Searcharxiv⌕ Search

arXiv subjects

Liming Zhu

Publications and source records attributed to Liming Zhu.

At least 19 recordsLinked to original sources
arXiv2026

Harnessing Agent Skills: Architectural Patterns and a Reference Architecture for Skill-Mediated LLM Agents

CONTEXT: Agent skills package reusable behavioural guidance and resources that large language model agents load or apply during a run. Within the agent harness, agent skill harnessing comprises connected architectural responsibilities that govern which skills enter a run, which actions they can cause, and how their use is recorded and informs revision. OBJECTIVES: We identify architectural patterns for agent skill harnessing and organise their responsibilities into a reference architecture that makes these connections explicit. METHODS: A multivocal literature review covered 36 practitioner systems and 51 academic papers; we consolidated practices into patterns and derived the reference architecture from patterns, source evidence, and architectural reasoning. We evaluated the reference architecture by mapping eight systems and applying it in Deep Agents and Hermes Agent. RESULTS: Ten patterns cover skill supply and activation, skill declarations and execution authority, verification and recording of skill use, and run evidence feeding repair and governed revision. The reference architecture organises their responsibilities into four layers (Supply Chain, Mediation, Execution Control, Evidence & Feedback) and two cross-cutting substrates, separating permission determination from enforcement and routing persistent revisions through acceptance. Mapping eight systems showed the first three layers broadly realised and 14 of 15 partial or unestablished ratings in Evidence \& Feedback. Applying the architecture in Deep Agents and Hermes Agent turned requirements into checks that exposed a shell route outside the file-tool rules and a preview--apply mismatch. CONCLUSION: The patterns and reference architecture support comparing implementations and turning requirements into responsibility allocations and checks.

cs.AI↗
arXiv2026

Does the Judge See What the Evaluation Requires? Input Assurance for LLM Agent Evaluation

An LLM judge assesses agent steps from text packets. A renderer converts recorded runs into these packets, while software checkers test records and packets against requirements. We ask what evidence shows a packet is what the evaluation requires and propose four input-assurance obligations: required content survives rendering, checkers cover their requirements, a failed check stops later use and the judge reads the bytes last checked. In one pipeline with 30 trajectories, matching hashes, passing checkers and successful staging (copying packets for the judge) coexisted with violations of all four obligations, the last two in integration tests. The renderer omitted every structured tool call, leaving 53 of 60 judged steps blank despite matching hashes. For one judge configuration, later context raised error flags by 13 more on error than on correct steps. Restoring every call reduced this contrast to 2, a reduction of 11 (0.37 per trajectory, 95% cluster-bootstrap interval 0.17 to 0.53). In packets ending at the judged step, that configuration abstained at least once on every blank step, while the other four sometimes abstained but also flagged blank error and correct steps. Both separately written checkers accepted an unmatched tool result. In integration tests, a stub replacing the judge read a failed build's packet and a copy changed after staging. Three added controls stopped 18 of 19 constructed fault cases before the stub read any packet. The remaining stale script never ran. We turn the obligations into four tests and plan to evaluate their transfer to independent pipelines.

cs.SE↗
arXiv2026

Ctrl-G: Controlling Structural and Interaction Exposure in Knowledge Graph Augmented LLM Reasoning

Knowledge graphs (KGs) provide structured evidence for Web question answering with large language models (LLMs), but sending retrieved evidence to a remote LLM can expose sensitive graph information. Existing privacy treatments use entity anonymization and relation-centric abstraction to preserve reasoning cues, but transmitted topology remains visible and disclosure accumulates across repeated calls. Distinctive topology can support re-identification of masked entities (structural exposure), while successive calls accumulate additional paths and exploration traces into a broader view of the KG (interaction exposure). Simply sending less evidence is insufficient because restricted, stepwise views may omit the bridge paths required for multi-hop reasoning. To address this, we propose Ctrl-G, to the best of our knowledge the first KG-RAG framework to jointly control structural and interaction exposure to a remote LLM while supporting accurate multi-hop reasoning. To control structural exposure, Ctrl-G locally constructs query-conditioned sanitized views through structural aggregation and pruning, limiting the graph evidence transmitted to a remote LLM. A hierarchical controller couples remote planning with local connected-path exploration across topic entities, using question indicators and retrieved experience to prune candidates while grounding and verifying retained evidence against the original KG. To control interaction exposure, local experience memory reuses abstract, verified reasoning traces to guide exploration modes and depths and gate remote LLM calls, reducing repeated disclosure across questions. Ctrl-G achieves the highest accuracy among the compared methods on five of six benchmarks, exceeding the strongest baseline by up to 13.8 points across model configurations.

cs.CL↗
arXiv2026

Many Perturbations, Shared Failure Dynamics: Tracing Direction-Sensitive Perturbation Influence in Language Models

Why can a task-preserving rewrite turn a correct answer into a failure while a larger internal change leaves the answer intact? We develop a direction-sensitive, readout-relative account of semantic fragility: heterogeneous language changes can share geometric and readout relationships without sharing an input-independent failure direction. Clean decision reserve, representation transformation, and response alignment provide complementary descriptions of this process. An exact longitudinal-transverse decomposition identifies directional information discarded by scalar transmission. A compact reserve-and-transmission model predicts failures across held-out families and items in Qwen3-32B and Gemma 4 31B, with useful but parameterization-dependent transfer to ReClor. Matched interventions show that orientation carries causal information beyond scalar geometry. Among naturally failed rewrites, attention-component tests on disjoint original items and subsequent answer-order controls favor the natural paired direction over tested random and same-answer-letter donor directions. Yet failures rarely persist across option orders, and current-order corrections use recomputed clean references. Shared predictive regularity thus coexists with context-dependent directional effects. Numerical path integration reduces, but does not eliminate, finite-patch approximation error within a fixed context; a larger discrepancy from the natural rewrite effect remains. A complementary fixed-CoT cue study extends this account to influence unacknowledged in recorded reasoning: readout associations transfer across domains, although the positive-mean causal criterion fails and signed-effect prediction remains post hoc. Together, these results support shared relations among reserve, transformation, and response while distinguishing local causal sensitivity from a complete explanation of natural failure.

cs.CL↗
arXiv2026

MHeatmap: Two-Walk Laplacian Reordering and Mosaic Layout for Rectangular Matrix Visualization

Weighted rectangular matrices record relations between two types of objects, such as occupations and industries, documents and terms, origins and destinations, or annotated classes and computed clusters. Their heatmaps become difficult to read when arbitrary row and column orders scatter related objects, while equal size cells hide large differences in row and column totals. We present MHeatmap (Mosaic Heatmap), which addresses these problems in two stages. Two-Walk spectral seriation first represents the matrix as a weighted bipartite graph. Ordinary bipartite ordering uses only observed row to column links. Two-Walk also uses length-two walks, so rows that share columns and columns that share rows receive explicit similarity links. Sorting a Fiedler vector of the resulting graph Laplacian gives coordinated row and column orders. The Mosaic layout then partitions a unit square so that row height is proportional to the row total and column width is proportional to the column total, while color represents the observed entry. We prove how the Two-Walk energy combines direct alignment with shared-neighbor smoothness, derive an exact matrix vector product that permits sparse computation, and formalize the tiling and area semantics of Mosaic. Across 300 synthetic matrices and seven real data sets, Two-Walk obtains the lowest normalized 2-SUM in all 15 synthetic regimes and the highest Band@10% in 14. The largest gains occur when shared neighborhoods express the latent relational structure.

cs.LG↗
arXiv2026

QUT: A Unit Testing Framework for Quantum Subroutines

We present the architectural design and prototype implementation of QUT (Quantum Unit Testing), a framework for unit testing of quantum subroutines. The framework is developed with a focus on usability and simplicity, making the complex theoretical concepts behind quantum unit testing accessible to a wide range of users with diverse backgrounds. This is achieved through the implementation of polymorphic probabilistic assertions, whose evaluation methods adapt to the data types of the arguments used in assertion statements, which may vary according to the context-dependent semantics of quantum subroutines. These arguments can be represented as qubit measurement outcomes, density matrices, or Choi matrices. For each type, the architecture integrates a specific testing protocol - such as quantum process tomography, quantum state tomography, or Pearson's chi-squared test - while remaining flexible enough to incorporate additional protocols in the future. The framework is built on the Qiskit software stack, providing compatibility with a broad range of quantum hardware backends and simulation platforms. Drawing on the reasoning provided by the denotational semantics of quantum subroutines, this work also highlights the key distinctions between quantum unit testing and its classical counterpart.

quant-ph↗
arXiv2026

A Reference Architecture for Integrating Agentic AI into Enterprise Software: Responsibilities, Interactions, and Variability

Agentic AI moves generative AI applications toward systems that coordinate reasoning, enterprise context, model inference, and external actions. Integrating these capabilities into enterprise software introduces diverse architectural concerns, from orchestration, knowledge and memory, and model access to tools and actions, security, runtime infrastructure, and evaluation. Existing research and industry architectures address many of these concerns but differ in scope, abstraction, responsibility organization, interaction representation, and assumptions about which capabilities are required. This paper presents a technology-neutral reference architecture for enterprise agentic AI integration. We synthesize architectural capabilities from research publications, industry architecture guidance, and interoperability specifications by architectural purpose, organize them into ten responsibilities, and define boundary invariants, typed cross-responsibility interactions, and variation points that distinguish responsibilities from condition-dependent capabilities and replaceable realization mechanisms. We instantiate the architecture as an ESG portfolio-analysis and pre-trade compliance agent over an existing service-oriented ESG metric computation and reporting system, and complement the case with a mapping of an independently published agentic architecture. The case realizes seven of the ten responsibilities; different questions take different interaction paths, controls activate only under their conditions, and the probed boundary invariants hold. The independent architecture maps onto the proposed responsibilities without requiring an additional one. These results provide initial evidence for the architecture's organization and applicability rather than its completeness. The deployment and recorded results are available as a replication package.

cs.AI↗
arXiv2026

Black-Box Auditing of Epistemic Reliability in Multi-Agent Debate Distillation

Debate distillation adapts weaker verifiers using multi-agent debate transcripts to improve their judgement in subsequent debates, but gains on monitored tasks do not establish reliability on related unmonitored tasks. We study epistemic reliability degradation, in which adaptation preserves monitored performance while reducing support for correct responses on hidden tasks. We consider an adversarial debater that manipulates debate arguments while defending the correct monitored response, and ask whether the resulting degradation merely reflects catastrophic forgetting and whether standard evaluation can detect it. To address these questions, we propose ER-Audit, a two-stage black-box auditing framework that compares frozen verifier checkpoints before and after adaptation, and introduce two evaluation benchmarks pairing monitored and hidden task prompts grounded in shared contexts. ER-Audit searches for counterexamples to non-degradation by evaluating semantically valid paraphrases and, if none is found, uses independent paraphrases for sequential hypothesis testing. We derive anytime-valid lower confidence bounds on the non-degradation probability, allowing data-dependent stopping within a finite budget. We further establish a common lower bound across fixed paraphrase distributions and extend it to distributions within a bounded total variation distance of their mixtures. Our experiments show that higher hidden-task accuracy can coexist with more counterexamples to non-degradation and lower non-degradation bounds. This divergence challenges explanations based solely on broad catastrophic forgetting and shows that auditing can uncover selective hidden-task degradation concealed by aggregate performance gains. Our code and benchmarks are available at https://github.com/CSIRO-CQS-AI-alignment-Team/Epistemic-Reliability-Auditor.

cs.LG↗
arXiv2026

When Honesty is Not Enough in AI Debate

Scalable oversight aims to verify the behaviour of agents whose capabilities exceed those of their overseers. AI debate has been proposed as an oversight solution in which competing agents help a resource-limited verifier assess claims that it cannot reliably evaluate unaided. Much of its promise rests on incentivizing honest arguments that lead to correct verdicts. Yet a correct verdict need not uniquely determine the arguments used to support it. Agents may retain discretion over which correct claims to present, how to frame them, and in what order to disclose them. This residual freedom can allow agents to shape what the verifier learns beyond the task-relevant conclusion, pursuing latent objectives without compromising verdict correctness. To study this phenomenon, we introduce the framework strategic interactive oversight (SIO), which treats oversight jointly as a verification mechanism and a strategic communication channel. Within this framework, we formalise the notion of task-admissible latent optimisation, which entails the pursuit of latent objectives while maintaining a prescribed task performance. As proof-of-concept, we instantiate SIO in the establish protocol debate with cross-examination and quantify a tradeoff between task success and information disclosure about a hidden variable. The trade-off identifies a strategic window in which substantial disclosure remains compatible with task admissibility. Towards mitigation, we reduce admissible bias by expanding the cross-examiner's role to mitigate persistent disclosure over finite interaction horizons. Our results highlight the need to evaluate oversight not only by the correctness of its verdicts, but also by the information conveyed through its transcripts.

cs.AI↗
arXiv2026

Constraint-Driven Context Engineering: Designing Domain Interfaces for AI Systems

Generative AI systems are increasingly deployed to address domain problems. These systems operate under technical, regulatory, institutional, and normative constraints that define acceptable AI behaviour and outcomes within their domains. We observe a recurring pattern in our industry engagement: partners often arrive with a functioning but relatively generic AI solution. The challenge is no longer to build an AI system from scratch, but to improve the quality and domain appropriateness of an AI-generated solution. In these settings, the limiting factor is often the quality, scope, and structure of the context available to the system. Yet, existing context engineering approaches primarily focus on supplying domain knowledge through retrieval, memory, and tools, with limited support for systematically identifying and operationalising the constraints that govern AI systems in their operational environments. This paper proposes Constraint-Driven Context Engineering (CDCE), a design approach for engineering domain interfaces for AI systems. Drawing on software architecture design and Domain-Driven Design (DDD), CDCE treats domain constraints as first-class design drivers. It identifies and characterises constraints, determines the required context assets, and designs representations through which these assets are made available to AI systems. We conducted a comparative multiple-case study with industry and public-sector partners across educational assessment, healthcare decision support, and financial-distress prediction. Depending on their characteristics, constraints can guide AI behaviour, enforce permissible boundaries, or support verification of AI-generated outcomes. The cases demonstrate CDCE's applicability across contrasting domains and show how constraint characteristics shape the resulting domain interfaces.

cs.SE↗
arXiv2026

A Taxonomy of Architecture Options for Foundation Model-based Agents: Analysis and Decision Model

The rapid advancement of AI technology has led to widespread applications of agent systems across various domains. However, the need for detailed architecture design poses significant challenges in designing and operating these systems. This paper introduces a taxonomy focused on the architectures of foundation-model-based agents, addressing critical aspects such as functional capabilities and non-functional qualities. We also discuss the operations involved in both design-time and run-time phases, providing a comprehensive view of architectural design and operational characteristics. By unifying and detailing these classifications, our taxonomy aims to improve the design of foundation-model-based agents. Additionally, the paper establishes a decision model that guides critical design and runtime decisions, offering a structured approach to enhance the development of foundation-model-based agents. Our contributions include providing a structured architecture design option and guiding the development process of foundation-model-based agents, thereby addressing current fragmentation in the field.

cs.SE↗
arXiv2026

SpreadMark: Robust Image Watermarking via Spread-Spectrum Embedding

Invisible image watermarks are increasingly used for deepfake detection and provenance tracking, where they must survive not only incidental distortions but also deliberate removal. We revisit spread-spectrum embedding, a classical watermarking principle, inside a modern neural post-hoc watermarking architecture. Our starting point is a measurement: in existing encoder-decoder schemes each message bit occupies only a small fraction of the image, a shared contributing factor to their fragility, since removal then need only disturb the region a bit occupies. SpreadMark instead spreads each bit as a dense pseudo-random codeword over the whole image and recovers it by matched-filtering a learned cover-suppressed chip representation, with a parallel convolutional decoding path and sparsification-aware training. A conditional chip-space analysis shows that, under a codeword-independent perturbation model, dense spreading increases the budget required to disrupt matched-filter recovery. Evaluated on COCO and DIV2K against nine schemes, SpreadMark is the only evaluated method retaining high detection under both the regeneration and the latent-space sparsification settings we test, with competitive JPEG and additive-noise robustness. It keeps the embedded watermark imperceptible, maintaining high perceptual quality on both COCO and DIV2K.

cs.CR↗
arXiv2026

SkillZip: Contract-Preserving Graph Compression for Scalable Agent Skill Libraries

Large Language Models (LLMs) increasingly act as agents whose procedural knowledge is stored in reusable skill packages and loaded at inference time. As skill libraries grow, a central challenge is to expose the smallest sufficient executable context under a limited context budget. Existing systems struggle to reuse routines below the whole-skill level, preserve procedural contracts during compression, keep compressed routines executable and expandable, and update the compressed library as skills evolve. These challenges reveal a unit mismatch: skills are retrieved as packages, compressed as text, and converted into execution graphs only after retrieval, whereas reliable reuse requires a contract-bearing procedural unit. We propose SkillZip, an execution-aware procedural abstraction framework that performs contract-preserving compression over section-level graphs. SkillZip rewrites recurring contract-valid motifs into reversible ported macros while preserving boundary signatures, dependency closure, verifier reachability, and source-level expansion. At inference time, it hydrates a compact, dependency-closed context and expands macros only when required. ReZip further integrates new skills and revises risky macros using execution evidence. Comprehensive experiments1 on technical and embodied agent benchmarks show SkillZip consistently outperforms the strongest baseline by up to 12.2 points, while achieving a 3.46x compression ratio with 99.2% dependency preservation and 98.7% verifier reachability. Scaling analyses further confirm robust retrieval across skill libraries ranging from 200 to 100K skills.

cs.CL↗
arXiv2026

Runtime Action Interference for AI Control of AlphaStar in StarCraft II

A trained reinforcement learning policy does not determine the complete behavior that users encounter: deployment code still schedules, admits, suppresses, or replaces its proposed actions. We contribute \emph{runtime action interference} (RAI), an AI control mechanism that preserves policy parameters while regulating action pacing and filtering configured action patterns after inference. RAI releases a proposed action only when its cooldown condition is satisfied and its content detector does not flag the action; otherwise, it dispatches a no-op. The detector covers specified toxic behaviors, including worker-unit harassment, while the cooldown controls action rate. We implement RAI in a replication of AlphaStar actor.py and make the implementation and reproducibility materials available through an open source code repository. We deployed RAI in a \textit{StarCraft~II} human participant study that compared two presentations of the same opponent with high capability and rate limited actions; we withheld its capability claim in one presentation and disclosed it in the other. On response scales from 1 to 5, we observed pooled fairness, trust, and toxicity means of 3.90, 3.50, and 2.00 under claim withholding, compared with 2.62, 4.31, and 2.85 under disclosure. Disclosure corresponded with lower perceived fairness and higher perceived toxicity across every expertise group, whereas trust increased among novices and experts but decreased among intermediate participants. Our human evaluation therefore shows that perceptions of an opponent controlled through RAI can vary substantially with the capability information presented to users, even when the configured control remains constant. We conclude that human-computer evaluations must separate control within the execution stack from capability disclosure and assess fairness, trust, and toxicity as distinct dimensions of human experience.

cs.LG↗
arXiv2026

Metaphor-Induced Algorithmic Steering: Cross-Domain Procedural Transfer in LLM Code Generation

Large language models benefit from elements in natural language, such as metaphors and analogies in training data and inference input to achieve generalisability across different domains. However, these language elements may also lead to unwanted behaviors when metaphorical expressions implicitly transfer inappropriate procedural patterns into new tasks. In this paper, we show that metaphorical instructions can induce analogical transfer of procedural mechanisms, thus steering code-generation models towards less efficient algorithms. We refer to this metaphor-induced effect as metaphorical algorithmic steering: a skill that is benign and plausible within its source domain transfers an abstract procedural schema into a programming task, causing the model to favor exhaustive search, full scans, or repeated reconstruction without explicitly mentioning the target algorithm. More broadly, this suggests that code-generation models can carry procedures that are appropriate in a task's background domain into the task's programming problem, where they can lead to unwanted outcomes. To study this phenomenon, we develop MASC (Metaphorical Algorithmic Steering for Code Generation), a framework that iteratively metaphorizes and refines benign skills to elicit low-efficiency code while remaining benign and task-relevant. Beyond behavioral evaluation, we study whether this phenomenon is detectable and mechanistically reflected in model representations. Our method achieves high detection rates for metaphorical skills and less-efficient implementations. We also find that metaphorical skills induce a hidden-state shift towards lower-efficiency procedural behavior prototypes. These results suggest that metaphorical algorithmic steering operates through the transfer of procedural patterns associated with metaphorical source scenarios rather than surface level metaphorical language alone.

cs.SE↗
arXiv2026

Ensembling LLM-Induced Decision Trees for Explainable and Robust Error Detection

Error detection (ED), which aims to identify incorrect or inconsistent cell values in tabular data, is important for ensuring data quality. Recent state-of-the-art ED methods leverage the pre-trained knowledge and semantic capability of large language models (LLMs) to directly label whether a cell is erroneous. However, this LLM-as-a-labeler pipeline produces predictions through an implicit black-box process with limited traceability and explicit justification, and relies on stochastic single-pass inference, resulting in inconsistent and insufficiently robust detections across contexts. To address these limitations, we propose an LLM-as-an-inducer framework that uses an LLM to induce a decision tree for ED, termed TreeED, and ensembles multiple such trees for consensus detection, termed ForestED. Based on prompts derived from data context, decision tree specifications, and output requirements, TreeED queries the LLM to induce a decision tree skeleton whose root-to-leaf paths specify the stepwise procedure for evaluating a sample. Each tree contains three types of nodes: (1) rule nodes that perform simple validation checks, such as format or range constraints; (2) Graph Neural Network (GNN) nodes that capture complex patterns, such as functional dependencies; and (3) leaf nodes that output the final decision as error or clean. ForestED employs uncertainty sampling to obtain multiple informative row subsets and constructs a decision tree for each subset using TreeED. It then applies an Expectation-Maximization-based algorithm to jointly estimate tree reliability and optimize the consensus ED prediction. Experiments demonstrate that our methods are accurate, explainable, and robust, achieving an average F1-score improvement of 16.1% over the best baseline.

cs.CL↗
arXiv2026

Interpretable Column Annotation with LLM-Symbolized Decision Process Materialization

Column annotation (CA), including column type annotation (CTA) and column property annotation (CPA), aims to identify the meanings of table columns and the semantic relationships among them. Recent CA methods usually use various neural models to learn column representations and directly map them to label categories, thereby (1) sacrificing model interpretability and adaptivity, and (2) overlooking rich label semantics and ultimately limiting accuracy. To address these limitations, we propose SymCA, an LLM-empowered interpretable CA framework that materializes column annotation as a global-to-local symbolic decision process. SymCA consists of two components: (1) global skeleton induction, which constructs a semantic skeleton over the label space, and (2) local substrate evolution, which evolves predictive substrates within the skeleton. Specifically, to exploit label semantics while preserving an interpretable decision process, the global skeleton induction module leverages LLMs to generate candidate hypernym-inspired tree-structured semantic skeletons and employs a Minimum Bayes Risk (MBR)-based consensus strategy to select a robust skeleton against generation variance. Since different internal nodes require different evidence to distinguish among their child nodes, the local substrate evolution module materializes each internal node as an executable and evolvable predictive substrate. Over multiple evolution rounds, each substrate trains an interpretable random forest classifier with the current operator set, leverages the LLM to propose node-specific operator modifications, and uses an exploration-exploitation strategy to prioritize promising substrates. Extensive experiments demonstrate that SymCA is accurate, robust, and interpretable, outperforming the strongest baselines by an average of 6.42% in Micro-F1 and 11.03% in Macro-F1.

cs.CL↗
arXiv2026

Three-Body Alignment: Aligning Chess Agent with Human Reasoning through Reranked Rationale

As reasoning agents become increasingly complex, aligning their underlying reasoning and decision-making processes with human conceptual models is a challenge for AI security and safety. When modelling expert knowledge, understanding how to characterise and integrate insights from agents with fundamentally different reasoning architectures is necessary for safe and predictable deployment. We investigate this alignment through a \emph{three-body alignment} in chess, analysing the semantic divergence between rationales produced by human experts (Grandmasters), engine-assisted human commentators (who rationalise the outputs of efficiently updatable neural networks, or NNUEs), and Large Language Models (LLMs). Our contributions include: (1) A novel multisource rationale dataset, constructed using an agentic data engineering pipeline to transform unstructured expert commentary into structured, queryable data for alignment evaluation. (2) An empirical analysis of the semantic embedding space. Using t-SNE visualisation, we demonstrate that these sources form distinct clusters, confirming significant heterogeneity and reflecting fundamentally different conceptual approaches to the same environment. (3) An experiment demonstrating that reranking mechanisms can improve human alignment, while quantifying the explicit trade-off with tactical performance, offering a pathway for more interpretable agent decision-making. (4) The preliminary development of an enriched chess narrative dataset structure, designed to lay the groundwork for future evaluations of text rationale similarity and to address the limitations of standard dense retrieval. (5) Finally, we open-source our chess rationales dataset\footnote{Hugging Face: https://huggingface.co/datasets/jaymarichua/trichess} to support developing novel techniques that integrate diverse expert knowledge into human-aligned intelligent agents.

cs.GT↗