Scientific content · draft v0.1Scope · umbrella SIF frameworkCanonical terminology · currentLast reviewed · 4 October 2026

Scientific reference

Framework

The Structural Invariant Framework (SIF) is a research framework for deterministic, evidence-backed inspection of specification-declared invariants and of the governing support on which bounded admissibility decisions depend.

Definition and scope

A bounded, evidence-backed inspection framework

SIF separates variable generation or transformation from external decision authority. It evaluates governed entities, declared relations and states against an applicable governing specification and records the evidence for each bounded decision.

SIF is not a general correctness oracle, a formal proof system, a certification regime or a general translation-quality metric.

Scientific problem

Plausibility does not settle admissibility.

A generated or transformed result may remain plausible while changing information that a governing specification treats as non-negotiable.

A decision can also lose its governing support after an upstream source, evidence item, configuration, relation or authority state changes.

SIF therefore separates two assurance questions: whether the produced result satisfies what is currently governed, and whether what currently governs still has valid support.

Governing model

From source to decision evidence

Bounded decision chainEach transaction applies only its declared scope and applicable checks.
  1. Governing source
  2. Applicable governing specification
  3. Admitted input / representation
  4. Applicable checks
  5. Deterministic decision
  6. Decision evidence

The governing source establishes where a controlled requirement comes from. The applicable governing specification represents what is governed, the relations and states that matter, the admissible realizations and the decision limits for a declared scope.

The deterministic mechanism then applies only the checks that are applicable to the admitted transaction. The decision remains bounded by specification coverage, the admitted representation and the evidence available to the transaction.

Governed objects

Entities, relations and states

A governed entity may be a technical concept, value, record, source, configuration item or other represented object. A declared relation may specify dependency, logical structure, source support, authority or another decision-relevant connection. A governed state may include accepted, affected, blocked, satisfied, violated or unresolved conditions, depending on the applicable mechanism.

Relations are first-class verification objects. Correct entities alone are not sufficient if the relation between them has changed.

Specification, Registry and authority

The specification governs a declared scope.

An applicable governing specification

An applicable governing specification is a versioned, source-grounded representation that defines what is governed and what decisions are permitted in a declared scope.

A Registry may govern selected concept identities or canonical forms, but it is only one component of a broader specification. The broader specification may also include protected carriers, context and applicability rules, governed values, declared relations, states, constraints, evidence requirements and decision limits.

Source authority
Identifies where a requirement comes from.
Decision authority
The deterministic mechanism that applies the admitted specification within its declared scope.

Structural invariants and verification

Deterministic inspection of declared requirements

A specification-declared structural invariant is a required property of a governed entity, relation or state within the applicable scope.

Structural verification is deterministic inspection of applicable specification-declared invariants against admitted input and recorded evidence.

This use of verification is deliberately bounded. It does not mean formal proof of a complete system, total semantic correctness or general quality assurance.

Two assurance surfaces

Output and governing-state assurance

Output Structural Verification

Does the result satisfy what is currently governed?

This branch includes SIF 1.0, SIF 2.0 and SIF Inspector. It operates on a produced or transformed result under an applicable governing specification.

Governing-State Assurance

Does what currently governs still have valid support?

This branch is developed through SIF Governance. It evaluates changes in represented sources, evidence, relations, states and authority and derives bounded consequences.

Applicability and context qualification

Detection alone is not authority.

Applicability asks whether a declared rule or invariant belongs to the current transaction. Context qualification asks whether an observed surface actually represents the governed meaning required for an intervention.

A lexical or structural match can exist without authorizing a governed action.

SIF 2.0 implementation note

Context Gate

Context Gate is the deterministic pre-intervention specificity safeguard used in SIF 2.0. It prevents a detected lexical match from authorizing intervention unless the governed domain meaning is supported by the applicable context.

Context Gate does not establish overall semantic correctness and is not a universal component required in the same form in every SIF implementation.

Decision and bounded intervention

The mechanism preserves unresolved support.

A SIF transaction may resolve to an admissible state, bounded authorized intervention, review or escalation, rejection or blocking, or an unresolved / OPEN state, depending on the applicable mechanism.

A repair or intervention does not define truth. It is permitted only when the governing specification authorizes a local deterministic correction within the declared scope.

OPENInsufficient evidence remains unresolved.

Coverage and representation boundary

Two limits on what a decision can establish

Specification coverage

A SIF decision can address only what the applicable specification is capable of representing and checking. Absence of an applicable check is not evidence of correctness.

Representation fidelity

Deterministic execution operates on an admitted representation. A reproducible runtime can therefore execute the consequences of an incomplete or unfaithful representation exactly.

Representation fidelity asks whether the admitted representation preserves the source-grounded entities, relations, states and claims required for the intended decision. This boundary is investigated explicitly in SIF Governance.

Terminology

Distinctions that bound interpretation

Specification ≠ Registry
A Registry may govern selected concept identities or canonical forms; it is only one component of a broader specification.
Relation ≠ dependency
Dependency is one relation class. Logical, source-support and authority relations may also matter.
Detection ≠ applicability
Detection identifies a surface. Applicability asks whether a declared rule or invariant belongs to the current transaction. A match alone does not authorize intervention.
Applicability ≠ context qualification
A rule may be relevant to the transaction while a detected surface still lacks the governed meaning required for intervention.
Structural admissibility ≠ overall correctness
Structural admissibility concerns applicable required invariants and prohibitive constraints under the declared specification; it is not overall correctness.
Execution correctness ≠ representation fidelity
Deterministic execution can reproduce the consequences of an admitted representation without establishing that the representation is complete or source-faithful.
Repair ≠ authority
Intervention is permitted by the specification; the repair mechanism does not define canonical truth.
OPEN ≠ failure
OPEN preserves insufficient support rather than inventing certainty.
Domain transfer ≠ universal portability
Architecture reuse and specification sufficiency are separate empirical questions.

Short examples

Relations, context and unresolved states

ConceptMinimal illustrationWhat it shows
Relation fidelity“7 years OR 1000 hours, whichever occurs first” mutated to “AND”.Logical relations are part of governed meaning, not implementation detail.
Context qualificationA lexical match such as “uncertainty” occurs outside the governed metrological meaning.Detection alone cannot authorize intervention.
OPENEvidence is insufficient to establish a dependency or final state.Unresolved support must remain explicit rather than being converted into unsupported PASS or INVALID.

Evidence and reproducibility

Claims remain linked to bounded evidence.

SIF evidence is versioned, scope-bounded and claim-specific. Each evidence item records the question tested, the declared or frozen configuration, the evaluated material, the observed result, the supported claim and the boundary beyond which the result is not generalized.

Engineering release status, publication status and scientific evidence status are tracked separately.

View Evidence