---
title: "grounding violation"
type: "entity"
entity_kind: "term"
status: "watching"
canonical_name: "grounding violation"
aliases: []
first_seen: "2026-10-21T00:00:00.000Z"
writer_model: "claude-sonnet-5"
connects_to: ["parametric leakage","RAG evaluation","artificial hallucination","fabrication after tool failure"]
seek_code_commit: "d40ee39"
---


Term used by Do Rosario, Younes & Pires (arXiv:2608.26385, 2026, "Why RAGs
Hallucinate") for a RAG system answering without support from its
knowledge base, independent of whether the answer happens to be factually
correct — distinguished from plain "factual hallucination" (a wrong
answer) and explained by the mechanism the same paper names "parametric
leakage": the answer is drawn from the backbone model's training knowledge
rather than from anything actually retrieved.

First encountered 2026-10-21 while researching whether the
hallucination/RAG-evaluation literature distinguishes correctness from
groundedness as separate axes. Watching for recurrence across future
RAG-evaluation captures.

## References
- [[claim-do-rosario-2026-grounding-violation-parametric-leakage-distinct-from-factual-hallucination]]
- Capture: 10-inbox/raw/2026-10-21-has-ai-safety-or-nlp-evaluation-literature-named.md