The paper introduces Search-to-Record, a database‑assurance task where search‑enabled large language models reconstruct corporate‑event records from public sources, and DelistBench, a 1,200‑record benchmark for security‑level delisting announcements. Five models were evaluated in both closed‑book and web‑enabled settings, showing that web access improves announcement‑date accuracy by 34.0–48.0 percentage points and event‑status accuracy by 2.8–21.7 points, with the best system achieving 81.5% overall joint accuracy within seven days. The study also demonstrates that low‑cost web systems can approach the best system’s accuracy and that risk‑based triage can identify low‑error subsets, though a significant portion of cases still requires review.
arXiv:2602. 07294v4 Announce Type: replace-cross Abstract: With the increasing deployment of Large Language Models (LLMs) in the finance domain, LLMs are increasingly expected to parse complex regulatory disclosures.
The paper proposes an incremental pooled LLM evaluation method for selecting retrieval models in production Retrieval-Augmented Generation (RAG) systems. By having a language model judge the union of documents retrieved by current candidates and expanding the pool only with new documents from added systems, the approach reuses judgments across all systems. Experiments on four benchmarks and a financial news QA deployment show strong correlation with gold-standard rankings, high preservation of pairwise orderings, and significant cost savings—up to 4.9× lower evaluation cost and 65–80% judgment reuse.
arXiv:2609.16564v1 Announce Type: new Abstract: Retrieval-augmented generation (RAG) pipelines may omit a source's material relationship to the query. We study a pre-generation triage layer that trea...
arXiv:2608. 05212v1 Announce Type: new Abstract: Deep search agents tackle challenging questions through long-horizon web interactions, a process that is both complex and fragile: small reasoning errors may propagate through long, noisy trajectories into fluent but incorrect answers.
The paper evaluates how modern large language models use internal web search to answer factual questions. Using 783 static queries and 288 dynamic queries, the authors find that enabling retrieval improves accuracy on static questions but hurts confidence calibration. On dynamic queries, models often retrieve but still achieve less than 70% accuracy, mainly due to poor query formulation and source selection, indicating that internal web search works better as a quick verification tool than a full information‑retrieval system.
arXiv:2607. 18240v1 Announce Type: new Abstract: Large language models (LLMs) can achieve strong fact-checking accuracy, yet forced binary decisions conceal a critical reliability problem: systems may issue confident verdicts even when supporting evidence is weak, sparse, or internally inconsistent.
arXiv:2609.33676v2 Announce Type: replace Abstract: LLM agents increasingly take consequential actions through interactions with users, policies, and external tools. Auditing these agents requires au...
The paper reports a case study of a large language model (LLM) coding agent tasked with building a multi‑component data system from a detailed specification. During a single session the agent introduced five defects, which were categorized by violated constraints and detection methods. The study also evaluates the agent’s retrieval‑filtering strategy on the HotpotQA benchmark, showing that filtering to a graph‑identified entity set yields higher recall than unfiltered search, with a statistically significant gap across all tested budgets.
arXiv:2607. 19453v1 Announce Type: cross Abstract: We audit whether candle-based machine-learning models can turn predictions of cryptocurrency extrema or short-horizon outcomes into positive Binance Spot paper policies after assumed costs.
arXiv:2607. 04281v1 Announce Type: cross Abstract: Semantic caching reduces the latency and cost of retrieval-augmented generation (RAG) by serving cached answers to semantically similar queries, but most existing methods do not model the time-varying freshness of open-web evidence.
FinRCA-Bench is a deterministic synthetic benchmark comprising 2,250 accounts‑payable‑to‑bank reconciliation cases that span 14 operational tables and include 1,500 injected failures across 15 causal categories. The benchmark hides root‑cause labels and record‑level evidence contracts from models, enabling independent evaluation of evidence retrieval versus reasoning accuracy. Experiments show that retrieval architecture dramatically influences performance, with retrieval improvements raising macro‑required‑record recall from 0.83% to 77.70% and exact 16‑class accuracy from 2.05% to 72.44%.
WildSEEK is a new dataset of 3,000 real user information‑seeking queries, manually annotated for risk‑sensitive domains and whether the query is factoid or analytical. The accompanying evaluation framework tests LLM responses against four failure criteria—sycophantic behavior, overreliance, a default US‑centric perspective, and poor handling of vulnerable populations—finding higher failure rates for analytical queries. The authors also train classifiers on WildSEEK to analyze over 1.8 million realistic queries, revealing that more than a third are high‑risk and often analytical.