knowledge management teams

Abstract: Agents powered by large language models (LLMs) have demonstrated remarkable progress in solving complex reasoning tasks. However, LLM agents often falter on long-horizon tasks due to cognitive overload, as their working memory becomes cluttered with expanding and irrelevant information, which dilutes their attention and hinders effective planning and reasoning. To mitigate this challenge, we introduce Cognitive Resource Self-Allocation (CORAL), a novel reasoning paradigm that empowers agents to proactively optimize their context. Implemented as an agent-callable working memory management toolset, CORAL allows an agent to checkpoint its task progress and verified facts, and adaptively initiate a new problem-solving episode by resetting cluttered context while preserving only the checkpointed information, enabling it to resume reasoning from a clean state. We further enhance the agent's checkpointing capabilities using a Multi-episode Agentic Reinforced Policy Optimization algorithm. On the GAIA benchmark, CORAL with an 8B model achieves 42.8% average accuracy, with particularly strong gains on the most challenging long-horizon tasks (Level 2 and Level 3). Crucially, these gains come at a substantially lower inference cost: CORAL consumes 6-7$\times$ fewer input tokens than a ReAct agent. It also generalizes to multi-hop Knowledge QA benchmarks and to the considerably harder BrowseComp, where it improves a strong proprietary agent by 8.0 points.

Abstract: Recent systems applying retrieval-augmented generation (RAG) and "deep research" iteratively generate queries, retrieve documents, and examine their contents, using the resulting information to guide subsequent search steps. When this process is carried out autonomously rather than by a human searcher, the system must make explicit decisions about which search trajectories to explore and what objective the search is intended to optimize. We introduce a simple framework for reasoning about such search policies, providing a principled approach to evaluation. The framework models the search environment as a set of ranked lists and represents behavior using two primitive actions: select, which advances judgment through an existing ranking, and spawn, which generates a new ranking through query generation or reformulation. Each action produces a document to inspect, yields information that updates the search state, and contributes a reward for a task-specific objective. Within this abstraction, a variety of existing retrieval strategies can be expressed as policies operating over a common environment, while common retrieval objectives can be represented through reward and weighting functions over the resulting search trajectory, and the resulting formulation admits a natural reinforcement learning interpretation. We illustrate the approach using high-recall information retrieval (HRIR), where the goal is to identify as many relevant documents as possible with limited judging effort, showing how the framework provides a unified way to describe and reason about strategies for both classical retrieval tasks and emerging LLM-guided search systems. Through experiments, we demonstrate that both the underlying retrieval method and the "deep research" policy employed can materially affect performance.

Abstract: We introduce idea traces, a novel thread-based representation of accretive content across scientific papers. Rather than capturing fixed themes from papers, idea traces surface similarities and deltas of faceted content that enable inspection of how ideas relate and evolve across the citation network. To extract traces, we contribute an LLM pipeline that deconstructs salient transitions between pairs of papers and collates the output into an idea dependency graph. We leverage idea traces in a visual system that supports idea-driven navigation starting from a seed paper, lifting the "plane" of synthesis from papers to ideas. To evaluate the usefulness and quality of generated ideas, we provide a case study that compares our approach with an LLM Q&A workflow, and interview 10 researchers who used our system and annotated 110 ideas. Experts found that our approach helped extract a useful space of connections that would have been difficult to make otherwise. Drawing on their feedback, we outline several promising directions for future research.

Abstract: Generative artificial intelligence is rapidly entering knowledge work, and its short-term productivity benefits are increasingly documented. Less is known about what happens, over time, to the knowledge, skills and judgment that remain in people and organizations. This integrative narrative review addresses that question through the construct of cognitive capital: the accumulated, renewable and depletable stock of knowledge, cognitive and metacognitive capabilities, and practices that enables an individual or an organization to learn, interpret information, exercise judgment, solve problems and adapt. We distinguish individual and organizational cognitive capital from the capability of the human-AI system, so that gains in joint performance and losses in human capability can be observed at the same time. Drawing on experimental, field and conceptual work, we propose four mechanisms through which interaction with AI affects cognitive capital: substitution of practice, scaffolding and feedback, metacognitive calibration, and organizational aggregation. Seven principles summarize the current evidence, including that assisted performance differs from retained competence, that the same technology can support or displace learning depending on design and use, and that individual gains can coexist with reduced collective diversity. For each mechanism we derive testable propositions, and we discuss five design levers for organizations as proposals to be validated rather than established practice. We conclude that cognitive capital offers a multilevel frame for governing AI adoption and outline a research agenda centered on longitudinal, multilevel measurement.

Abstract: A retrieval-augmented language model answers from two sources, the retrieved context and its own knowledge, and the two disagree whenever the context is incomplete or wrong. Existing methods settle the disagreement with a rule fixed at design time: follow the context, amplify it in proportion to how far it moves the model, or trust the model's memory. Each rule is confidently wrong in some deployment. In a decision-theoretic model of two sources, the score of any fixed rule is linear in the share of reliable contexts while the best attainable score is convex, so every fixed rule is beaten wherever it is not exactly optimal; and a symmetric divergence between the two sources, the signal adaptive decoding uses, carries no information about which one is right. Resolving a conflict therefore needs two inputs that no existing method keeps: evidence about each source separately, and a prior on how often the context is right. We test these claims on GRACE, a benchmark that intervenes on the knowledge-graph context of a fixed question, so that what the model knows, what the context says and what is true are measured separately. There, every baseline from four families is strong on right contexts or on wrong ones, never on both. We then build the two inputs: the model states its knowledge as notes, a small policy voices each note with a calibrated confidence, a reader answers in two voices, and a selector commits or abstains under a prior $π$ applied as an exact prior shift. With $π$ set to the true share of right contexts, the system is above the best baseline at every mixture of GRACE contexts, by up to 26 points. Its overall score is 45.3 and 35.0 (Qwen3.5-35B-A3B, Llama-3.1-8B), against 29.4 and 27.5 for the best baseline. On preregistered ClashEval it beats same-backbone CK-PLUG and TCR by 12 and 26 points. Where the model lacks the knowledge, the gains vanish, as the theory predicts.

Abstract: This paper proposes a five-level organization of message formation and understanding within thinking and activity in which ASI participates. Its central thesis is that a language model's context is fundamentally incapable of presenting the foundations of utterance formation for examination and verification. These foundations are established in advance by acquired schemas, systemic representations of the world of activity, and connections accepted as true within them. Neither context nor trained LLM weights, in themselves, attest to the presence of deliberately constructed systemic representations. The five levels distinguish systemic representations of the world of activity, a particular situation cut out from those representations, its narrative description, the full sense of a message, and text. A situation's schema-model is not identical to a story about it. A recipient reconstructs the full sense from an abbreviated text through jointly constructed and agreed-upon representations and joint activity. Reverse work proceeds from the full sense through an extended description and the situation's schema-model to the reconstruction of systemic foundations. The theoretical foundations draw on the systems methodology of G. P. Shchedrovitskiy, the structural differential of A. Korzybski, and the Meaning-Text theory of I. A. Mel'chuk. The contribution is a specific systemic-activity organization of content for examining and adjusting message foundations and controlling the consequences of their transformation. Further research concerns large graphical models of activity on reconstructible systemic foundations, distinct from text generation from context.

Abstract: Information retrieval has long been focusing on connecting people with information they need to complete their tasks. Recently, the meaning of this has changed dramatically as more and more AI agents now consume, retain, and act on information on a user's behalf. Consequently, information management now also takes place \emph{inside} the agent in the form of memory management. In this perspective paper, we argue that \emph{agentic memory management} (AMM), which studies the structure, analysis, organization, storage, searching, retrieval, and \emph{use} of information in an intelligent agent's memory, is a natural extension of IR research and an area in which the IR community can make distinctive contributions. Specifically, we first present a system-level framework for AMM and introduce a computational taxonomy of memory objects as \emph{constants}, \emph{functions}, and \emph{operators}, each with distinct retrieval semantics. Then, we compare AMM with established IR research and identify their difference and connections. Last but not least, we outline a research agenda covering benchmarks, candidate principles of memory utility, optimization procedures that mix parametric and non-parametric updates, and more. We hope this paper could attract more people, both inside and outside the IR communities, to think more about AMM problems and research from IR perspectives.

Abstract: Recent advances in search agents have enabled large language models to tackle complex tasks by retrieving and synthesizing external information. However, existing search agents often lack search-state awareness, i.e., the ability to recognize what information has been obtained and what remains missing, leading to hallucinated reasoning and prematurely terminated search. To address this limitation, we propose Self-Assessment Reinforcement Learning (SARL), a reinforcement learning framework in which models learn to recognize their own search state during training and improve their search behavior accordingly, without requiring additional assessment at test time. SARL achieves this through a dual reward design that captures the corrective effect of each assessment while penalizing excessive reliance on it, jointly encouraging effective and efficient search trajectories. Under this setup, SARL achieves a 24.3% relative improvement over the strongest baseline, with average gains of 35.9% on multi-hop reasoning and 27.9% on deep research benchmarks against the strongest per-benchmark baselines. SARL further surpasses methods built on auxiliary supervision or multi-agent coordination, despite operating as a standalone RL framework with no inference-time scaffolding. Beyond final performance, SARL maintains training stability under extended search horizons, a regime where existing methods consistently degrade, enabling sustained gains as task complexity grows.

Abstract: Retrieval-augmented generation over multi-viewpoint queries must preserve distinct answers and positions from the retrieved evidence while operating under a fixed context budget. We introduce LMVU, a lightweight ranker that learns to prioritize candidates by the marginal viewpoint coverage they add to the current selection. Training uses viewpoint annotations from each benchmark's development split to compare candidates within the same selection state. At inference, LMVU repeatedly scores budget-feasible candidates as the context grows, without accessing these annotations. We evaluate two forms of recovered viewpoint coverage: RVC$_1$ assigns each viewpoint full credit for at least one supporting source unit, while RVC$_2$ assigns half credit for one and full credit for two or more distinct source units. Exact oracles measure achievable coverage under the same evidence pool and budget. Across four viewpoint-annotated benchmarks at a 1,024-token budget, LMVU achieves gains of up to 0.047 and 0.052 points in RVC$_1$ and RVC$_2$, respectively, over the strongest non-oracle baseline for each benchmark and metric, and reaches the exact viewpoint-breadth ceiling on two benchmarks.

Abstract: This technical report documents the design and implementation of Scriptorium, a digital library search interface designed to support multi-human, multi-AI collaboration: full cross-user transparency of search activity combined with a Manager--Worker architecture of seven role-specialized generative AI agents. A companion short demonstration paper describes Scriptorium's interaction design; because a demonstration format is constrained to a few pages, it necessarily omits most implementation detail. This report documents Scriptorium's complete system architecture, data model, collaborative-interaction design rationale, and AI orchestration mechanism, including verbatim agent prompts, the deterministic guardrails that constrain them, the language model and retrieval configuration, and the steps needed to run the system, so that the design can be inspected, critiqued, reproduced, and reused by other researchers. We also discuss the system's current limitations and the user study we are planning to evaluate its central hypothesis: that combining collaborative transparency with scoped multi-agent AI assistance increases serendipitous cross-pollination between collaborators' research threads.

Abstract: Long-running agents rely on memory of their past interactions for task state, such as what has been completed and which conditions still hold. Such state is usually a derived state. No record states it, and it must be computed by rules from multiple records of an evolving history. Existing memory benchmarks score only the final answer, with no reference for the intermediate states behind it. A wrong answer therefore cannot be traced to the state that was missed, and a state that silently became outdated goes undetected. We present MemDerive, a diagnostic benchmark for agent memory over evolving data. It specifies each scenario symbolically, derives exact gold answers with Datalog, records the gold derivation DAG and the supporting events of every query, and renders the history in natural language with fictional entities and distractors. Its 10 scenarios contain 4,692 events and 1,503 queries whose difficulty varies along access, evolution, and composition. On this ground truth, MemDerive evaluates answer reliability, diagnoses failures by probing the nodes of the gold derivation DAG, and tracks whether answers stay correct as events arrive. Our evaluation of four memory systems reveals failures that final-answer benchmarks hide. On derived state, every memory system falls behind simple BM25 retrieval, although the same answer model reaches 96.01% when given the supporting events. Most failures are ungathered, as a system holds the missing intermediate state but does not retrieve it for the question that needs it. Memory systems also rarely propagate implicit changes, which no new record states. These results suggest three principles for memory design. Gather inputs along the derivation, store rules with their versions, and maintain derived states through recorded dependencies.

Abstract: LLM-based search agents demonstrate strong capabilities in long-horizon information seeking and reasoning, typically trained via reinforcement learning. However, existing approaches predominantly rely on trajectory-level outcome rewards, providing limited supervision for complex search and reasoning processes. Although step-level credit assignment offers a promising solution, obtaining reliable and task-grounded supervision for open-ended search remains challenging. In this paper, we propose Evidence-Guided Group Relative Policy Optimization (EG-GRPO), a framework that derives step-level supervision from the evidence annotations available in synthetic deep search data. EG-GRPO identifies the first supporting citation for each required evidence claim and uses the newly acquired evidence to construct step-level process rewards. By representing the accumulated evidence as an evolving knowledge state, EG-GRPO further performs state-conditioned relative advantage estimation and combines the resulting step-level signal with the standard trajectory-level outcome advantage. Experiments on five challenging deep search benchmarks show that EG-GRPO generally outperforms standard GRPO and existing reward-enhanced and step-level RL baselines across different model scales. These results demonstrate that task-grounded evidence annotations can provide effective process supervision for credit assignment in deep search RL.

Abstract: Prompt engineering has become a central mechanism for guiding the behavior of Large Language Models (LLMs) across a wide range of reasoning and language understanding tasks [1], [2]. However, existing studies frequently evaluate prompting techniques using different LLMs, datasets, or task settings, making direct comparisons across methods and domains difficult [3]. Consequently, in this paper we present a controlled cross-domain evaluation of Zero-Shot prompting, Few-Shot prompting, Retrieval-Augmented Generation (RAG), and automatic prompt optimization based on a PE2-inspired optimization pipeline [4]-[6]. All experiments were conducted with the same LLM and evaluated on six benchmark datasets covering arithmetic reasoning, logical reasoning, commonsense reasoning, and factual verification. The results show that no prompting strategy consistently outperforms the others across all domains, and pairwise statistical comparisons of the highest-performing configurations revealed no significant differences. RAG achieves the highest overall average accuracy of 78.33% when 300 examples are available for retrieval, while Few-Shot Prompting achieves comparatively high accuracies across several structured reasoning tasks. Zero-Shot Prompting remains competitive on tasks with self-contained inputs, whereas the PE2-inspired optimization approach produces stable but not systematically superior results compared to the other evaluated prompting techniques. Furthermore, increasing the number of task-specific examples does not consistently improve performance across prompting techniques, indicating strong domain and task dependency. These findings suggest that prompting techniques should be selected adaptively based on task characteristics rather than treated as universally applicable optimization strategies.

Abstract: Most retrieval-augmented systems treat literature search as a single retrieval step followed by summarization. Real literature research works differently: the investigator's understanding of the question changes as they read, new subquestions appear, and evidence is weighed, not just collected. We present Optimist Deep Research, an open-source agent that runs literature research as an iterative evidence-acquisition process over an explicit, mutable research state. A research question is broken into a tree of subquestions, each served by search intents that are translated into source-specific queries over arXiv, OpenAlex, Crossref, PubMed, Semantic Scholar and the web. Results are merged into a canonical, provenance-preserving document registry through a conservative four-tier deduplication hierarchy. The registry is expanded by citation snowballing until a saturation criterion is met, and then screened, ranked and allocated to subquestions round-robin. Documents are read in parallel into a structured evidence store of claims with support, location, type and strength. A critic reviews the state along eight dimensions and returns research actions rather than a binary verdict: it can add, split, rephrase, deprioritize or close subquestions, so the research plan is an output of the investigation as well as an input. The final report is written section by section from the evidence store, and every citation resolves to a retrieved document. We describe the design, the deterministic machinery that keeps a language-model pipeline auditable, and a case study that read 28 of 237 retrieved documents in about three and a half minutes. We also report weaknesses the case study exposed.

Abstract: Graph-based retrieval-augmented generation (RAG) exploits relationships among evidence to support multi-hop question answering. However, conveying retrieved graphs through non-unique linear textual contexts to Large Language Models (LLMs) leads to graph-structure degradation and inefficient KV-cache reuse. Meanwhile, graph-structure order optimization is already NP-hard, and the additional KV-cache reuse objective further complicates the search by introducing potentially conflicting orders. To tackle these issues, we propose FuseRAG, a cost-efficient graph-based RAG framework that shifts most retrieved evidence from explicit textual contexts to latent representations. Our main idea is to separate reusable evidence content from query-aware structural interactions in latent space, while retaining only a few explicit facts as semantic anchors. The resulting latent representations are injected into decoder self-attention, where they reshape attention over evidence and thereby influence hidden states and next-token probabilities. This design preserves graph structure through latent interactions, enables evidence reuse beyond exact token-prefix matching. Experiments on three multi-hop question-answering datasets with Llama-3-8B and Qwen3.5-9B show that FuseRAG improves answer quality while achieving up to 2.7x and 2.2x TTFT speedups over the fastest graph-based baseline, respectively.

Abstract: DeepSearch agents achieve strong performance on complex information-seeking tasks through iterative search and reasoning. However, they often issue intent-misaligned, under-constrained, or redundant search queries, resulting in poor evidence coverage and accumulating retrieval noise. To mitigate these limitations, we propose QREF, a plug-and-play Query Refiner and Evidence Filter. Specifically, the Query Refiner rewrites search queries according to the current reasoning intent while ensuring complementary coverage across each search batch. The Evidence Filter retains evidence relevant to both the current search batch and the reasoning objective, reducing distracting retrieval results. We initialize the Query Refiner with supervised fine-tuning and optimize it through reinforcement learning based on interactions with a real-world web search environment, reducing ineffective queries. Comprehensive experiments demonstrate that our approach improves the performance of DeepSearch agents across four challenging benchmarks, achieving average relative Pass@1 improvements of 33.55% on WebSailor and 11.04% on TongyiDeepResearch without modifying architectures or retraining their reasoning policies. Compared with the original DeepSearch backbone without QREF, the RL-trained refiner reduces the fraction of query batches that lack discriminative retrieval anchors by 95%. Compared with the SFT-trained refiner, it also lowers the filter-based ineffective query ratio.

Abstract: Iterative retrieval-augmented generation (RAG) accumulates knowledge from retrieved evidence to answer complex questions, but existing methods struggle to reconstruct entity correspondences and inferential dependencies across retrieved passages as new information emerges. We propose RCER, an iterative RAG method that reconstructs these cross-evidence relations directly from accumulated source passages. Inter-round relational evidence acquisition identifies entity correspondences and relational dependencies to drive dynamic problem decomposition and iterative retrieval. Reasoning context reconstruction independently constructs a coherent reasoning context through joint selection of complementary passages and derivation of cross-evidence inferences. Cross-evidence compositional reasoning leverages these inferences to establish entity correspondences and compose relations across selected passages for answer derivation. Experiments show that RCER outperforms the best baselines by 2.83 EM and 3.06 F1 points on average across three multi-hop QA benchmarks and achieves superior performance on narrative and implicit reasoning tasks.

Abstract: Many real-world questions demand not a single fact but an organized collection of them, yet existing factual knowledge benchmarks almost exclusively target single answer retrieval. We introduce ListQA, a benchmark of 9,045 human-curated, cross-validated questions (audited error rate of 1.7%) that require LLMs to recall multiple facts and compose them into structured lists of 3-10 elements. Questions range from flat lists ("Which NCAA teams went undefeated between 2000 and 2024?") to hierarchical lists with sub-attributes ("...and what was their record and result?"), spanning eight categories across 60+ countries. An evergreen design (explicit temporal constraints or historically immutable facts) keeps ground truth valid without periodic updates. For evaluation, we frame element matching as a linear sum assignment problem: optimal bipartite matching paired with LLM-based semantic grading jointly captures factual accuracy, hallucination, and format compliance. We benchmark 34 models (8 families, 42 configurations across standard and thinking enabled inference) and find: the best standard-inference model scores below 37% LLM-Judge-F1, multi-faceted questions are harder across the board, and providing the source Wikipedia page lifts scores dramatically, confirming that models can organize facts but struggle to retrieve them from parameters alone. Extended analyses cover base vs. instruction-tuned checkpoints and token efficiency. We release the dataset and evaluation code.

Abstract: We argue that tacit knowing is better modelled as an operating regime of one memory substrate than as a kind of content. The substrate is VaCoAl, an algebro-deterministic XOR-and-shift hyperdimensional memory over GF(2), realized as CASRAM (commercial SRAM-CAM) and PyVaCoAl (DRAM-CAM).In VaCoAl/PyVaCoAl a directed acyclic graph (DAG) reads two ways. Repairing every collision (Rescue mode) makes the trace exactly auditable but ranks no path higher; tolerating collisions (Don't Care mode) yields a path-integrated confidence, CR$_2$, that silently favours direct routes. This audit-valuation complementarity is a formal counterpart of Polanyi's claim that tacit and explicit knowledge are not sharply divided. It separates contingent tacitness, which a new query brings out (AlphaGo's game tree), from regime-bound tacitness, statable only in a regime where it is gone (AlphaGo's hunches). (1) Polanyi's from-to structure reads as non-commutative binding, the Identikit as role-filler composition, recognition as majority-vote coincidence. (2) Explicating the valuation destroys it, as attending to his fingers can make Polanyi's pianist falter, yet a closed-form law states it. Polanyi, Nonaka, Tsoukas and Collins each capture part of this but narrowly; even Nonaka and von Krogh's continuum cannot tell contingent from regime-bound tacitness. (3) Collins's three tacit knowledges become strata, one outside the substrate. (4) Our mentor-student DAG is a tacit-transmission graph. (5) Terence Tao's remark that AI-solved problems spawn few follow-ups is diagnosed as rootless explicit knowledge: checkable results whose route no one can follow and that no one yet understands well enough to extend, teach or judge. To re-root them, LLMs generate, VaCoAl/PyVaCoAl keeps an auditable trace with its confidence, and people value, a role virtue epistemology shows cannot be offloaded. Testable predictions follow.

Abstract: Deep Research agents rely on long-horizon reasoning and iterative tool use to solve complex information-seeking tasks. However, adaptive decision-making over these trajectories remains challenging, as agents may stop prematurely when further evidence is needed, continue unnecessarily after sufficient evidence has been gathered, or fail to redirect unproductive trajectories. Existing solutions either require costly model retraining and dense turn-level supervision, or rely on heuristic control rules that are not directly optimized for downstream research utility. This motivates a lightweight, utility-aware alternative for Deep Research control. We formulate adaptive rerouting in long-horizon Deep Research as a state-conditioned trajectory correction problem. Based on this view, we introduce RIVO, a lightweight plug-in framework for adaptive trajectory correction. The agent estimates its research state from lightweight signals and determines whether trajectory redirection is needed. Once triggered, the plug-in modifies the Host token logits to steer generation toward a more promising research direction. Theoretical analysis shows that, under a value-alignment condition and bounded local curvature, a sufficiently small logit intervention can locally improve expected downstream research utility. We further construct BrowseCore, a curated dataset with a paired local corpus, for efficient and controlled Deep Research experimentation. Building on this resource, the plug-in is trained from downstream outcomes, enabling it to learn corrections that improve subsequent research and final task performance. Extensive experiments across multiple Deep Research benchmarks and control settings demonstrate the effectiveness of our method, showing that lightweight logit intervention can improve long-horizon research without retraining the underlying Host model.

Abstract: A stored fact may stop being valid when a user changes a preference or an event replaces an earlier one. We study whether a language model can learn these changes in its parameters and answer correctly from only a question and the current time. Engram Injection stores facts in LoRA adapters and supplies the clock through FiLM conditioning on a frozen Qwen2.5-3B-Instruct model. We compare it with additive clock conditioning, prompting with the full dated history, and retrieving four dated records. We train six matched adapters across three seeds after fixing a sampler that could admit test-age examples into training. The training audit reports no such examples among 127,760 evaluated training inputs. On 62 pairs of questions asked before and after an update, at an elapsed age of seven days excluded from training, both adapter methods answer zero pairs correctly under the primary blinded assistant review with order code 1 held fixed. Full-history prompting answers 59/62 pairs correctly, and retrieval answers 62/62. Separate experiments show that adapters can store facts and that removable memory units can support local deletion. Combining fact selection with time-based version selection remains unsuccessful. These results separate learning a fact from learning when to use it.

Abstract: Existing Open-Ended Deep Research (OEDR) systems primarily generate reports from scratch, making them inefficient for scenarios where research reports need to be continuously maintained as new information emerges. We introduce \textbf{Incremental Open-Ended Deep Research (Incremental-OEDR)}, a research setting that treats a report as an evolving research state and incrementally updates it by preserving valid knowledge, revising outdated or incomplete content, and incorporating newly available information. To support this setting, we propose \textbf{Structured Harness}, which represents reports as structured collections of outlines, sections, and supporting evidence, and provides structured retrieval, a persistent structured evidence pool, and structured generation for selective report updating and evidence reuse. We further establish a temporal evaluation framework spanning ten years, with \emph{Single-Step Task} and \emph{Long-Chain Task} to evaluate incremental updates over both individual transitions and long-term update chains. Extensive Experiments on DeepResearch Bench and DeepConsult under both the Open-source Configuration (OC) and Proprietary Configuration (PC) show that Incremental-OEDR maintains competitive report quality while substantially improving report continuity and reducing research costs. As shown in Figure~\ref{fig:profile}, it achieves up to 0.51 higher content-level ROUGE-L F1, 0.63 higher outline-level EM F1, 33\% lower token consumption, and 61\% fewer search calls than OEDR on DeepResearch Bench. For more details, please refer to our project page: https://ioedr-project.github.io/.

Abstract: Retrieval-augmented generation systems increasingly rely on document-structure treatments: structure-aligned chunking, LLM-generated chunk contexts, heading-path metadata, and hierarchical two-stage retrieval. Separate studies support each on different corpora, embedders, and metrics, and none control for a shared confound: any text prepended to a chunk perturbs its embedding. We present a mechanism-isolating ablation testing all four treatments under one protocol, matching chunk sizes across conditions and adding a semantically null placebo---heading paths that are structurally valid but shuffled across documents. We score retrieval with a coverage-aware nDCG and test four pre-registered contrasts via document-clustered bootstrap with Holm correction, on two distant corpora: 200 Wikipedia Featured Articles (951 queries) and 1,585 QASPER papers (4,303 questions). Organization helps, and the cause is content, not tokens: structure-aligned chunks with real heading paths beat contextualized fixed windows (+0.022 / +0.012 cov-nDCG@10) and the placebo (+0.010 / +0.016). Naive two-stage hierarchical retrieval hurts (-0.033 / -0.015), traceable to first-stage section recall. Gold structure beats LLM-induced structure on Wikipedia but not on QASPER. Effects are small ($dz$ 0.06-0.11) but Holm-significant and consistent across corpora.

Abstract: Organizations have long sought to make workers' expertise reusable by others. Agentic AI changes the nature of such reuse by enabling AI systems to act on workers' knowledge with limited human involvement. This raises the question of how this shift shapes workers' willingness to share their expertise. We conducted an experiment with knowledge workers who created materials incorporating domain knowledge and decided whether to authorize human or AI reuse. We find that AI and human reuse differ primarily in whether workers choose to share their knowledge, rather than in what they choose to share. Participants' reasoning shifts from prosocial considerations when sharing with humans toward concerns about loss of control, replacement, and downstream governance when sharing with AI. These findings suggest that AI reuse may intensify tensions between organizational knowledge reuse and contributors' interests. We discuss implications for workplace AI and knowledge management policies that preserve workers' rights and agency.

Abstract: Natural-language access to RDF knowledge graphs is a core Semantic Web ambition. Large language models (LLMs) have advanced Text-to-SPARQL, yet on unfamiliar graphs they often generate valid queries that misrepresent the populated data model. QRAKEN is a training-free, ontology-agnostic neurosymbolic pipeline grounding generation in empirical graph evidence rather than schema expectations. An offline distiller produces TTQL, a compact description of populated multi-hop patterns, conditional frequencies and path-conditioned literal examples, plus a class-property co-occurrence matrix. Online, TTQL guides the LLM, while deterministic syntax, vocabulary and data-model checks provide diagnostics for iterative refinement. On CK25 (First International Text2SPARQL Challenge), under matched-condition recomputation on a QLever snapshot, QRAKEN achieves strict F1 of 0.643 $\pm$ 0.026 with GPT-4.1 mini and 0.652 $\pm$ 0.012 with GPT-5.4: relative gains of 30% and 32% over the strongest recomputed participant, outperforming systems using the same base model family. Ablations identify TTQL patterns as the dominant driver (+0.31 strict F1 over a shape-only baseline); the refinement loop provides a cheap safety net, rejecting triple patterns unsupported by the co-occurrence matrix. Compared with auto-derived SHACL, TTQL yields 64% higher strict F1, supporting the value of empirical patterns beyond schema exposure. With two local 35B 4-bit open-weight models at zero marginal cost, the same pipeline matches the strongest recomputed participant, and TTQL advantages over shape-only and SHACL baselines persist. Results on a single, relatively small benchmark provide an initial empirical signal; monolithic TTQL injection on very open cross-domain graphs remains the main limitation.

Abstract: A long-running assistant cannot keep everything it has seen, so it summarises. Summarising is not neutral: what is kept is chosen against some notion of what the record is for, and that choice is made once, before anyone knows which of the user's standing goals will ask. Goals rarely disagree about what happened. They disagree about which parts of it were worth the space. Once the history is too long to re-read, the summary replaces the stream, and whatever it left out is gone. We ask what a memory should summarise for when it serves several standing goals at once. Three policies answer differently: summarise with no goal in view, write one summary covering every goal, or write one summary per goal and read them together. We compare them across several models and event streams, holding the read step fixed so that only the write differs. The goals do pull apart: summaries written for different goals overlap each other less than a summary overlaps a rewrite of itself. Per-goal summaries win on relevance, completeness and accuracy, and the all-goal summary loses even to the neutral one written at a fraction of its budget. Interpreting at write pays off, but only for the goal that later asks.

Abstract: Large language models (LLMs) increasingly serve as general-purpose interfaces to factual knowledge, but their parameters do not automatically reflect information that changes after pretraining. Knowledge editing (KE) provides a targeted alternative to costly retraining by modifying selected knowledge and preserving unrelated knowledge and general capabilities. Conventional KE uses structured factual triples, whereas unstructured KE (UKE) uses free-form passages containing multiple facts. Nonetheless, existing UKE editors exhibit a failure mode known as context reliance: edited LLMs can often reproduce the editing passage but fail to reliably recall its individual facts without the original passage context. We identify context-induced difficulty underestimation under the standard passage-level editing objective: later facts receive increasingly rich ground-truth context and consequently incur lower initial losses, making them appear easier to learn. In response, we propose FOVEATED, a plug-and-play framework that constructs focused views of each sentence by randomly shifting the Rotary Position Embedding (RoPE) positions assigned to the keys of its preceding context. The perturbation is applied during editing and removed afterward, leaving the model's native positional encoding unchanged at inference time. We instantiate FOVEATED for both direct-optimization and locate-then-edit editors. We theoretically analyze how FOVEATED counteracts context-induced difficulty underestimation and empirically demonstrate consistent improvements across five KE editors, two LLM backbones, and three benchmarks.

Abstract: Retrieval-Augmented Generation (RAG) is now the standard way to ground Large Language Models (LLMs) in external knowledge, yet the design space of retrieval pipelines is large and the trade-offs between variants are not well understood, especially on domain-specific corpora at realistic scale. In this work, we present a controlled comparison of six retrieval strategies for scientific question answering: (i) classic top-k dense retrieval, (ii) LLM-based query rephrasing, (iii) query rephrasing followed by LLM-based reranking, (iv) multi-query fusion via Reciprocal Rank Fusion (RRF), (v) an agentic tool-call pipeline in which the generator decides for itself whether to retrieve, and (vi) late-interaction retrieval with ColBERTv2. All six pipelines share the same generator (Meta-Llama/Llama-3.1-8B-Instruct), prompt, and evaluation protocol; the five single-vector pipelines additionally share SPECTER2 embeddings and a Chroma vector store; and all six retrieve from the full corpus of 463,971 arXiv papers dated 2024-2025. To support reproducible, large-scale evaluation, we also release a synthetic question dataset of 19,484 problem-statement and methodology questions generated by Llama-3.1-8B-Instruct from a random sample of 10,000 papers across academic domains (query generation succeeded for 9,742 of them), and every strategy is evaluated on this same query set. We describe the architecture and implementation of each pipeline, release the code and the synthetic question dataset, and evaluate each strategy with an LLM-as-a-judge protocol along multiple quality dimensions, together with direct gold-paper retrieval metrics. The result is an open testbed for studying the cost and quality trade-offs of RAG design choices on a research-literature corpus, and a basis for future work on faithfulness, retrieval robustness, and agentic retrieval.

Abstract: When conversing with large language models (LLMs), users often begin with a simple question and build towards a multi-hop question through follow-up turns. Retrieval-augmented generation (RAG) and its graph-based variant (GraphRAG) have become the dominant approaches for grounding LLM responses in external evidence, yet both are evaluated almost exclusively on single-turn, fully specified queries. We systematically investigate this evaluation mismatch through a large-scale simulation study. Building on prior work on multi-turn LLM evaluation, we transform questions from multi-hop question answering (QA) benchmarks into underspecified conversations and evaluate ten LLM assistants with eight retrieval systems across 1.5 million simulated conversations. Our findings reveal that multi-turn interaction causes widespread performance degradation, incurring relative performance drops of up to 21% and increasing unreliability by 47%, making RAG systems simultaneously less accurate and less reliable. We identify two distinct failure modes behind this degradation. Systems are either lost in translation, where conversational rephrasing distorts the retrieval query, or lost in conversation, where retrieval succeeds but the LLM fails to synthesize evidence distributed across turns.

Abstract: Understanding how two entities are connected often requires tracing multi-hop relations across documents to identify intermediate entities and supporting evidence that explain a connection. This is a task that appears frequently in scientific research and other knowledge-intensive analyses. We formalise this setting as multi-hop relation inference: given two known endpoint entities, we aim to recover the bridge entities and evidence-grounded reasoning chains that connect them across a document corpus, and to generate an explanation grounded in the retrieved evidence. Existing multi-hop RAG systems typically seek an unknown answer entity rather than explicitly recovering the connection between two known endpoints and graph-based approaches often rely on costly LLM-extracted knowledge graphs that limit scalability to large document collections. We introduce ConRAG, which builds a lightweight entity-document graph from entity co-occurrence and LLM-based entity filtering. Its connective retrieval infers and semantically ranks paths between two endpoints. On MuSiQue and 2WikiMultiHopQA, ConRAG consistently improves bridge entity and reasoning chain recovery over strong RAG baselines, while reducing graph-indexing token cost by up to roughly 1.5 orders of magnitude. Our results show that endpoint-constrained path retrieval provides an effective and index-efficient approach to evidence-grounded relation discovery.

Abstract: Most methods that optimize LLM prompts and agent workflows assume that task-specific output schemas, extraction instructions, and evaluation criteria are predefined. For scientific extraction agents, however, a short task goal may not fully determine these components, while specifying them manually is costly. We study the upstream problem of constructing the task-specific configuration from a weak specification containing only a short goal and unannotated reference documents. Rather than treating automatic construction as a fixed preprocessing step, our framework constructs a task-specific schema, extraction instructions, and base training rubrics, then keeps schema construction and extraction instructions editable during optimization. Failure-focused updates concentrate textual-gradient feedback on lower-scoring documents, while training-time evaluation criteria adapt to recurring failures. On a heterogeneous-catalysis literature corpus, automatic construction remains improvable, and optimizing both schema construction and extraction instructions performs best across all four judge-rubric settings, with ablations and blinded human evaluation supporting the proposed formulation.

Abstract: In autonomous deep research, a user provides a task and relevant background, then leaves the agent to conduct an extended investigation without further human intervention. We study how this initial user information is reflected in intermediate actions and how these actions relate to final recommendations. We introduce DRaligned, a counterfactual behavioral evaluation framework built on PDR-Bench. By varying one task-relevant user factor while keeping the remaining context fixed, we compare acquisition requests, working drafts, and final reports. Source-grounded extraction, blinded local judgments, and deterministic aggregation yield coarse directional measurements while leaving ambiguous cases unresolved. Our experiments show that strong user-specific delivery can emerge from a largely shared research process: agents investigate similar broad questions but allocate requests differently, and final recommendations distinguish user conditions more clearly than explicit requests do. Reports can also integrate user factors that were not jointly visible during acquisition. In readable draft-to-report comparisons, recommendations often retain their coarse user-specific direction despite substantial rewriting. Final directional differences recur across tested agent models, execution harnesses, and evaluator models, even as execution paths vary. These findings describe how initial user information shapes autonomous research and clarify the relationship between the process an agent follows and the recommendations it delivers.

Abstract: Scientific literature search often begins with an open research problem rather than a known target paper or a fixed candidate set. We introduce INSPIRE, a benchmark for evaluating agents that search prior literature to make progress on solution-redacted research problems. Each instance pairs a research brief with a target-specific cutoff three months before a later paper and evaluates ranked outputs against graded cited antecedents from that paper's realized research lineage. Search proceeds over an open corpus, while the identity of the target paper and membership of its cited antecedents remain hidden from the agent. Beyond end-to-end retrieval quality, INSPIRE uses logged search trajectories to distinguish three coupled stages: resource exposure, whether useful antecedents are surfaced during search; selection, whether exposed antecedents are retained; and ranking, how effectively retained papers are ordered. Across 476 computer-science targets under a shared search interface and budget, the strongest evaluated agent achieves 0.284 nDCG@10. Results show that current agents more readily recover an isolated antecedent than assemble a broader portfolio of relevant prior work. The stagewise analysis identifies resource exposure as the largest observed bottleneck, with further losses in selection and ranking. We additionally construct replay-valid hindsight demonstrations and show that they improve held-out search without changing test-time information, establishing that the benchmark provides an actionable learning signal. INSPIRE therefore enables both end-to-end comparison and stage-resolved diagnosis in a setting where the agent must construct its own working criterion of relevance.

Abstract: Based on an analysis of the role of language in thought and communication, this article proposes a new concept for designing corporate knowledge bases. The concept integrates the probabilistic vector space of a corporate vocabulary, reflect-ing industry specifics, subject focus, terminology, and culture, with traditional ontological modeling. This combination enables the efficient extraction of knowledge from accumulated corporate documents while strictly accounting for specific business processes. Consequently, this concept bridges statistical and semantic (cause-and-effect) methodologies. Furthermore, analyzing the projec-tions of probabilistic spaces and causal relationships can help identify bottlenecks in business logic. As a dynamic system, language functions as a separate, inde-pendent layer within the overall information architecture. Introducing a dynamic component into the probabilistic space of word distribution allows it to be mod-eled as a multidimensional solution space for various problem formulations. In this context, input data defining the problem conditions serve as control parame-ters for dynamic transformations.

Abstract: Document rerankers determine what evidence reaches the downstream model in RAG and deep research, yet mainstream rerankers select by relevance matching, and individually relevant documents rarely constitute the complete, complementary, non-redundant set a complex information need demands. Prior work rewards a set by its aggregate rubric score, shifting the objective from ranking documents to composing sets. Yet that score is one scalar shared by every document in the set, so the supervision is sparse: a redundant document is rewarded with the rest whenever the set scores well, and a decisive one penalized with the rest whenever it does not; credit assignment leaves contributors indistinguishable from free riders. On-policy distillation could densify this supervision, but existing methods give every rollout the same fixed guidance, too prescriptive for strong rollouts and too abstract for weak ones. We therefore propose AdaTutoRank, a setwise reranker trained with Adaptive Tutoring Optimization (ATO) under a three-level hierarchy of nine rubric dimensions, which supplies silver labels for the cold start, rewards for reinforcement learning, and hints for distillation. ATO draws three hint forms of increasing specificity from the policy's own frozen snapshot: the rubrics alone, a self-selector's sibling-set chosen under rubrics, and a self-reflector's reflection contrasting the rollout with that sibling-set; each rollout receives the form matched to its quality. Re-scoring that rollout under the hint-conditioned frozen teacher and the hint-free snapshot distills the hint's effect into a token-level advantage that complements the group-relative outcome advantage. Across ten benchmarks spanning RAG, deep research, and setwise evaluation, AdaTutoRank attains the best overall performance while issuing fewer retrieval calls.

Abstract: Generative AI is reshaping the cultural infrastructures through which knowledge is found, synthesized, and held accountable. To make sense of this shift, scholars and policymakers reach for historical analogies of technologies such as the printing press, steam power or electricity. But these comparisons are typically imprecise about which property of the technology carries the comparison, and imprecise analogies produce imprecise governance by designing interventions against the wrong property of the system. This paper offers a diagnostic framework for analyzing how generative AI can transform epistemic and cultural practice. This paper offers a diagnostic framework for analyzing how generative AI can transform epistemic and cultural practice. We decompose each intervention into three coordinates: the epistemic site at which a technology acts, the governing logic by which it organizes its object, and the technical mechanism through which the logic is instantiated. This framework allows us to distinguish between structural cultural consequences, which follow from the mechanism itself, from contingent ones, which remain open to design and institutional choice. Applying the framework to information discovery and knowledge synthesis, we show how the shift from indexicality to inference and from editorial authority to statistical consensus produces specific, traceable cultural effects and reveals governance levers that gestalt analogy obscures.

Abstract: Evaluating claim admission in shared agent memory is challenging because repeated claims may be mistaken for independent evidence. An agent may copy or paraphrase a retrieved belief, while admitting a false claim exposes subsequent agents to it. To study this problem, we introduce the Correlated Promotion Benchmark (CPB), which evaluates whether candidate claims should be admitted to shared memory.CPB-Static constructs a frozen test split from publicly annotated sources with fixed gold actions. CPB-Live runs multi-agent teams over a shared store, records all writes and retrievals, and tracks source lineage defined by each scenario. A separate consumer answers from the store alone. We evaluate eight admission policies across four agent families. Our results show that policies which deduplicate sources reject many true claims alongside false ones, whereas policies preserving answer coverage admit nearly as many false claims as unrestricted sharing. Gating on declared source type reduces false adoption to 0.06--0.09, compared with 0.22--0.47 for other answering policies. Once an uncontested false belief enters memory, the consumer asserts it in 0.97--0.99 of probes across all families. No non-oracle policy consistently rejects false claims across verbatim copies, paraphrases, and paraphrases declared authoritative. These findings reveal the limitations of admission policies without access to source lineage.

Abstract: As researchers tackle interdisciplinary problems, they face the need to deepen expertise in primary areas while rapidly acquiring knowledge in secondary domains. Generative AI (GenAI) is increasingly positioned to meet this need, from general-purpose chat assistants to Deep Research tools marketed as autonomous research agents. Prior work has examined how researchers use GenAI to support single-discipline or general research tasks. However, we know little about the goals and GenAI practices in interdisciplinary research. We conducted a longitudinal study and semi-structured interviews with 15 interdisciplinary researchers to examine how interdisciplinary researchers actually orchestrate GenAI. Findings show that researchers leaned on GenAI to fill knowledge gaps while maintaining epistemic agency for novelty discovery. We also uncovered an expertise paradox: GenAI outputs were hardest to verify when most needed. Our empirical insights motivate GenAI designs that calibrate verification to researchers' expertise, nudge toward cross-domain synthesis, and adapt prompting and outputs to disciplinary conventions.

Abstract: Given a large corpus, the questions one might ask can vary -- from "When was the first human heart transplant?" to "What are all the contradictory claims in this literature?" -- but what makes some questions more challenging than others? In this work, we define a notion of Corpus Task Complexity (CTC) that characterizes tasks by how their difficulty grows with corpus size; for instance, a retrieval query only requires a single linear pass over a corpus, while finding contradictions requires checking a quadratically growing set of claim pairs. Observing that prior work has largely only studied tasks whose difficulty grows linearly with corpus size, which we call low CTC tasks, we introduce 10 new tasks belonging to a class of high CTC whose difficulty grows quadratically or more in corpus size. We find that high-CTC tasks not only grow much more challenging on average at longer contexts for LCLMs, they reverse many modeling conclusions drawn solely from low-CTC evaluations. For instance, efficient block-sparse and hybrid attention approaches consistently match full attention performance on low-CTC tasks, but degrade much more on high-CTC tasks. Large-corpus high-CTC reasoning thus remains an open challenge as full attention is too costly to scale, motivating future research on these tasks. We release our code, data, and 22-task suite (CTC-Bench), to facilitate future research in this area.

Abstract: As autonomous AI agents take on every stage of scientific inquiry, research output is expanding far beyond human review capacity. Yet scientific communication still relies on natural-language prose: an informal medium prone to ambiguity, hidden assumptions, and untracked limitations that machines cannot reliably audit. We introduce Lara, a machine-checkable language and protocol for checking and revising support for research claims. By turning research arguments into executable artifacts, Lara provides an epistemic kernel for autonomous science: it enables automated validation pipelines for research agents, lets declared bridges connect arguments across papers into an auditable network, and allows both humans and machines to recheck the standing of an encoded claim in milliseconds. In a Lara program, authors explicitly declare their claims, supporting evidence and assumptions, and known objections or limitations. A lightweight, deterministic checker adjudicates these interactions, assigning each claim a reproducible status: "justified", "defeated", "contested", or "gap", which marks a claim whose support is incomplete and locates the unanswered question. Case studies cover empirical review, a philosophical debate without measurements, and the loss of support when an assumed axiom is withdrawn. We establish the metatheory of claim checking and cross-context argument transport, and mechanize the semantic guarantees in Lean 4 (roughly 117,000 lines), leaving three arguments on paper. The audited public metatheory is "sorry"-free and uses only Lean's three standard axioms; some executable examples additionally trust native evaluation.

Abstract: Traditional Web search follows a human-facing paradigm in which users inspect ranked documents and synthesize information themselves. In AI Search, retrieved documents instead serve as inputs to a generation model, shifting the retrieval objective from ranking documents by Search Satisfaction to constructing reliable context for correct answer generation. We formulate this shift as answer-oriented context construction through a three-stage framework: (1) Answer Support identifies candidate documents that contribute information to answer generation; (2) Content Trustworthiness assesses whether this information provides a reliable basis for correct answers from source, temporal, and factual perspectives; and (3) Context Organization selects, consolidates, and structures retained information under a finite context budget for consistent and robust generation. We further develop an industrial workflow spanning prior and posterior optimization and establish a systematic evaluation protocol covering both retrieval-side context and final answers. Experiments show consistent improvements at both Retrieval and Answer levels, demonstrating the effectiveness of the framework and its industrial implementation.

Abstract: Large Language Models (LLMs) have recently shown strong potential for producing context-rich text embeddings for retrieval. Most existing methods either treat embedding learning as passive feature extraction or exploit LLM reasoning through instruction following for better embedding optimization. However, specialization toward embedding objectives can suppress useful reasoning generation or produce retrieval-irrelevant text. We refer to these two forms of degradation as reasoning collapse. To address this issue, we propose CoFree (Collapse-Free Reasoning Embedding), a two-stage framework that progressively integrates LLM reasoning into query and document embedding optimization while preserving reasoning quality. At the first stage, CoFree applies reference-guided supervised fine-tuning to restore the reasoning ability and retain representational strength of the foundation embedding model. At the second stage, we introduce dual rewards, an embedding-oriented reward and a reasoning-oriented reward, to guarantee fine-grained reasoning of the relevance toward the embedding goal in reinforcement learning. This endpoint-coupled optimization transforms embedding learning from static alignment into a high-quality reasoning-guided search process for retrieval. Extensive experiments demonstrate the effectiveness of CoFree, with CoFree-4B achieving an average absolute improvement of 2.8 nDCG@10 points over Qwen3-Embedding-4B across 22 datasets from MTEB and BRIGHT. Online experiments in a real-world retrieval system further show consistent gains. Code, RTED, and model checkpoints will be made publicly available.

Abstract: We introduce document embedding geometry as a quantitative observable of conceptual reorganization and develop a counterfactual ablation framework for measuring how individual concepts influence the organization of scientific knowledge, providing a quantitative framework for detecting scientific revolutions. The observable is defined by the geometric perturbation induced when removing documents associated with a candidate concept from the embedding space before and after its historical emergence. Statistical validation is performed using five historical case studies spanning physics, mathematics, and machine learning: special relativity, Gödel's incompleteness theorems, the Higgs mechanism, deep learning, and the attention mechanism underlying transformer architectures. Across the historical case studies, the framework identifies measurable geometric signatures associated with conceptual reorganization, while the validation studies expose important limitations arising from document assignment and sparse historical data. These results establish embedding geometry as a medium for quantifying conceptual reorganization, providing a new approach for studying how scientific fields restructure over time.

Abstract: Deep research agents are increasingly capable of web search, tool use, multimodal evidence analysis, and information synthesis. However, existing benchmarks mainly evaluate medium-horizon exploration and rarely test whether agents can sustain long, dependency-heavy research processes. We introduce Mr.LHDR (Multimodal real-world Long-Horizon Deep Research), a benchmark for evaluating real-world deep research over long, irreducible chains of interdependent evidence across eight categories. Each question is constructed from a hidden Node-Relation graph and requires an average of 12.1 necessary intermediate conclusions with a mean dependency depth of 10.4 before reaching a short, unique, and verifiable answer. Questions incorporate multimodal evidence, including images, maps, PDFs, logos, charts, tables, and video frames, with at least one non-text element that changes the reasoning state. Mr.LHDR evaluates both final answers and the correctness of intermediate conclusions under annotated dependencies. We evaluate general models, deep research systems, and agent frameworks using Overall Accuracy (OA), Strict Accuracy (SA), Checklist Score (CS), and Dependency-Aware Checklist Score (DACS). Results show that even the strongest system achieves only 43.1% OA and 34.3% SA, indicating that final-answer accuracy substantially overestimates complete research success. Removing images reduces DACS by 12.6 points, demonstrating the importance of multimodal evidence, while SA consistently declines as reasoning chains become longer. These findings reveal sustained, dependency-consistent evidence integration, rather than isolated fact retrieval, as a key bottleneck for current deep research agents.