Insider Knowledge: How Much Can RAG Systems Gain from Evaluation Secrets?

RAG systems are increasingly evaluated and optimized using LLM judges, an approach that is rapidly becoming the dominant paradigm for system assessment. Nugget-based approaches in particular are now embedded not only in evaluation frameworks but also in the architectures of RAG systems themselves. While this integration can lead to genuine improvements, it also creates a risk of faulty measurements due to circularity. In this paper, we investigate this risk through comparative experiments with nugget-based RAG systems, including Ginger and Crucible, against strong baselines such as GPT-Researcher. By deliberately modifying Crucible to generate outputs optimized for an LLM judge, we show that near-perfect evaluation scores can be achieved when elements of the evaluation - such as prompt templates or gold nuggets - are leaked or can be predicted. Our results highlight the importance of blind evaluation settings and methodological diversity to guard against mistaking metric overfitting for genuine system progress.

Incorporating Q&A Nuggets into Retrieval-Augmented Generation

RAGE systems integrate ideas from automatic evaluation (E) into Retrieval-augmented Generation (RAG). As one such example, we present Crucible, a Nugget-Augmented Generation System that preserves explicit citation provenance by constructing a bank of Q&A nuggets from retrieved documents and uses them to guide extraction, selection, and report generation. Reasoning on nuggets avoids repeated information through clear and interpretable Q&A semantics - instead of opaque cluster abstractions - while maintaining citation provenance throughout the entire generation process. Evaluated on the TREC NeuCLIR 2024 collection, our Crucible system substantially outperforms Ginger, a recent nugget-based RAG system, in nugget recall, density, and citation grounding.

RoutIR: Fast Serving of Retrieval Pipelines for Retrieval-Augmented Generation

Retrieval models are key components of Retrieval-Augmented Generation (RAG) systems, which generate search queries, process the documents returned, and generate a response. RAG systems are often dynamic and may involve multiple rounds of retrieval. While many state-of-the-art retrieval methods are available through academic IR platforms, these platforms are typically designed for the Cranfield paradigm in which all queries are known up front and can be batch processed offline. This simplification accelerates research but leaves state-of-the-art retrieval models unable to support downstream applications that require online services, such as arbitrary dynamic RAG pipelines that involve looping, feedback, or even self-organizing agents. In this work, we introduce RoutIR, a Python package that provides a simple and efficient HTTP API that wraps arbitrary retrieval methods, including first stage retrieval, reranking, query expansion, and result fusion. By providing a minimal JSON configuration file specifying the retrieval models to serve, RoutIR can be used to construct and query retrieval pipelines on-the-fly using any permutation of available models (e.g., fusing the results of several first-stage retrieval methods followed by reranking). The API automatically performs asynchronous query batching and caches results by default. While many state-of-the-art retrieval methods are already supported by the package, RoutIR is also easily expandable by implementing the Engine abstract class. The package is open-sourced and publicly available on GitHub: http://github.com/hltcoe/routir.

FACTUM: Mechanistic Detection of Citation Hallucination in Long-Form RAG

Retrieval-Augmented Generation (RAG) models are critically undermined by citation hallucinations, a deceptive failure where a model confidently cites a source that fails to support its claim. Existing work often attributes hallucination to a simple over-reliance on the model's parametric knowledge. We challenge this view and introduce FACTUM (Framework for Attesting Citation Trustworthiness via Underlying Mechanisms), a framework of four mechanistic scores measuring the distinct contributions of a model's attention and FFN pathways, and the alignment between them. Our analysis reveals two consistent signatures of correct citation: a significantly stronger contribution from the model's parametric knowledge and greater use of the attention sink for information synthesis. Crucially, we find the signature of a correct citation is not static but evolves with model scale. For example, the signature of a correct citation for the Llama-3.2-3B model is marked by higher pathway alignment, whereas for the Llama-3.1-8B model, it is characterized by lower alignment, where pathways contribute more distinct, orthogonal information. By capturing this complex, evolving signature, FACTUM outperforms state-of-the-art baselines by up to 37.5% in AUC. Our findings reframe citation hallucination as a complex, scale-dependent interplay between internal mechanisms, paving the way for more nuanced and reliable RAG systems.

NeuCLIRBench: A Modern Evaluation Collection for Monolingual, Cross-Language, and Multilingual Information Retrieval

To measure advances in retrieval, test collections with relevance judgments that can faithfully distinguish systems are required. This paper presents NeuCLIRBench, an evaluation collection for cross-language and multilingual retrieval. The collection consists of documents written natively in Chinese, Persian, and Russian, as well as those same documents machine translated into English. The collection supports several retrieval scenarios including: monolingual retrieval in English, Chinese, Persian, or Russian; cross-language retrieval with English as the query language and one of the other three languages as the document language; and multilingual retrieval, again with English as the query language and relevant documents in all three languages. NeuCLIRBench combines the TREC NeuCLIR track topics of 2022, 2023, and 2024. The 250,128 judgments across approximately 150 queries for the monolingual and cross-language tasks and 100 queries for multilingual retrieval provide strong statistical discriminatory power to distinguish retrieval approaches. A fusion baseline of strong neural retrieval systems is included with the collection so that developers of reranking algorithms are no longer reliant on BM25 as their first-stage retriever. NeuCLIRBench is publicly available.

Overview of the TREC 2024 NeuCLIR Track

The principal goal of the TREC Neural Cross-Language Information Retrieval (NeuCLIR) track is to study the effect of neural approaches on cross-language information access. The track has created test collections containing Chinese, Persian, and Russian news stories and Chinese academic abstracts. NeuCLIR includes four task types: Cross-Language Information Retrieval (CLIR) from news, Multilingual Information Retrieval (MLIR) from news, Report Generation from news, and CLIR from technical documents. A total of 274 runs were submitted by five participating teams (and as baselines by the track coordinators) for eight tasks across these four task types. Task descriptions and the available results are presented.

MURR: Model Updating with Regularized Replay for Searching a Document Stream

The Internet produces a continuous stream of new documents and user-generated queries. These naturally change over time based on events in the world and the evolution of language. Neural retrieval models that were trained once on a fixed set of query-document pairs will quickly start misrepresenting newly-created content and queries, leading to less effective retrieval. Traditional statistical sparse retrieval can update collection statistics to reflect these changes in the use of language in documents and queries. In contrast, continued fine-tuning of the language model underlying neural retrieval approaches such as DPR and ColBERT creates incompatibility with previously-encoded documents. Re-encoding and re-indexing all previously-processed documents can be costly. In this work, we explore updating a neural dual encoder retrieval model without reprocessing past documents in the stream. We propose MURR, a model updating strategy with regularized replay, to ensure the model can still faithfully search existing documents without reprocessing, while continuing to update the model for the latest topics. In our simulated streaming environments, we show that fine-tuning models using MURR leads to more effective and more consistent retrieval results than other strategies as the stream of documents and queries progresses.

Distillation for Multilingual Information Retrieval

Recent work in cross-language information retrieval (CLIR), where queries and documents are in different languages, has shown the benefit of the Translate-Distill framework that trains a cross-language neural dual-encoder model using translation and distillation. However, Translate-Distill only supports a single document language. Multilingual information retrieval (MLIR), which ranks a multilingual document collection, is harder to train than CLIR because the model must assign comparable relevance scores to documents in different languages. This work extends Translate-Distill and propose Multilingual Translate-Distill (MTD) for MLIR. We show that ColBERT-X models trained with MTD outperform their counterparts trained ith Multilingual Translate-Train, which is the previous state-of-the-art training approach, by 5% to 25% in nDCG@20 and 15% to 45% in MAP. We also show that the model is robust to the way languages are mixed in training batches. Our implementation is available on GitHub.

PLAID SHIRTTT for Large-Scale Streaming Dense Retrieval

PLAID, an efficient implementation of the ColBERT late interaction bi-encoder using pretrained language models for ranking, consistently achieves state-of-the-art performance in monolingual, cross-language, and multilingual retrieval. PLAID differs from ColBERT by assigning terms to clusters and representing those terms as cluster centroids plus compressed residual vectors. While PLAID is effective in batch experiments, its performance degrades in streaming settings where documents arrive over time because representations of new tokens may be poorly modeled by the earlier tokens used to select cluster centroids. PLAID Streaming Hierarchical Indexing that Runs on Terabytes of Temporal Text (PLAID SHIRTTT) addresses this concern using multi-phase incremental indexing based on hierarchical sharding. Experiments on ClueWeb09 and the multilingual NeuCLIR collection demonstrate the effectiveness of this approach both for the largest collection indexed to date by the ColBERT architecture and in the multilingual setting, respectively.

Language Fairness in Multilingual Information Retrieval

Multilingual information retrieval (MLIR) considers the problem of ranking documents in several languages for a query expressed in a language that may differ from any of those languages. Recent work has observed that approaches such as combining ranked lists representing a single document language each or using multilingual pretrained language models demonstrate a preference for one language over others. This results in systematic unfair treatment of documents in different languages. This work proposes a language fairness metric to evaluate whether documents across different languages are fairly ranked through statistical equivalence testing using the Kruskal-Wallis test. In contrast to most prior work in group fairness, we do not consider any language to be an unprotected group. Thus our proposed measure, PEER (Probability of EqualExpected Rank), is the first fairness metric specifically designed to capture the language fairness of MLIR systems. We demonstrate the behavior of PEER on artificial ranked lists. We also evaluate real MLIR systems on two publicly available benchmarks and show that the PEER scores align with prior analytical findings on MLIR fairness. Our implementation is compatible with ir-measures and is available at http://github.com/hltcoe/peer_measure.