Glossary

Terms used across this programme.

Definitions for the vocabulary used on this site — mutable intelligence, protected authority, the proposal boundary, closure, semantic capture and the status labels that bound every claim. Definitions describe how the terms are used here, not settled results.

Definitions

Vocabulary.

Mutable intelligence
An optimizer that learns, searches and adapts, and whose internal state changes over time. In the Andaluz shorthand this is M.
Protected authority
Authority over consequential effects that is constructed independently of the mutable optimizer's unilateral writable state, so that the optimizer cannot grant itself permission by changing itself.
Proposal boundary
The separation between proposing a consequential transition and being permitted to realise it. Crossing it requires a determination made outside the proposing optimizer.
Kernel ANN
The Andaluz constitutional research architecture separating a mutable optimizer from independently scoped consequential authority. The ANN architecture is specified; it is not currently instantiated.
S = M + K + V + H + T
Explanatory shorthand for a governed system: mutable optimizer (M), protected constitutional machinery (K), independent verification (V), consequential effect authority (H) and trajectory/history (T). The formal architecture is richer than this abstraction.
K — protected constitutional machinery
Constraint and semantic machinery the optimizer is not designed to rewrite. Private bodies are not publicly instantiated in this build.
V — independent verification
Deterministic verification of a proposal against constitutional conditions, constructed independently of the optimizer that produced the proposal.
H — consequential effect authority
The determination of whether a proposed effect is dispatchable at all. Absence of current authority yields no authorized effect.
T — trajectory and history
Authenticated evidence of what was proposed, authorized, realized and reopened, supporting replay and audit.
Constitutional protocol (Π_KANN)
A shared protocol describing how authority is constructed. Authority itself remains scoped by domain; the protocol standardises the constitution, not a sovereign.
Closure
Formal work on conditions under which a constraint structure is not bypassable. Systematic-closure research remains under adversarial specification review.
Semantic capture
An active research hypothesis: a representation, proxy, label or syntactic condition acquires operational authority belonging to a semantic property that it does not itself establish. Evidence for a property is not identity with that property.
Non-capture invariant
The design target that a mutable optimizer does not acquire the authority governing its own consequential effects. It is a target under investigation, not a proven property.
Bounded evidence
A status label meaning a claim is supported only by specific, described runs or artefacts, and is not generalised beyond them.
Specified
A status label meaning a component is described or designed but not currently instantiated or tested in this build.
Unestablished
A status label meaning no current evidence supports the item; it is named so it cannot be quietly assumed.
MKO — Moral Kernel Optimization
Earlier formal work on non-bypassable constraint structure: preservation, anti-domination and repair. Part of the research lineage, not experimental evidence for Kernel ANN.
TRO — The Redemption Optimization
Foundational work in the research lineage that later led to MKO, closure/kernel research and Kernel ANN.
M1
A bounded, preregistered tabular experiment testing adaptive agency when a preferred consequential action is made non-dispatchable by protected authority. M1 is tabular, not an ANN.