The substrate primitives

The 9 domain-agnostic primitives that constitute the LUU strategic-judgment substrate, plus the falsifiable substrate criterion and the strategic-extensions exclusion list.

What the substrate is

The strategic ontology is two things at once. Descriptively it is an ontology — it names what kinds of strategic-analytic artifacts exist. Operatively its kernel is a judgment substrate — the minimum load-bearing primitive set without which calibrated, auditable, adaptable judgment-under-uncertainty cannot proceed in any domain.

The substrate is a projection of the ontology, not the whole ontology. Strategic-domain extensions (multi-actor models, time horizons, goal hierarchies) ride above the substrate, not within it.

The nine primitives

Primitive	Role
`Claim`	Proposition asserted with calibrated confidence
`Evidence`	Typed signal bearing on a claim, with weight and provenance
`Source`	Origin of evidence with credibility attributes
`Alternative`	Competing hypothesis or option to a parent claim
`UncertaintyEstimate`	Calibrated probability or distribution on a claim
`ChangeReason`	Recorded rationale for any state mutation
`Attribution`	Binding between claim, evidence, and source
`DecisionFrame`	The question, options, and stakes a judgment is over
`InformationGap`	Known unknowns acknowledged in a frame

The substrate criterion (falsifiable)

A primitive P belongs in the kernel iff there is no coherent judgment-under-uncertainty process that omits P and still satisfies calibration, auditability, and adaptability.

Test by removing P, attempting to construct a judgment pipeline, and observing whether all three properties survive.

Programmatic check

import { runSubstrateCriterion } from '@luu/judgment-substrate';

const result = runSubstrateCriterion();
console.log(result.ok); // true if all 9 primitives defend
                         // at least one minimum property

What's NOT in the substrate

The following are strategic extensions — they ride above the substrate and are not required for non-strategic judgment domains (e.g. medical differential diagnosis, legal tort liability, scientific peer review):

Actor, ActorRelationship, TheoryOfMind, Goal, GoalHierarchy, TimeHorizon, Scenario, Storyline, EnvironmentalDynamic, NarrativeTension, ExchangeProfile, PerspectivePack, RigourTier, PostureMode, DomainTopology, MultiHorizonForecast, SimulationMode.

Adding any of them to the kernel requires demonstrating that omitting them collapses calibration / auditability / adaptability for any judgment domain, not just strategic cognition.

Empirical validation

The substrate has been empirically validated via a 5-phase ladder with pre-registered acceptance thresholds locked 2026-05-02:

Phase 0 — Data audit: minimum row counts and coverage requirements per primitive
Phase 1 — Predictive validity: regression with cluster-robust SE, robustness checks
Phase 2 — Hard-cap calibration: threshold sweep across 30 / 35 / 40 / 45 / 50
Phase 3 — Marginal contribution: multivariate regression with VIF + partial-R²
Phase 4 — Controlled ablation: primitive-keyed ablation harness with smuggling detector
Phase 5 — Cross-domain: medical, legal, scientific instantiations

See the upstream repo for the full pre-registration document and acceptance criteria.

Why this matters

Falsifiability: a primitive is testable for membership. Reviewers can ask whether its removal collapses the system.
Defensibility: "judgment substrate" is a stronger claim than "yet another ontology" and harder to copy.
Architectural clarity: the boundary is load-bearing in code, not just in prose.
Benchmark separability: substrate competence (Tier 1) tests independently of strategic-extension competence (Tier 2), so failures localise.

JudgmentEnvelope<T> — the universal cross-boundary container that carries substrate metadata
Exchange profiles — the canonical interchange formats every tool emits
Tool catalog — JSON catalog mapping every tool to the primitives it produces