RAG-E: Quantifying Retriever-Generator Alignment and Failure Modes

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Randl, Korbinian, Rocchietti, Guido, Henriksson, Aron, Abedjan, Ziawasch, Lindgren, Tony, Pavlopoulos, John
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917232489005056
author Randl, Korbinian
Rocchietti, Guido
Henriksson, Aron
Abedjan, Ziawasch
Lindgren, Tony
Pavlopoulos, John
author_facet Randl, Korbinian
Rocchietti, Guido
Henriksson, Aron
Abedjan, Ziawasch
Lindgren, Tony
Pavlopoulos, John
contents Retrieval-Augmented Generation (RAG) systems combine dense retrievers and language models to ground LLM outputs in retrieved documents. However, the opacity of how these components interact creates challenges for deployment in high-stakes domains. We present RAG-E, an end-to-end explainability framework that quantifies retriever-generator alignment through mathematically grounded attribution methods. Our approach adapts Integrated Gradients for retriever analysis, introduces PMCSHAP, a Monte Carlo-stabilized Shapley Value approximation, for generator attribution, and introduces the Weighted Attribution-Relevance Gap (WARG) metric to measure how well a generator's document usage aligns with a retriever's ranking. Empirical analysis on TREC CAsT and FoodSafeSum reveals critical misalignments: for 47.4% to 66.7% of queries, generators ignore the retriever's top-ranked documents, while 48.1% to 65.9% rely on documents ranked as less relevant. These failure modes demonstrate that RAG output quality depends not solely on individual component performance but on their interplay, which can be audited via RAG-E.
format Preprint
id arxiv_https___arxiv_org_abs_2601_21803
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle RAG-E: Quantifying Retriever-Generator Alignment and Failure Modes
Randl, Korbinian
Rocchietti, Guido
Henriksson, Aron
Abedjan, Ziawasch
Lindgren, Tony
Pavlopoulos, John
Computation and Language
Retrieval-Augmented Generation (RAG) systems combine dense retrievers and language models to ground LLM outputs in retrieved documents. However, the opacity of how these components interact creates challenges for deployment in high-stakes domains. We present RAG-E, an end-to-end explainability framework that quantifies retriever-generator alignment through mathematically grounded attribution methods. Our approach adapts Integrated Gradients for retriever analysis, introduces PMCSHAP, a Monte Carlo-stabilized Shapley Value approximation, for generator attribution, and introduces the Weighted Attribution-Relevance Gap (WARG) metric to measure how well a generator's document usage aligns with a retriever's ranking. Empirical analysis on TREC CAsT and FoodSafeSum reveals critical misalignments: for 47.4% to 66.7% of queries, generators ignore the retriever's top-ranked documents, while 48.1% to 65.9% rely on documents ranked as less relevant. These failure modes demonstrate that RAG output quality depends not solely on individual component performance but on their interplay, which can be audited via RAG-E.
title RAG-E: Quantifying Retriever-Generator Alignment and Failure Modes
topic Computation and Language
url https://arxiv.org/abs/2601.21803