RAG Evaluation

Retrieval-augmented generation (RAG) systems often serialize user queries, retrieved documents, metadata, system labels, and task instructions into one natural-language prompt. We study a source-authority boundary failure in this design: attacker-authored retrieved text can impersonate metadata, provenance, authority, or disclosure-policy signals that appear control-relevant to the model. We call this pattern Document-Authored Control-Signal Impersonation (DACSI). DACSI is a non-imperative, metadata-like payload subclass within indirect prompt injection. Its central lesson is simple: document-authored labels are data, not policy. Command-style injection asks the model to ignore, override, or violate policy; DACSI asks whether untrusted document text can be misattributed as an authorized control signal when RAG prompt rendering collapses trusted and untrusted text into the same natural-language channel. We evaluate DACSI across six model settings, prompt-pressure levels, injection baselines, signal taxonomies, RAG-mediated pipelines, system-control probes, a source-authority attribution probe, and synthetic canary formats. We interpret the evidence by model regime rather than as six equal replications: DeepSeek V4 Pro and Qwen3.5-397B provide the cleanest positive lift, DeepSeek V4 Flash is a high-susceptibility setting, GPT-5.5 and Gemini 3.1 Pro Low are strong-boundary probes with selected residual risks, and GLM-4.7 is a saturated leakage boundary case. Across these regimes, DACSI warrants separate evaluation because it uses a command-free metadata/provenance/policy surface, follows a RAG-specific source-authority path, and responds to source/channel separation. The source-authority probe is behavioral attribution evidence, not proof of an internal mechanism.

Large language models answer knowledge-intensive questions using both parametric memory and retrieved evidence, but neither source is uniformly reliable. Retrieval can fill knowledge gaps, yet distracting passages may override correct closed-book answers. We study this post-generation conflict as answer-level source arbitration: given Direct and RAG answers from the same frozen model, decide which source to trust. We propose TRUSTMARGIN, a training-free, plug-and-play arbitration layer that scores the two existing candidates with the model's own likelihoods. It combines a parametric-prior margin, which tests whether memory accepts the retrieved answer, with an evidence-binding margin, which discounts passage-only salience and measures question-specific support. TRUSTMARGIN selects between Direct and RAG without fine-tuning, external judges, or additional generation. Across 2WIKIMQA and CWQA with three LLaMA scales, TRUSTMARGIN consistently improves over Direct generation and BM25-RAG, recovers part of the Direct/RAG oracle gap, and generalizes to multiple training-free RAG pipelines.

Retrieval-Augmented Generation (RAG) is widely used to improve the factual reliability of large language models (LLMs) by grounding answers in retrieved evidence. In misinformation-rich environments, however, retrieved content may include plausible but incorrect information, raising concerns about the reliability of RAG-based information access systems. In this work, we propose an evaluation protocol to systematically test how the RAG system handles conflicts between parametric knowledge and evidence retrieved from context with varying amounts of misleading information. We target correct answers to factoid questions that the model responds to correctly, even when there is no retrieval, and use this to test the system with clean, poisoned, and mixed evidence. The proposed analytical framework combines parametric override and confidence metrics to assess when and how misleading information affects the generation process of LLMs. This study aims to provide insights into the robustness of RAG systems in information disorder scenarios.

Multi-turn retrieval-augmented generation (RAG) is challenging due to evolving user intent, conversational noise, and strict context limits. We propose a training-free hybrid retrieval pipeline for SemEval-2026 Task 8 that combines dense and sparse retrieval with controlled query rewriting and cross-encoder reranking. On the official test set of Task A, our system achieves 0.5453 nDCG@5, ranking third among 38 teams and outperforming the strongest baseline score of 0.4795. For Task C, we reuse the documents retrieved for Task A and apply a lightweight generation pipeline guided by the official prompt, achieving 0.5312 as the harmonic mean of relevance and faithfulness and ranking 15th among 29 teams. All retrieval components are open-source, while query rewriting and answer generation rely on LLM APIs.

Final-answer accuracy, retrieval recall, and citation overlap do not reveal how much answer advantage a long-context or retrieval-augmented language model actually recovers from supplied evidence. A model may answer from parametric priors, fail to use evidence that is present, or cite relevant text without converting it into the final answer. This paper introduces a four-condition diagnostic protocol for evidence-utilization evaluation under matched examples, models, prompts, and scoring rules. The protocol compares no-evidence, full-context, retrieved-evidence, and oracle-evidence reference conditions, and uses Oracle-Reference Normalized Context Utilization (ONCU) as a denominator-valid estimate of recovered oracle-reference evidence advantage. The empirical study evaluates five local open-weight models from the Qwen, Gemma, Llama, and Mistral families over Controlled-ONCU-safe16K, HotpotQA-ONCU, and 2WikiMultiHopQA-ONCU, comprising 18,000 ONCU-compatible predictions. Results show a task-dependent diagnostic pattern: controlled synthetic settings expose reduced recovery when the same evidence is embedded in long input rather than supplied compactly, while realistic multi-hop reconstructions show that full-context inputs outperform the tested retrieved inputs in denominator-free answer and evidence metrics, with ONCU supporting the same direction on oracle-improving groups. Sensitivity audits with stronger retrieval settings narrow some gaps but do not overturn the scoped interpretation. The main contribution is therefore not a single utilization ratio, but a matched diagnostic protocol that separates no-evidence answerability, oracle-evidence recoverability, full-context recovery, retrieval-conditioned recovery, denominator validity, and companion answer/evidence diagnostics.

Retrieval-Augmented Generation (RAG) enhances Large Language Models (LLMs) by grounding their responses in external knowledge, but conventional pipelines rely on static, single-step retrieval that limits performance on complex queries. This paper presents an Agent-Orchestrated Adaptive RAG framework that introduces dynamic query decomposition, iterative retrieval, and a bounded self-reflective evaluation loop. We evaluate the system across two complementary datasets: a domain-specific DevOps knowledge base and the multi-hop reasoning benchmark MuSiQue. Using metrics that include overall score, citation accuracy, mean reciprocal rank, and topic coverage, we find that query decomposition yields consistent gains in the structured domain (overall score $+0.04$, MRR $+0.17$ on DevOps) but degrades ranking precision on the multi-hop benchmark, while the reflection mechanism improves citation accuracy at a substantial latency cost. These contrasting results show that agentic enhancements are not universally beneficial and must be applied selectively according to query and domain characteristics. Our findings argue for adaptive, cost-aware orchestration rather than uniformly aggressive reasoning pipelines.

Retrieval-augmented QA pipelines often route retrieved passages through an LLM \emph{rewriter} before a smaller reader, lifting F1 by tens of points on multi-hop benchmarks; this gain is typically credited to improved evidence quality. We ask whether that lift is causally driven by the gold answer string appearing in the rewritten context rather than by curation per se, using a controlled intervention audit. For each rewritten context we re-run the reader after one of four controlled edits to the compile output: removing the gold answer span, replacing a length-matched random non-answer span (placebo), or injecting the gold into rewrites where it was absent (at the prefix or at a midpoint sentence boundary). Across twelve completed (cell, baseline) intervention runs spanning three reader families (Qwen2.5-7B, Qwen3.5-35B, GLM-4.7), two datasets (HotpotQA, 2WikiMultihopQA), and three compiler arrangements (MA-only, MB-only, MA$+$verify), removing the gold answer drops reader F1 by $28$ to $64$ points beyond the length-matched placebo on paired \texttt{answer-in-compile} strata, and prepending the gold into rewrites that lacked it raises F1 by $+0.7$ to $+9.7$ points in $10$ of $12$ (cell, baseline) combinations. A companion five-sentinel audit shows the conventional single-\texttt{[MASK]} probe is itself sentinel-fragile: on 2Wiki it reports a $+4.12$~F1 ``non-leakage residual'' that flips to $-3.33$ to $-7.81$~F1 under four alternative sentinels and fails an equivalence test for three of those four ($1/4$~pass). We do not propose a new rewriter or mitigation; we release the intervention runner and the sentinel panel so that other rewriter-gain claims can be tested against the same standard.

Summarizing the latest medical literature to guide clinical decision-making is essential for evidence-based medicine and high-quality patient care. Yet clinicians face increasing challenges due to limited time with patients and a rapidly growing volume of published articles. Although retrieval-augmented large language models (LLMs) have shown promise in clinical summarization, human evaluations of their effectiveness in synthesizing broader scientific literature and direct comparisons to expert-written syntheses remain scarce. We constructed a RAG-based agentic AI framework using three state-of-the-art LLMs: Sonnet, GPT-4o, and Llama 3.1. A headache specialist created 13 questions, three for prompt optimization and ten for evaluation. Ten headache specialists across the United States and Canada each wrote a summary for one question, yielding four summaries per question (expert, Sonnet, GPT-4o, and Llama). The experts, blinded to authorship, critically evaluated the summaries, excluding the topic for which they wrote a summary, based on correctness, completeness, conciseness, and clinical utility, scoring each from 1 to 10 using standardized rubrics. They also ranked the summaries by preference and indicated whether they believed each summary was written by an expert or an LLM. Our study, comparing LLM- and expert-written literature summaries evaluated by headache specialists, showed that expert-written summaries were preferred, although experts sometimes found it challenging to distinguish between human- and AI-generated summaries. We also identified key expert-valued features beyond standard evaluation metrics that can guide future refinement of both human and AI literature summarization pipelines.

Many real-world questions over business, legal, and scientific corpora are natural-language versions of database-style queries over records latent in text. Existing retrieval-augmented generation (RAG) systems are optimized primarily for semantic relevance, but retrieving plausible passages does not guarantee correct query execution. We introduce QO-Bench, a diagnostic benchmark for query-operator question answering over typed event tuples. The benchmark covers 22,984 news articles and 614 corporate events, with 18 query templates instantiating 785 questions. Each gold answer is deterministically computed from typed event tuples and scored by recall, with answers matched to the gold tuples by exact match rather than an LLM judge. This design enables operator-level diagnosis such as joins and intersection. We evaluate RAG, ReAct RAG, GraphRAG, and information-extraction-to-SQL under matched conditions, with a long-context oracle ceiling to isolate retrieval failure. A two-axis framework -- index-time preservation versus query-time execution -- predicts where each paradigm fails, and the results bear it out: systems retrieve relevant text but discard the typed values operators need, and the deployable paradigm ranking inverts across operators, with similarity retrieval leading on filter/project and extraction-to-SQL on intersection and counting. Even given the gold evidence, a long-context oracle stays far from saturated, so operator execution -- not retrieval alone -- is a core bottleneck that a stronger answer model does not remove. QO-Bench reframes the goal from passage relevance to query-operator-preserving retrieval. The benchmark, predictions, and code are released at https://github.com/ZHANG-MENGAO/qo-bench.

AI systems increasingly need to combine two demanding capabilities: navigating multi-session conversation history and performing deep reading comprehension within long documents. Yet no existing benchmark evaluates both simultaneously. We introduce MemoryDocDataSet, a synthetic benchmark of 50 micro-worlds and 1,000 QA pairs in which each instance comprises 3-5 personas, a temporal event graph spanning months of activity, 3-5 real long documents (20,000-50,000 tokens each sourced from the Caselaw Access Project), multi-session conversations grounded on those documents, and 20 question-answer pairs across five reasoning categories. The defining feature is the Hybrid source tag: questions requiring a system to first navigate conversation history to identify which document is relevant, then extract the answer from within that document. Hybrid questions account for 75.1% of the dataset. Dataset quality is characterised through a prompt-sensitivity self-consistency analysis using LLM-as-judge, yielding a median Cohen's $κ= 0.634$ across all 50 micro-worlds. We evaluate six baseline configurations spanning truncated context, long-context LLMs, retrieval-augmented generation (RAG), and memory systems. The best baseline (RAG-Both) achieves 0.358 overall F1 and 0.342 on Hybrid. Document-only retrieval (RAG-Doc) collapses to 0.267 on Hybrid despite achieving 0.453 on Doc-only questions, demonstrating a clear joint-retrieval gap that motivates architectures unifying conversational memory with long-document navigation. We release the dataset, generation pipeline, and all baseline implementations.

Recent advances in multimodal retrieval-augmented generation (MM-RAG) have shifted toward minimal parsing, relying on page-level images for producing retriever embeddings and for answer generation. While efficient, this trend often neglects explicit handling of the rich, structured information in complex enterprise documents, instead depending on pre-trained embeddings or vision-language models to implicitly capture such structure. In this work, we take a more direct approach: MM-BizRAG proactively extracts and represents document structure via a document structure-aware split that dynamically routes documents through orientation-specific ingestion pipelines, applying explicit layout-aware parsing for vertically structured documents (e.g., reports) and holistic page-level representations for horizontally structured documents (e.g., slide decks). A unified LLM-driven artifact transformation pipeline with placeholder-based positional alignment preserves natural reading order, while inference-time multimodal assembly decouples retrieval representations from generation context, enabling richer, more grounded answers without any finetuning requirement. Through experiments on a large, heterogeneous enterprise dataset and two public benchmarks (SlideVQA and FinRAGBench-V), MM-BizRAG consistently outperforms state-of-the-art vision-centric baselines by up to 32% points, with especially strong gains on report-style layouts. Furthermore, we introduce FastRAGEval, a single-call LLM Judge metric for fine-grained generative recall that halves RAGChecker's cost while achieving stronger human alignment.

Medical question answering is a high-stakes setting where factual errors can have serious consequences. Retrieval-augmented generation (RAG) is widely viewed as a promising solution, and prior work has reported substantial gains for large medical QA models. We revisit this assumption across a broad range of open-weight instruction-tuned models spanning 7B to 72B parameters. Across five models, ten biomedical QA datasets, four retrieval methods, and four retrieval corpora, we find that retrieval yields only small and inconsistent improvements over a no-retrieval baseline, typically within 1-2 points. In contrast, the choice of backbone model has a much larger effect than the choice of retriever or corpus, and expert and layman retrieval sources perform similarly in most settings. These results suggest that the main bottleneck is not retrieval quality alone, but the model's limited ability to use retrieved evidence effectively.

Retrieval-augmented generation systems for legal question answering typically retrieve passages based on semantic similarity and provide them to a language model, which then generates cited answers. Prior work assumes that highly ranked passages are most likely to be usefully cited by the model. Perturbation-based attribution methods, such as C-LIME, have been used exclusively for post-hoc explanation. However, on the AQuAECHR benchmark, semantic similarity does not correlate with passage attribution. Within a retriever's candidate pool, similarity-based ranking performs worse than random selection at surfacing gold citation paragraphs. To address this limitation, a lightweight cross-encoder is trained on continuous perturbation-based attribution scores to re-rank passages prior to generation. This approach is evaluated on the AQuAECHR benchmark, using two language models and five-fold cross-validation. The re-ranker substantially improves citation faithfulness and alignment with gold expert answers. Notably, two re-rankers trained independently on different models converge beyond their raw attribution agreement. This finding indicates that the cross-encoder reduces model-specific noise and produces a shared relevance signal that partially transfers across models, although same-model re-ranking remains more effective. These results demonstrate that perturbation-based attribution provides a practical, model-agnostic training signal for citation-aware retrieval.

Chunked-document retrieval is a common component of retrieval-augmented generation (RAG) systems. Documents are split into overlapping chunks, embedded, and indexed with approximate nearest-neighbor search such as hierarchical navigable small world graphs (HNSW). Overlap improves boundary coverage but induces a practical failure mode: top-k retrieval often returns near-adjacent chunks that repeat evidence and waste prompt budget. We propose Self-Conditioned Positional HNSW (SCP-HNSW), a lightweight modification that appends a low-dimensional positional code to chunk embeddings and uses a two-pass query procedure to estimate and apply a query-specific document-position prior. SCP-HNSW leaves HNSW graph construction and traversal unchanged while adding an auditable minimum-index-gap selector for final context construction. We also integrate industrial review artifacts for generated evidence quality: a 770-review text-evidence audit with 318 fully labeled reviews and a 70-case OCR audit with 350 ratings. The text audit shows that 574 of 770 projected reviews are rated 3/5, only 39 fall in the 1-2 range, and narrative reviewer detail appears much more often than structured issue flags. The OCR audit shows slice-level pass rates from 95% for clean chat screenshots to 45% for handwritten/blurry captures, with moderate to strong agreement. These results motivate overlap-aware, audit-friendly RAG retrieval and identify the remaining controlled retrieval ablations needed for causal performance claims.

Retrieval-Augmented Generation (RAG) has demonstrated significant capabilities in enhancing the performance of Large Language Models (LLMs). One of the key tasks in RAG systems is the chunking process. Traditionally, fixed-size chunking and semantic chunking have been the standard approaches. However, interest in chunking strategies has been increasing, leading to a growing number of proposed methods that often claim improved performance over these conventional techniques. Many of these approaches are tailored to specific use cases and data types, with limited evidence of their effectiveness across diverse scenarios. As a result, it remains challenging to directly compare different techniques and assess their relative strengths. To the best of our knowledge, this study is the first to systematically evaluate the effectiveness of a wide range of chunking methods and emphasize the underlying challenges of chunking strategies in RAG systems. While chunking is commonly treated as a simple preprocessing step, we show that it introduces a range of impactful and often overlooked issues.

Retrieval-Augmented Generation (RAG) systems are often evaluated using final answer accuracy, even though their failures can originate from preprocessing, retrieval, context packing, or generation. This paper presents a controlled empirical study of RAG sensitivity, robustness, and stability across 56 experimental runs. We evaluate how chunk size, retrieval depth (top k), embedding-based reranking, probabilistic retrieval noise, and repeated seeded runs affect retrieval, context packing, and generation behavior. Using a fixed 500-question QA subset mapped to 20,958 unique corpus contexts, we analyze both final answer metrics and intermediate failure modes. Across these experiments, retrieval-oriented metrics improved under broader retrieval settings, while downstream exact-match and F1 scores often behaved non-monotonically. We also observe preprocessing-induced answer loss under smaller chunk sizes, progressive degradation under retrieval corruption, and higher observed variance in broader retrieval regimes. These findings suggest that RAG evaluation should include sensitivity, robustness, stability, and multi-stage failure analysis rather than relying only on final answer accuracy.

Retrieval-Augmented Generation (RAG) is the current industry standard for grounding AI in real-world facts. Traditional retrieval methods rely on keyword matching and topic proximity, ranking content based on how closely it sounds like the user's query. What they do not measure is how many verified facts the content actually contains. This structural gap, termed the Expert Blindness Effect, causes standard RAG pipelines to consistently bury high-density factual evidence in favor of lexically dominant text on the same topic. To address this gap, this paper introduces Factual Density (FD*), a novel retrieval optimization signal that measures the proportion of verified atomic claims relative to total token count. Using the NexusAgentics Ghost Audit preprocessing pipeline, raw text is scored for factual specificity using probabilistic factuality analysis to filter content before corpus ingestion. An initial formulation introduced a severe document-length confound (Pearson R = -0.8636, p = 2.27e-07). Implementing Z-score normalization within length bins resolved this bias, validating FD* as a length-independent density signal (p = 0.0749). Evaluated against the HealthFC benchmark (750 health claims labeled Supported, Refuted, or No Evidence by medical experts), FD*-optimized retrieval was the only condition to achieve 100% systematic review saturation in top-5 results, surfacing Cochrane evidence that standard cosine similarity ranked outside the top ten. Ground truth verification confirmed 25 mappings across seven HealthFC-supported claims. While full statistical validation across n=50 queries remains future work due to constraints on corpus-benchmark alignment, these findings establish factual density reranking as a low-cost, high-impact intervention for improving factual precision in health RAG architectures.

In existing memory benchmarks for Large Language Models (LLMs), the evaluated dialogue sessions often lack long-term semantic consistency, and the underlying personas tend to be flat and static. Furthermore, in real-world scenarios, interactions between users and assistants involve more diverse, heterogeneous data streams, such as documents and emails. These shortcomings significantly limit the realism and effectiveness of current evaluations. To address these limitations, we introduce RHELM (Realistic, Heterogeneous, and Evolving Long-term Memory). Driven by meticulously crafted user profiles and a novel LOOP (pLan-rOllout-evOlve-Prune) module, we construct realistic dialogues across diverse interaction scenarios that exhibit dynamic temporal evolution and long-term coherence. Crucially, these dialogues are deeply integrated with heterogeneous external sources synchronized with the user's temporal event trajectory. The resulting benchmark encompasses challenging question-answer pairs spanning seven inquiry types, with each question mapping to at least one of 27 critical memory characteristics that we identify as essential yet underexplored in current research. Comprehensive experiments across full-context models, retrieval-augmented generation (RAG) methods, and representative memory frameworks reveal that contemporary approaches still expose critical weaknesses in complex, real-world settings, particularly in resolving multi-source aggregation and real-world contextual reasoning.

Retrieval-Augmented Generation (RAG) supplements a language model's input with retrieved documents, yet most RAG pipelines inherit retrieval components designed for human readers. How retrieved content should be represented when the consumer is a large language model (LLM) rather than a human is less well understood. Recent work has proposed transformations of retrieved content and identified properties that affect generation, but each examines a single transformation or property in isolation, leaving open which features of a document's representation matter most. We address this with a controlled comparison: holding retrieval fixed, we vary only the representation of retrieved documents, comparing an original baseline against thirteen transformations spanning selection, summarisation, and reformulation, in query-dependent and query-independent variants. Across these fourteen representations we measure question-answering accuracy for four generators, and for each representation we also measure answer retention: whether a known answer-bearing document still supports its answer after transformation. We find that answer retention is the primary determinant of generator accuracy; notably, when retention is high, a representation's wording, structure, length, and query-dependence have limited effect. This suggests that accuracy gains attributed to specific mechanisms in prior work may be partly explained by how well those mechanisms preserve answer-bearing content, an attribution that cannot be settled without controlling for retention.

RAG-based legal assistants have been growing in popularity, but LLM hallucinations remain a key issue and potentially undermines justice. While benchmarks have been developed to evaluate progress, many rely on synthetic queries rather than realistic legal scenarios. Moreover, Canadian law remains underrepresented in existing evaluations. To address this gap, we introduce CanLegalRAGBench, a Canadian legal QA benchmark based on realistic queries and expert-annotated answers grounded in case law. Our evaluation shows that retrieval performance is sensitive to design choices and that open-source embedding models are competitive with closed source models. However, it also reveals the limitation of automatic evaluations that penalize systems for retrieving alternative relevant documents. We also find that generated answers often diverge from gold responses, either with hallucinations or by producing overly detailed or irrelevant content, with 8-29% of claims not being supported by the retrieved documents. We hope this benchmark will help drive continued progress in addressing limitations of legal RAG systems.

Creating spoken dialogue datasets is methodologically challenging, and these challenges are amplified when the goal is to build multilingual, multi-parallel datasets at scale. This work introduces HEALTHDIAL, a large-scale, multilingual, and multi-parallel dataset for developing and evaluating retrieval-augmented generation (RAG)-based spoken dialogue systems. The dataset comprises 6,000 information-seeking dialogues (1,500 per language) grounded in trusted content from the World Health Organization (WHO) and 163 hours of user speech recorded from native speakers of diverse dialects across four official WHO languages: Arabic, Chinese, English, and Spanish. Each speaker is annotated with demographic (e.g., gender, age) and sociolinguistic (e.g., primary language, region of origin) variables. We report benchmark results across key dialogue tasks, which reveal consistent performance disparities across languages, even among high-resource ones. To support future research, we release the dataset, a prototype system, and a toolkit for data collection and system evaluation.

Retrieval-augmented generation (RAG) systems expose numerous design choices spanning query rewriting, chunking, retrieval depth, reranking, and context compression. In practice, these choices are often configured through heuristics, hindering systematic evaluation and reproducibility across settings. We argue that this challenge is best formulated as RAG architecture search. To support controlled and reproducible study of this problem, we introduce the RAG Intelligence Search Engine (RAISE), a comprehensive framework and benchmark for RAG hyperparameter optimization, which evaluates optimization methods for RAG pipelines under standardized search spaces and budgets. RAISE implements 13 search algorithms and evaluates them across seven public text and multimodal datasets using three random seeds. Our experiments show that optimization performance is highly task-dependent: methods that perform strongly on one dataset may not generalize consistently across others, cautioning against interpreting aggregate rankings as evidence of universally superior strategies. RAISE provides a common experimental substrate for fair, reproducible, and systematic research on RAG hyperparameter optimization.

Large Language Models (LLMs) have advanced financial automation through Retrieval-Augmented Generation (RAG), yet hallucinations remain a critical barrier to deployment in high-stakes environments. Existing benchmarks focus on single-turn, English-centric tasks, leaving the multi-turn dynamics and linguistic-regulatory nuances of the Korean financial domain unaddressed. We introduce K-FinHallu, the first benchmark for hallucination detection in multi-turn Korean financial RAG. We construct multi-turn dialogues from authentic Korean financial documents and inject hallucinations under a proposed hierarchical taxonomy based on context answerability that explicitly accounts for justified abstention. Benchmarking frontier and open-source LLMs as hallucination detectors, we find that even the strongest models struggle with fine-grained financial diagnostics and refusal behavior. While fine-tuning an 8B model on our training split yields performance competitive with frontier LLMs, justified abstention remains the weakest axis across all evaluated models.

A retrieval-augmented generation (RAG) system deployed over a multi-author institutional corpus can give a different answer to the same question depending on which source it retrieves -- a failure mode the dominant single-gold-answer paradigm cannot diagnose. We argue that source-dependence is a missing axis of NLP evaluation, and that auditing it means shifting the unit of evaluation from answer correctness to the inter-source relationship. We make this concrete in transplant patient education, where institutional sources demonstrably disagree, releasing three artefacts: TransplantQA, a benchmark of real patient questions, each answered by grounding generation in multiple institutional handbooks as candidate sources; HERO-QA, a hierarchical retrieval strategy that grounds and audits each answer; and a structured-output judge that scores inter-source relationships on a validated 5-label taxonomy. At scale, better retrieval reveals far more disagreement than prior estimates suggested -- understating its prevalence, not its intensity. The framework is domain-agnostic and transfers to legal and educational RAG: measuring source-dependence is a responsibility for deployed multi-source NLP generally.

Users of search-augmented LLMs rely on citations as evidence that responses are grounded in real sources, and rarely verify the cited pages themselves. Millions of queries per day now pass through these systems, making citation quality a silent determinant of whether users are informed or misled-yet existing benchmarks each address one facet in isolation, leaving the joint structure that determines citation trustworthiness unmeasured. We construct CITETRACE, a large-scale dataset that traces the full citation chain from user query through retrieved source to generated answer: 11,200 real-world queries from 28 communities paired with 112,000 responses from ten models across five providers, yielding 761,495 evaluable citation pairs. We design a three-dimension evaluation framework that scores each citation on intent-purpose alignment, source suitability, and answer-source fidelity, using expert-validated predefined matrices and a five-level fidelity rubric; the framework applies to any system that produces citation-bearing responses. Applying this framework at scale, we identify a systematic pattern we call VERIFIED MISGUIDANCE (VM): models cite real, accessible sources yet fail along one or more dimensions, producing a fidelity-suitability trade-off in which faithful models select inappropriate sources and vice versa. Across our pool, 30.6% of citations distort their sources and 27.1% originate from domain-inappropriate sources; at the response level, up to 96% of users encounter at least one structurally misleading citation. Provider-level differences explain 88-96% of citation-quality variance, suggesting that source selection is governed more by factors beyond individual model capability than by the LLMs themselves. Together, CITETRACE and its evaluation framework provide the first resource for diagnosing structural citation failures in deployed search-augmented systems.

We study quality-latency-resource trade-offs in a documentation-grounded retrieval-augmented generation (RAG) system that uses Low-Rank Adaptation (LoRA) of the generator. We build a manually verified benchmark of 5,144 question-answer pairs over the official Kubernetes documentation and combine it with a fixed hybrid-retrieval pipeline (BGE-M3 dense, BGE-M3 native sparse, Reciprocal Rank Fusion, cross-encoder reranking). Over this benchmark we ablate 20 LoRA configurations on Llama-3.2-3B-Instruct and Llama-3.1-8B-Instruct across rank and target-module choices, and evaluate each on token-level F1, LLM-judged groundedness and correctness (pass@4), inference latency, inference memory, and training cost, all reported with bootstrap 95% confidence intervals. Pareto analysis shows that LoRA adapters acting only on the q and v attention projections consistently dominate the front, while the 3B/8B choice mainly defines operating regime. A param-matched control comparison further indicates that the q/v advantage is structural rather than purely parametric. The benchmark, selected adapters, and code are available at https://github.com/EugPal/rag-lora-tradeoffs.

Cited RAG evaluation often treats visible sources as a grounding signal, but a real, topically relevant citation can still under-warrant the attached wording. We study this diagnostic failure as citation laundering: a related source is presented as warrant for an over-strong claim. We introduce FORCEBENCH, a contrastive stress test for evidence-force calibration. Each item holds a cited passage fixed and pairs an evidence-calibrated claim with a localized force-raised variant across five operational axes: relation, modality, scope, temporal validity, and numeric specificity. A calibrated evaluator should score the evidence-calibrated claim higher. Headline experiments use a fixed, locality-filtered 198-pair evaluation set. A citation-presence sanity check is uninformative by design; token and entity overlap still violate monotonicity on 32.8--36.4% of pairs. Across four reported model judges, standard generic support prompting is insufficient for this force-calibration stress test (aggregate MVR 47.2%), while explicit warrant-strength prompting lowers MVR to 24.5% but remains imperfect. We release the benchmark, prompts, outputs, and plug-in pipeline so citation evaluators can report monotonicity violation rate and force sensitivity alongside conventional support metrics.

Retrieval-augmented generation (RAG) systems are often compared by asking a large language model (LLM) judge which answer is better. For multi-hop RAG, this has become a measurement problem as much as a modeling problem: the same score can reflect retrieval quality, answer length, lexical overlap, or a statistical test that ignores clustered data. We ask what happens when these choices are made explicit. We propose a minimum measurement standard for LLM-as-a-judge comparisons in RAG. The standard fixes the top-100 candidate pool, evidence budget, answer cap, generator, and prompt; it also requires pre-registered hypotheses, cluster-aware inference, an exact cluster sign-flip check when feasible, and second-judge replication. Clustered benchmarks can overstate progress; the field should adopt this standard. We stress-test it with Genetic Algorithm Decoder for Multi-hop Evidence Composition (GADMEC), an evolutionary evidence selector, on 400 multi-hop questions in computer science/machine learning (CS/ML) and Materials Science. The protocol changes the empirical story. A binomial test makes all four semantic-baseline comparisons look significant; cluster-aware inference leaves only one Bonferroni-significant result. BM25 beats pure semantic GADMEC under the same budget, while a lexical-semantic hybrid recovers in CS/ML and narrows the Materials Science gap.

Retrieval-augmented generation (RAG) systems can respond incorrectly even when the correct passage was retrieved. The model must still read the retrieved passages and identify which one contains the answer among others that look relevant. This passage-reading model is called the reader. Does it fail simply because the context is longer or because the other passages genuinely compete with the correct one? We introduce and demonstrate a matched-control protocol for RAG reading: we keep the number and length of passages fixed, but replace hard competitors with less competitive real passages. We apply this control across two compact open models on SQuAD. This replacement partially restores performance, with the strongest effects on F1 and answer inclusion. For Phi-2, this recovers +6.0 EM points, +7.0 answer-inclusion points, and +0.057 F1. For Qwen2.5-1.5B, it recovers +4.5 EM points, +9.0 answer-inclusion points, and +0.068 F1. To track how performance changes as competitors accumulate, we also report retention curves and summarize them with a right-censored half-life when the curves do not cross half-retention. Together, these results show the protocol isolates a competition effect distinct from context length, though the effect is clearer for F1 and answer inclusion than for exact match, and also varies with snippet length.

Modern retrieval-augmented generation(RAG) deployments increasingly rely on caching to reduce token cost and time-to-first-token(TTFT). Prefix-level KV reuse is now standard in serving stacks such as vLLM, and chunk-level and position-independent reuse have been pushed further by recent systems(RAGCache, TurboRAG, CacheBlend, EPIC, ContextPilot, PCR, LMCache). Output-level semantic answer caches, by contrast, remain fragile: similar prompts can map to different correct answers, retrieved evidence drifts as the corpus is updated, and adversarial collision attacks have been shown to hijack cached responses. We argue that the right framing for cached answer reuse is not how to reuse faster but when reuse is safe. We propose GroundedCache, an evidence-validated cache router that admits a cached answer only when 4 cheap gates simultaneously hold: query similarity, retrieved-evidence overlap, source-version validity, and lexical (or judge-based) support of the cached answer by the freshly retrieved evidence. We build a six-regime workload that stress-tests cache safety rather than only hit rate, and introduce an operator-facing metric, the unsafe-served rate (USR), fraction of all queries that received a wrong cached answer. Across 2 datasets and 12,000 real-LLM generations(Qwen2.5-7B-Instruct on vLLM with Automatic Prefix Caching), GroundedCache drives USR to 0.0% on every HotpotQA regime(vs. 15-35% under naive caching) and to 1.5% on mtRAG document drift(vs. 51.5%), a 34x reduction on the design-point adversarial regime and 3-10x reductions across the other mtRAG regimes, while end-to-end p50 latency stays within 1.04-1.07x of a no-cache RAG baseline. A per-gate ablation isolates the lexical support gate as the load-bearing safety mechanism on both datasets, with the remaining gates providing defense-in-depth at near-zero cost. We release the implementation, workload, and evaluation harness.