
§4.3 Entities
Primary primitives: Entity, Authority, Capacity, Account. Charter record: network.
Entities are things that persist across distinctions — identity maintained through change. Without entities, authority cannot be assigned, accountability cannot be resolved, capacity cannot be assessed, and delegation cannot function.
A governed world model implements entities through entity identity (persistent across state changes), authority binding (what the entity is permitted to decide, under what delegation), capacity assessment (what the entity can realistically handle), accountability binding (structural traceability through the delegation chain, not organizational chart), and an actor taxonomy (a closed taxonomy of actor types with defined authority ceilings — human, AI-automation, AI-agentic, AI-assistive).
Entities separate into two strata: an accountability stratum (authority and account — located, individuated, non-poolable) and a capability stratum (entity and capacity — poolable, summonable on demand). A capability is invoked under an accountable entity's authority for a single interaction and never inherits standing authority; this is what keeps accountability located even when capability is shared. The separation is forced by the rule that computational actors hold only delegated, never inherent, authority (§3.1, axioms 4–5).
§4.4 Constraints
Primary primitives: Constraint, Commitment. Charter record: controls.
Constraints are constitutive limits, not guardrails. A guardrail is advisory; a constraint is structural. Constraints operate through prohibitions (what the system will not do) and commitments (what the system binds itself to do). Without constraints, drift is undetectable and governance is advisory.
A governed world model implements constraints through negation clauses (every directive record names at least one class of action that would constitute drift even if the positive statement were satisfied), commitment registers (auditable inventory of structural bindings that produce findings when unmet), and constraint-first validation (constraints validated before content when processing new records or revisions).
§4.5 Time
Primary primitives: Cycle, Decision. Charter record: strategy.
Time is ordering with directionality — the axis along which recursion operates. Without time, decisions have no sequence, learning is impossible, and governance collapses to a single frame with no history, no trajectory, and no accountability for how the current state came to be.
A governed world model implements time through cycles (structural commitments to re-examine governance state at defined intervals — the forcing function that prevents governance decay) and decisions (discrete temporal events where governance state changes, each carrying a before-state and after-state). Cycles produce decisions; decisions produce temporal ordering. Together they give the world model temporal legibility.
§4.6 Observation
Primary primitives: Orientation, Environment Interface, Context, Interpretation. Charter record: interface.
Observation is a model from somewhere, for someone, with limited information. Observation carries four primitives because seeing is structurally complex: orientation (position and directional bias), environment interface (what can be perceived from that position), context (surrounding and conditioning factors), and interpretation (how signals become meaning).
A governed world model implements observation through orientation declaration (every observation-producing record declares its position and biases), interface specification (the perception surface — what the system can and cannot perceive), context capture (conditions at observation time, immutable from capture), and interpretive transparency (how signals became meaning, made explicit and auditable).
§4.7 Relations
Primary primitives: cross-cutting via object properties. Charter record: ecosystem.
Relations are the connective tissue between entities — what makes a world a world rather than a collection of isolated objects. Relations has no primary primitives of its own because it operates through the relationships between primitives. Without relations, governance is local only — each entity, record, and decision exists in isolation, and constraint propagation, authority delegation, and accountability resolution all fail for lack of channels.
A governed world model implements relations through hierarchical relations (parent-child, directed acyclic graph terminating at apex — carry authority delegation, constraint inheritance, reconciliation obligations), class membership relations (cross-cutting, carry capability inheritance, conflict-resolution weights, forcing-function discipline), and cross-reference relations (named associative links — the audit trail).
§4.8 Transition Logic
Primary primitives: Work, Activation. Charter record: workflow.
Transition Logic is the rules of change. It is second-order recursive: the rules of change themselves change. Without transition logic, governance declarations are inert — purpose exists but nothing moves toward it, constraints exist but nothing tests against them, evidence obligations exist but nothing produces evidence.
A governed world model implements transition logic through work specification (what governed activity produces, what inputs it requires, what constraints it operates under, what evidence it generates — to re-performance standard), activation governance (trigger conditions, review triggers, termination conditions — preventing premature start, indefinite continuation, and ungoverned re-activation), and transition rules (the rules by which work moves between states, themselves governed with full lineage).
§4.9 Memory
Primary primitives: Evidence, Learning. Charter record: knowledge.
Memory is retained, accessible history. Without memory, every governance cycle starts from zero — findings are forgotten, corrective actions are untracked, indicators have no trend line, and the three follow-up obligations are unanswerable.
A governed world model implements memory through evidence retention (at granularity sufficient for re-performance and duration sufficient for cycle-over-cycle comparison), lineage preservation (prior versions, amendments, supersession chains), learning lifecycle (emergence, capture, triage, crystallization or decay — each a governed transition), and knowledge representation (retained evidence in a form supporting retrieval, comparison, and audit).
§4.10 Uncertainty
Primary primitives: cross-cutting via truth-type system. Charter record: queries.
Uncertainty is what the model does not know, represented structurally. Uncertainty is the second cross-cutting ingredient. Without it, all claims carry equal epistemic weight, audit cannot evaluate evidence quality, and the system cannot prioritize investigation because its unknowns are invisible.
A governed world model implements uncertainty through truth-type classification (authoritative, declared, derived, opaque — every claim carries an epistemic status), investigative query discipline (acknowledged uncertainty becomes structured investigation with lifecycle terminating in a decision), and uncertainty propagation (truth types propagate through derivation chains — a conclusion cannot be more authoritative than its least authoritative input).