Semantic Grounding Index: Geometric Bounds on Context Engagement in RAG Systems
Marin J · 2025 · Preprint
WASTE classifies this as Failed Experiment Report · AI classification, approximate
An experimental approach did not work as intended — learn what to avoid before investing the same effort.
Abstract (excerpt)
When retrieval-augmented generation (RAG) systems hallucinate, what geometric trace does this leave in embedding space? We introduce the Semantic Grounding Index (SGI), defined as the ratio of angular distances from the response to the…
Excerpt shown for reference under fair use — read the full paper at the publisher.
Hosted by the publisher — may require access.
About to run something similar?
Run an AI Precheck on your own design to catch failure modes like this one before you spend the time. Your first desk check is free.
Related failures
The Two-pore channel (TPC) interactome unmasks isoform-specific roles for TPCs in endolysosomal morphology and cell pigmentation
Failed Experiment ReportA Markov-Middleton Model for Bursty Impulsive Noise: Modeling and Receiver Design
Failed Experiment ReportThe Reference of Proper Names: Testing Usage and Intuitions
Failed Experiment ReportAn investigation of self‐efficacy crossover between mentors and protégés within mentoring dyads
Failed Experiment ReportLymphatic endothelial cells are a replicative niche for Mycobacterium tuberculosis
Failed Experiment ReportStatistically comparing EEG / MEG waveforms through successive significant univariate tests: How bad can it be?
WASTE indexes this work — it does not host or republish it. Failure-type classification is automated and approximate.
Metadata source: Europe PMC · DOI 10.32388/1pq8it