Interpretive variability designates the observable dispersion of machine-generated interpretations of the same entity, claim, source, product, organization, or category across controlled execution contexts.
The term is normative when it is used to qualify whether output variation remains inside an admissible corridor of meaning or creates a fidelity, source, category, or recommendation problem.
Definition
Interpretive variability must be assessed through observable outputs, declared context, source evidence, and comparison against a canon or baseline when one exists.
It may arise from retrieval differences, source selection, model behavior, query formulation, language, region, session state, time, downstream summarization, or delivery-layer fixation. It must not be reduced to randomness, ranking volatility, or personalization folklore.
Minimum distinction
- Variation: outputs differ.
- Interpretive variability: outputs differ in a way that may affect meaning, source use, category, proof, or recommendation.
- Interpretive drift: the difference moves away from a canon, baseline, or declared rule.
- Hallucination: the output asserts unsupported or false content.
Normative rule
Rule IV-1: no system, audit, or public claim may infer durable AI visibility or interpretive stability from a single query, model, context, region, or session.
A claim of stability requires repeated observation, preserved evidence, and a stated qualification method.
Out of scope
Interpretive variability does not prove that a crawler obeyed a policy, that a model used training data, that a ranking factor exists, or that an organization can guarantee uniform AI answers.
Delivery-layer note
Interpretive variability also includes cases where expected variation is suppressed by a delivery layer. Semantic caches, approved-answer stores, routing, or orchestration may repeat a prior reconstruction. Such repetition must not be interpreted as fidelity unless the delivery path and freshness state are known.