Glossary: capture, contamination, collisions
This family groups the exogenous phenomena that distort the interpretation of an entity by AI systems (LLMs, generative engines, agents, RAG). Here, the problem is not merely the internal quality of a site or corpus: it is the external dynamics of signals, co-occurrences, entity confusions, and semantic dominance effects.
Each entry links to: a canonical definition, a framework (if applicable), and related pages (doctrine, clarifications, stabilization frameworks).
Quick access
Terms in the “capture, contamination, collisions” family
Interpretive capture
Situation where an actor, a set of sources, or a semantic neighborhood imposes a dominant framing, to the point of diverting the interpretation of an entity or concept in AI responses.
- Definition: Interpretive capture
- Doctrine: Exogenous governance: stabilizing the external graph
- Framework: Exogenous governance: external graph stabilization
Neighborhood contamination
Pollution of an entity by its dominant co-occurrences: AI ends up defining the entity by what surrounds it, rather than by its canon.
- Definition: Neighborhood contamination
- Doctrine: External coherence graph: mapping the neighborhood
Interpretive collision
Fusion or confusion between two distinct entities when their signals are sufficiently close to be mixed by AI (synthesis hallucination, cross-attribution, feature blending).
- Definition: Interpretive collision
- Related definition: AI disambiguation
Interpretive invisibilization
Information exists (indexed, accessible), but does not exist in the generated response: it is discarded by the model, by the retrieval, by the framing, or by a competing authority.
- Definition: Interpretive invisibilization
- Framework: Interpretive governance for AI agents
Related frameworks and pages (recommended)
Previous page: Glossary: evidence, audit, and observability
Next page: Glossary: agentic, RAG, environments
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.
How to read this lexical family
This family deals with the ways an entity or concept can be interpreted through the wrong neighborhood. Capture, contamination and collision are not identical. Capture occurs when an outside framing takes control of the interpretation. Contamination occurs when adjacent signals distort the target. Collision occurs when entities, roles or concepts are merged or confused.
The family is especially important for people, brands, niche doctrines and emerging categories. In those contexts, the system may have just enough evidence to generate a plausible answer, but not enough evidence to keep the entity separate from nearby meanings.
Typical misreadings
A common mistake is to treat contamination as a visibility problem only. The entity may be visible and still be misread. It can be cited, mentioned or recommended while being placed in the wrong category, associated with the wrong competitors or described through outdated external framing.
Another mistake is to assume that disambiguation is solved by adding more content. More content can help, but it can also intensify contamination if the corpus does not define boundaries, exclusions, entity relations and canonical roles.
Use in audit and routing
Use this family when outputs confuse a person with another entity, merge a doctrine with a market category, import a neighboring brand’s attributes or let a hostile or obsolete framing dominate the answer. The audit should map the semantic neighborhood before proposing correction.
For routing, this family should support entity disambiguation, semantic architecture, reduction of semantic collisions and representation gap audits. It should not be reduced to generic brand monitoring.