SIF Glossary

A source-qualified model accepted for a particular Governance evaluation scope and interface. Admission does not prove that all source meaning was represented faithfully.

AIP evaluates correct governed anchor instances relative to the expected instances in the declared evaluation subset. SIF 1.0 used the name Anchor Identity Precision; SIF 2.0 RC3 uses anchor-instance accuracy. The exact denominator and scoring conditions are defined by each experiment.

A Registry-governed technical concept with a defined identity and authorized canonical form. An anchor instance is an expected occurrence of that concept in evaluated material.

Whether a declared rule or invariant applies to a specific input, event and scope. Finding a lexical match does not by itself establish applicability.

In Governance research, a mapping from a qualified representation to executable control inputs under a specified contract. A generated binding is not expert semantic approval.

Consequences supported by declared dependencies within a qualified scope. Unknown affected-set membership stays unresolved rather than being guessed.

A local action or correction permitted by the governing specification under defined conditions. Repair does not independently establish canonical authority.

A specification-authorized realization of a controlled concept in the applicable domain and context. A model's preferred wording is not its own canonical authority.

The SIF 2.0 deterministic pre-intervention safeguard requiring domain-specific contextual support before a lexical match can trigger intervention. It does not verify general semantic correctness.

Checking that an observed expression has the governed meaning required for an intervention. A rule can apply even when a particular lexical match is not context-qualified.

A supported relationship in which one object or decision depends on another. Not every traceability reference, source-support relation or link is an engineering impact dependency.

Applying specified checks or consequences reproducibly to admitted inputs. Determinism does not establish that the underlying representation or governing specification is complete or correct.

A structured set of evaluation records connecting cases, outcomes and reported metrics. A redacted public ledger permits metric checks without revealing original texts or private runtime code.

A control activation or intervention where the specified negative case expected none. The measured rate applies to the declared negative-control set.

A fixed-version input, output or evidence item preserved for a named evaluation or replay. A frozen demonstration is not new live execution.

The external, versioned, source-grounded rules and admissibility conditions against which a result is checked in a declared scope. A Registry is only one component.

Recalculating a published numerical result from released evidence fields and documented formulas or scripts. It does not rerun SIF or independently validate original annotations.

A deliberately defined case in which an intervention or governed concept should remain inactive, including NULL cases without an expected governed anchor. It is not a positive accuracy test.

An explicitly unresolved state when the available source or dependency support is insufficient. OPEN is not automatically a runtime defect, an invalid relation or a release authorization.

Traceable origin, identity and version of records and sources supporting a claim. A cryptographic hash establishes artifact identity, not the factual truth of its content.

A versioned resource for selected governed concepts and authorized canonical forms. It does not replace the broader governing specification.

What an independent researcher can check using the released code and data. Metric recomputation, static replay and a full experimental rerun are different levels of verification.

Satisfaction of applicable specification-declared structural requirements under stated conditions. It does not imply complete translation correctness or overall quality.

A specification-declared property of a governed entity, relation, or state that must hold within the applicable scope.

A technical object such as a value, unit, formula, constant, symbol or tag protected from unauthorized change under the specified evaluation. Literal preservation does not establish mathematical equivalence.