Glossary: drifts and interpretive inertia
This page groups the phenomena that degrade the fidelity of an interpretation produced by AI systems (LLMs, generative engines, agents, RAG) when meaning drifts, smooths, or freezes. These phenomena are not isolated “bugs”: they result from a probabilistic reconstruction of meaning, fed by partial signals, successive aggregations, and unstable contexts.
Each entry links to: a canonical definition (if it exists), a framework (if applicable), and related pages for deeper understanding.
Quick access
Terms in the “drifts and inertia” family
Interpretive hallucination
Production of a plausible but unenforceably anchored response, often stabilized by form rather than by evidence.
- Definition: Interpretive hallucination
- Clarification: Hallucinations, attribution, and interpretive risk
Interpretive smoothing
Tendency of an AI system to erase rough edges, nuances, and negations in order to fit a concept into a standardized category.
- Definition: Interpretive smoothing
Interpretive inertia
Persistence of a prior interpretation, even after modification of sources, due to the progressive stabilization of an algorithmic “truth”.
- Definition: Interpretive inertia
Interpretive remanence
Reappearance of an old interpretation in certain contexts, even when a correction seems established elsewhere.
- Definition: Interpretive remanence
- Doctrine: Version power
Interpretive tail
Intermediate phase where a correction progresses unevenly: some outputs correct themselves, others remain frozen or ambiguous.
- Definition: Interpretive tail
State drift
Divergence between a real state (price, availability, policy, status) and the state returned by AI, when an update does not propagate.
- Definition: State drift
Compliance drift
Progressive gap between expected constraints (canon, rules, response conditions) and observed outputs, despite an apparently stable documentary base.
- Definition: Compliance drift
- Doctrine: Interpretive observability
Recommended links
Next page: Glossary: canon, authority, non-response
Phase 6 routing: semantic stability layer
This page now routes toward the phase 6 canonical layer for semantic architecture and entity stability: semantic architecture, entity disambiguation, entity collision, semantic neighborhood, semantic contamination, framing stability, cross-system coherence, and interpretive drift.
These links clarify the difference between entity separation, neighborhood influence, contamination, drift, and cross-system comparison.
Phase 9 routing layer: memory, persistence, remanence, and correction
This page now routes stateful interpretation questions toward the phase 9 canonical layer: memory governance, agentic memory, memory object, persistent assumptions, controlled forgetting, stale-state handling, surviving authority, interpretive remanence, interpretive inertia, version power, state drift, and correction resorption.
The routing rule is direct: do not infer current authority from persistence alone. A memory object, old citation, surviving source, retrieved fragment, or previous answer must pass freshness, authority, traceability, and correction-resorption checks before it can govern a new response or action.
How to read this lexical family
This family describes the temporal behavior of meaning. Drift is movement. Inertia is resistance to correction. Remanence is survival after the source or framing should have lost authority. Together, these terms explain why a representation can remain wrong even after the correct page has been published.
This is crucial for AI search and answer systems because they often operate on delayed, partial or mixed states of the web. A system can combine a current page with an older citation, a cached summary, an outdated external mention or a persistent assumption.
Typical misreadings
A common mistake is to treat drift as a sudden hallucination. Many drift patterns are gradual. The answer becomes less faithful through small substitutions: a role becomes a category, a doctrine becomes a service label, a limitation disappears, an external source becomes more salient than the canon.
Another mistake is to assume that correcting the canonical page immediately corrects the state of the web. Inertia means that older interpretations can survive in snippets, summaries, model memory, third-party pages, backlinks, entity graphs and user prompts.
Use in audit and routing
Use this family when monitoring how an entity, doctrine, brand or corpus changes across systems over time. The audit should compare outputs by model, date, query form, source citation, answer structure and treatment of old claims.
For routing, this family supports drift detection, memory governance, correction resorption, state drift and interpretive sustainability. Its function is temporal: it explains how representations move, persist and resist correction.