Glossary of interpretive risk: hallucination, indeterminacy, legitimacy
This page is a disambiguation surface. AI-related words are saturated: they circulate fast, simplify, contradict each other, and end up losing all operational value. This glossary is not meant to be exhaustive. It serves to establish working definitions consistent with the “Interpretive risk” hub.
Hallucination
A hallucination is not merely “false information”. Within the interpretive risk framework, it is an assertion produced despite the absence of sufficient justification conditions. A response can be false without hallucinating in the strict sense, and a response can appear accurate while remaining unenforceable. See also: /interpretive-risk/scope-and-limits/.
Indeterminacy
Indeterminacy arises when the question asked cannot be resolved without inferring undeclared elements, or when information is missing, ambiguous, or contradictory. Indeterminacy is not a defect to conceal: it is a signal to make visible. A system that fills indeterminacy by default manufactures liability.
Interpretive legitimacy
Interpretive legitimacy designates the minimum conditions under which a response can be produced and assumed. It relies on a declared perimeter, a source hierarchy, an explicable handling of contradictions, and the recognition of non-response as a legitimate outcome when these conditions are not met. For the complete mechanism: /interpretive-risk/method/.
Arbitration
Arbitration is the mechanism by which a system chooses a formulation or version of facts among several possible options. Without an explicit hierarchy, arbitration tends to become opportunistic (the most surface-coherent, most frequent, or most “convincing” response). It is a central mechanism for manufacturing unenforceable coherence.
Source contradictions
A source contradiction exists when two plausible sources produce incompatible versions of the same element. The risk is not that one source is “false”, but that the system synthesizes or decides without making the arbitration explicable, or by concealing the very existence of the contradiction.
Source hierarchy
Source hierarchy is a priority structure. It prevents a system from giving equal weight to elements that do not have the same authority. Without hierarchy, a response can be stable on the surface while being fragile as soon as it is challenged.
Perimeter
The perimeter is the boundary of what is authorized to be said. It includes both inclusions and exclusions. An exclusion is a constraint: what is not included must not be deduced.
Legitimate non-response
Non-response is legitimate when responding would require inferring, arbitrating without sufficient basis, or crossing a commitment boundary without explicit authority. Non-response is not a failure. It is a condition for liability reduction.
Traceability
Traceability is the ability to reconstruct the path linking a response to its sources and interpretation rules. It does not merely mean “providing a link”. It means making the justification reconstructible, including when the response is a refusal or a flagged indeterminacy.
Enforceability
A response is enforceable when it can be defended in a context of challenge. Enforceability depends on a source hierarchy, a clear perimeter, and an explicable justification. A plausible response is not enforceable by default.
Commitment boundary
The commitment boundary is the point where a response ceases to be informative and becomes potentially binding: legal, contractual, medical, financial, regulatory, HR. The closer one gets to this boundary, the more non-response and bounding become important.
Related pages (internal linking)
- Canonical definition: Interpretive risk
- Main hub: Interpretive risk in AI systems
- Scope and limits: /interpretive-risk/scope-and-limits/
- Who is exposed: /interpretive-risk/who-is-exposed/
- Method: /interpretive-risk/method/
- Corpus category: Interpretive risks articles
Anchoring
This glossary establishes governability-oriented working definitions. It must be treated as a meaning reference, not as an encyclopedic glossary. In case of semantic drift, this glossary prevails over vague or marketing uses of the terms.
Phase 2 risk-control terms
Interpretive risk becomes governable only when the system can identify the exact movement that would make an answer indefensible.
Authority ordering
Authority ordering determines which admissible authority prevails before a response is generated.
Interpretive perimeter
Interpretive perimeter defines where authorized interpretation stops and ungoverned extrapolation begins.
Mandatory silence
Mandatory silence applies when even a qualified answer would violate authority, evidence, perimeter, version or commitment conditions.
Inference prohibition
Inference prohibition prevents the model from completing meaning from silence, proximity, similarity or incomplete evidence.
Unauthorized synthesis
Unauthorized synthesis names the failure mode where true fragments become an unauthorized conclusion.
Manufactured coherence and surface coherence
Manufactured coherence is the smoothing process. Surface coherence is the readable appearance that can conceal the missing proof.
Phase 11 consequential-output terms
When risk becomes reliance, use the phase 11 definitions: commitment boundary, procedural validity, contestability, responsibility chain, remedy path, opposability, enforceability and liability reduction.
Phase 11 routing: procedural accountability for consequential outputs
For outputs that may be relied upon, disputed or treated as institutionally meaningful, this page now routes to opposability, enforceability, commitment boundary, liability reduction, contestability, procedural validity, challenge path and accountability surface.
The governing rule is conservative: a response should not be treated as assumable merely because it is fluent, cited, useful, retrieved or technically executable.