Extracting structured, machine-readable compliance criteria from regulatory documents remains an open challenge. Single-pass language models hallucinate structural elements, lose hierarchical relationships, and fail to resolve inter-document dependencies. We introduce RegReAct, a self-correcting multi-agent framework that decomposes regulatory information extraction into seven specialized stages, each with an Observe–Diagnose–Repair (ODR) loop that validates outputs against the source, correcting not only model hallucinations but also cross-reference errors in the regulations themselves. To ensure structural accuracy, RegReAct constructs a typed criterion graph; to ensure completeness, it resolves external dependencies by retrieving, summarizing, and embedding referenced legal content inline, producing self-contained outputs. Applying RegReAct to three EU Taxonomy Delegated Acts, we construct a dataset comprising 242 activities with over 4,800 hierarchical criteria, thresholds, and enriched source summaries. Evaluation against a GPT-4o single-pass baseline shows that RegReAct outperforms it across all structural and semantic metrics.

Existing benchmarks treat multi-turn conversation and reasoning-intensive retrieval separately, yet real-world information seeking requires both. To bridge this gap, we present a benchmark for reasoning-based conversational information retrieval comprising 707 conversations (2,971 turns) across eleven domains. To ensure quality, our Decomposition-and-Verification framework transforms complex queries into fact-grounded multi-turn dialogues through multi-level validation, where atomic facts are verified against sources and explicit retrieval reasoning is generated for each turn. Comprehensive evaluation reveals that combining conversation history with reasoning doubles retrieval performance (Baseline .236 → History+Reasoning .479 nDCG@10), while reasoning-specialized models substantially outperform dense encoders. Despite these gains, further analysis highlights that implicit reasoning remains challenging, particularly when logical connections are not explicitly stated in the text. [https://github.com/RECOR-Benchmark/RECOR]

Reasoning-intensive retrieval remains difficult for small models. Compact public ColBERTs are usually trained on general-purpose corpora and underperform reasoning-tuned 150M+ baselines on BRIGHT (CITATION) by several nDCG@10 points; no public reasoning-tuned ColBERT exists at edge scale. We introduce SmallReason-ColBERT, a 32M late-interaction retriever that closes much of this gap with three ingredients: a varied-length contrastive warmup on ReasonIR-VL, a hard-negative distillation polish on merged ReasonIR-HQ and BGE-Reasoner data, and a single-layer per-query-token importance head trained on top of the frozen base. The head is trained with an un-normalised weighted MaxSim score and evaluated with its length-normalised form. In a controlled re-training, replacing this training objective with the symmetric normalised score causes the loss to stall and costs 3.59 nDCG@10. The full recipe reaches 21.41 mean nDCG@10 on BRIGHT, within 1.21 of the 150M Reason-ModernColBERT (22.62) and above all ≤ 33M ColBERTs we evaluate. Through ablations on capacity, initialisation, and score variants, we further show that the learned head outperforms fixed IDF weighting and that simply thresholding the learned gates is harmful.

Multiple-choice questions (MCQs) are a standard format for evaluating large language models (LLMs), yet the popularity of answer options can confound evaluation. Modern LLMs systematically prefer popular but incorrect options over less popular correct ones, a vulnerability we call popularity bias. This pattern aligns with confidence miscalibration: model confidence remains high even as accuracy collapses for popular options. To systematically isolate this phenomenon, we introduce PopMCQ, a benchmark with six controlled strategies that vary option popularity while keeping the correct answer fixed. In our most adversarial setting, where all distractors are more popular than the correct option, models choose popular but wrong answers 66% of the time. To mitigate this bias, we propose PopDebias, a lightweight inference-time correction that estimates and removes a popularity prior from model predictions. It requires no fine-tuning, is label-free at test time (using only a small calibration split for parameter fitting), and adds negligible computational cost. Experiments on 22 open-source LLMs (0.5B to 32B parameters) show consistent improvements, with accuracy gains up to 54.1 percentage points under strong popularity pressure.

Late-interaction vision-language retrievers represent each document page as many visual token embeddings and score queries with MaxSim. In systems such as ColPali, ColQwen, ColNomic, and Nemotron ColEmbed, the document embeddings are produced without seeing the query, so the same page is represented identically for a table lookup, a chart question, and a layout-sensitive evidence request. We introduce Argus, a family of query-conditioned late-interaction retrievers built on Qwen3.5-VL. Argus adds a region-aware Mixture-of-Experts module: the query encoder produces both retrieval embeddings and a compact context vector, the document page is pooled into spatial regions, and a query-aware router selects latent experts per region before MaxSim. The output remains a multi-vector index compatible with ColPali-style retrieval, but the document representation is now dependent on the query (i.e., 𝐃(q)). All Argus models use a 1024-dimensional retrieval head, compared with the 2560-dimensional and 4096-dimensional heads of recent state-of-the-art systems, and are trained on roughly 9% of the available public supervision rather than the full pool. The 9B model reaches 92.67 NDCG@5 on ViDoRe V1 and 86.0 NDCG@5 on the combined V1+V2 leaderboard, the highest reported value for an open late-interaction model on the combined leaderboard. Wrapped in a Qwen3.6-27B agentic retrieval pipeline on ViDoRe V3, Argus-9B further improves its NDCG@10 from 60.28 to 64.80 over public tasks, showing that the same retriever serves both as a strong standalone system and as a search primitive for iterative LLM agents.

Existing temporal QA benchmarks focus on simple fact-seeking queries from news corpora, while reasoning-intensive retrieval benchmarks lack temporal grounding. However, real-world information needs often require reasoning about temporal evolution and synthesizing evidence across time periods. We introduce TEMPO, the first benchmark combining temporal reasoning with reasoning-intensive retrieval across 13 domains. TEMPO features: (1) 1,730 complex queries requiring deep temporal reasoning such as tracking changes, identifying trends, or comparing cross-period evidence; (2) step-wise retrieval planning with 3,976 decomposed steps and gold documents mapped to each step for multi-hop evaluation; and (3) novel temporal metrics including Temporal Coverage@k and Temporal Precision@k measuring whether results span required time periods. Evaluation of 16 retrieval systems reveals substantial challenges: the best model (DiVeR) achieves only 32.0 NDCG@10 and 71.4% Temporal Coverage@10, demonstrating difficulty in retrieving temporally complete evidence. We believe TEMPO provides a challenging benchmark for improving temporal reasoning in retrieval and RAG systems.

The growing demand for corporate sustainability transparency, particularly under new regulations like the EU Taxonomy, necessitates precise data extraction from large, unstructured corporate reports, a task for which Large Language Models and Retrieval-Augmented Generation (RAG) systems require high-quality, domain-specific question-answering datasets. To address this, we introduce SustainableQA, a novel dataset and a scalable pipeline that generates comprehensive QA pairs from corporate sustainability and annual reports by integrating semantic chunk classification, a hybrid span extraction pipeline, and a specialized table-to-paragraph transformation. To ensure high quality, the generation is followed by a novel automated assessment and refinement pipeline that systematically validates each QA pair for faithfulness and relevance, repairing or discarding low-quality entries. This results in a final, robust dataset of over 195,000 diverse factoid and non-factoid QA pairs, whose effectiveness is demonstrated by initial fine-tuning experiments where a compact 8B parameter model outperforms much larger state-of-the-art models. These results demonstrate the potential of SustainableQA as a resource for developing and benchmarking advanced knowledge assistants capable of navigating complex sustainability compliance data

Although Large Language Models (LLMs) show strong potential for zero-shot document ranking, current listwise approaches face three critical limitations. First, context length constraints prevent processing many documents simultaneously. Second, sequential generation creates bottlenecks that cannot run in parallel. Third, ranking results depend heavily on initial document order, leading to inconsistent performance. We introduce BracketRank, a reasoning-driven competitive elimination framework that addresses these challenges through systematic group competition. Our method uses adaptive grouping to automatically optimise group sizes based on LLM context limits, reasoning-enhanced prompts that require explicit relevance explanations, and bracket-style elimination where documents compete through winner and loser brackets. This structure ensures every document has fair advancement opportunities regardless of initial positioning while allowing for parallel processing across competition stages. We evaluate BracketRank on TREC DL 19, TREC DL 20, and eight BEIR benchmark datasets. Results show that BracketRank achieves 77.90 NDCG@5 on TREC DL 19 and 75.85 NDCG@5 on TREC DL 20, outperforming RankGPT and other state-of-the-art methods. On BEIR datasets, BracketRank reaches 54.66 average NDCG@10, demonstrating robust performance across diverse domains.

In this work, we present a systematic and comprehensive empirical evaluation of state-of-the-art reranking methods, encompassing large language model (LLM)-based, lightweight contextual, and zero-shot approaches, with respect to their performance in information retrieval tasks. We evaluate in total 22 methods, including 40 variants (depending on used LLM) across several established benchmarks, including TREC DL19, DL20, and BEIR, as well as a novel dataset designed to test queries unseen by pretrained models. Our primary goal is to determine, through controlled and fair comparisons, whether a performance disparity exists between LLM-based rerankers and their lightweight counterparts, particularly on novel queries, and to elucidate the underlying causes of any observed differences. To disentangle confounding factors, we analyse the effects of training data overlap, model architecture, and computational efficiency on reranking performance. Our findings indicate that while LLM-based rerankers demonstrate superior performance on familiar queries, their generalisation ability to novel queries varies, with lightweight models offering comparable efficiency. We further identify that the novelty of queries significantly impacts reranking effectiveness, highlighting limitations in existing approaches.