The Architecture

Six constraints. Three axes. One criterion for semantic validity.

The Core Claim

Any semantically complete claim must satisfy exactly six constraints. Not approximately. Not usually. Exactly six.

Fewer than six and the claim has gaps—places where it can fail without warning.

This isn't theory. It's geometry.


The Six Constraints

#ConstraintQuestionFunction
1ReferentialWHAT is being claimed?Establishes determinate content
2ContextualUnder what CONDITIONS?Specifies scope and applicability
3PremissiveOn what GROUNDS?Provides warrant and support
4InferentialWHY does this follow?Shows logical connection
5ConstrainingWhat are the LIMITS?Defines boundaries and exclusions
6TeleologicalWhat is this FOR?Establishes purpose and relevance

Every valid claim answers all six. Every invalid claim is missing at least one.


Why Six?

Consider the simplest three-dimensional enclosure: a tetrahedron.

Why Six? The tetrahedron: simplest three-dimensional enclosure 1. Referential 2. Contextual 3. Premissive 4. Inferential 5. Constraining 6. Teleological Claimant (who asserts) Subject (what is discussed) Grounds (what supports it) Purpose (what it is for) = Component (vertex) = Constraint (edge) — the relation between components Remove any edge → structure collapses → no enclosure → meaning leaks out

Figure 2

Remove any edge and the structure collapses. It no longer bounds a volume. It's no longer closed.

Semantic validity works the same way. The six constraints are the six edges. Remove any one and meaning leaks out—the claim can be true in ways you didn't intend, false in ways you can't detect.


The Three Axes

Every claim exists in a three-dimensional space:

The Three Axes of Semantic Space Every claim exists in three-dimensional space. Current systems flatten this to one. USER (who is asking) Expertise level Intent Context of use SUBJECT (what is claimed) Domain Certainty level Settled vs. disputed METHOD (how it is derived) Inference type Confidence warranted Deduction ≠ speculation Current systems: Input (one axis) → Output (one axis). Valid systems: track all three throughout.

Figure 3

AxisWhat It TracksWhy It Matters
USERWho's asking, expertise level, intentSame answer isn't right for everyone
SUBJECTWhat's being discussed, certainty levelDifferent topics require different rigor
METHODInference type, confidence warrantedDeduction ≠ speculation

Current systems flatten this to one axis: input → output.

Valid systems track all three throughout processing.


The Validation Process

The Validation Process GENERATED OUTPUT Any claim, response, or assertion SIX-CONSTRAINT VALIDATOR Referential — Is it determinate? Contextual — Are conditions specified? Premissive — Are grounds provided? Inferential — Does it follow? Constraining — Are limits acknowledged? Teleological — Is purpose clear? 6/6 PASS Output as valid Semantically complete <6 FAIL Revise, hedge, or refuse Incomplete — do not emit A system that refuses when it cannot satisfy the constraints is more trustworthy than a system that always produces output. Refusal is a feature, not a failure.

Figure 4

Claims that fail any constraint have three options:

  1. Revise — Fix the missing constraint
  2. Hedge — Acknowledge the incompleteness explicitly
  3. Refuse — Decline to output until constraint can be satisfied

Option 3 is a feature, not a failure. A system that refuses when it shouldn't answer is more trustworthy than one that always produces output.


Inference Type Discrimination

Not all inferences are equal. The architecture tracks four types:

TypeOperationConfidenceMarker
DeductionNecessary conclusion from premisesCertain (if valid)"This follows necessarily..."
InductionGeneralization from instancesProbable"Evidence suggests..."
AbductionInference to best explanationPlausible"The best explanation is..."
InterpolationPattern completion from structureRecognition-dependent"This appears consistent with..."

Current systems conflate all four. The architecture requires tagging. Users know what kind of inference produced the claim.


Closure Authority

Not every output should be finalized by the system. The architecture includes a closure registry:

Output TypeClosure Authority
Factual lookupSystem closes
CalculationSystem closes
Creative generationSystem closes
Medical adviceHuman review
Legal determinationHuman closes
Ethical judgmentHuman closes
Safety-criticalHuman closes
UncertaintyExplicit hedge

The system routes outputs to appropriate closure. It doesn't decide what humans should decide.


Patent Coverage

Ten patent families protect this architecture:

FamilyCoverage
ATetrahedral validity structure
BForm-preserving memory
CInference discrimination
DSemantic state signaling
EConstraint governance
FClosure authority
GIntegrated system
HSemantic condensation and reconstitution
IHexis-governed cognitive pedagogy
I-ExtExtended pedagogical and diagnostic applications

Provisionals filed. Full specifications available under NDA.


Next Steps

  1. Review THE PROOF — See how this dissolves known problems
  2. Study THE VOCABULARY — Learn the precise terms
  3. Contact for integrationthemis@echosphere.io
Six constraints. Three axes. One criterion. Valid or not. Now you can tell.