Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Thursday 3 September

Computation and Language · Cross-listed from cs.LG

Conformity Breaks Conformal Prediction

Yibo Hu and Hanyu Su

A conformal certificate can be valid when an LLM answers alone and invalid when the same LLM sees peers that unanimously assert a wrong answer. The question is unchanged; the model's score for the correct answer changes. We call this a score-mechanism shift: clean calibration certifies how the model scores answers alone, but not how it scores them under peer pressure. We show that this shift silently breaks conformal prediction in multi-agent LLM systems. Across open-weight models and multiple-choice QA tasks, coverage falls from a calibrated 90% to 74% under unanimous-wrong peers at the standard alpha = 0.10 operating point. The average hides a sharper failure: by targeting the low-confidence items the certificate still covers, an attacker nearly halves coverage on that subgroup, from 87% to 47%, while the monitored average remains much higher. The failure also reaches the decision layer: a system that should escalate when uncertain can instead become confident enough to act on the attacker's wrong answer. Standard conformal fixes do not solve the problem, because the question distribution has not changed; the model's scoring behavior has.
Read the full abstract

Computation and Language

GRACE: Graph-Grounded Reflective Agent Copilot Engine for Expert-in-the-Loop Knowledge Expansion

John Seon Keun Yi, Joshua R. Minot and Dokyun Lee

Large language models deployed in high-stakes settings frequently generate plausible but ungrounded claims. Standard retrieval-augmented generation (RAG) pipelines offer limited remedy, since they retrieve isolated passages without tracking cross-document evidence relationships or quantifying uncertainty. We introduce GRACE (Graph-grounded Reflective Agent Copilot Engine), a framework that deconstructs LLM responses into atomic claims and grounds them against trusted knowledge priors within a weighted bipartite graph. Edge weights encode the closeness of each claim to the priors, enabling weighted centrality analysis that classifies claims as Grounded, Refuted, or Boundary. Such classification identifies not just hallucinations but also novel or contested claims at the frontier of the model's knowledge. To efficiently allocate human or agent resources, we formulate a Return on Attention (RoA) objective that defers a claim to expert review only when its priority-weighted uncertainty exceeds the cost of verification. Claims verified by experts are promoted to new evidence anchors, closing a validator-LLM evolutionary loop that expands the knowledge base across iterations. We evaluate GRACE across multiple language models and on datasets spanning both general and domain-specific knowledge. Our results show that our knowledge base serves as a reliable foundation for retrieval that outperforms RAG baselines, and that the RoA framework efficiently selects valuable boundary knowledge for expert verification. These findings demonstrate that graph-structured representations combined with expert-in-the-loop verification can mitigate hallucination at the system level rather than at the generation level. Code available at https://github.com/johnsk95/grace_code
Read the full abstract

Computation and Language

What Attention Recalls and Recurrence Controls in Hybrid Language Models

Kirill Afendulev, Alexey Dontsov, Elena Tutubalina et al. (+1 more)

Hybrid language models combine attention with a fixed-size recurrent state, but the role of each channel remains unclear. We introduce two cache-level interventions. Split-prefill keeps only the KV cache or only the recurrent state from a prefilled context, then generates an answer. State-swap pairs the KV cache from one context with the recurrent state from another in a single forward pass. On Qwen3.5 and Falcon-H1, the two channels split sharply by function. Exact retrieval survives only through attention (64-98% of full accuracy) and collapses to zero through recurrence. Output language and persona reverse the pattern: both survive recurrence (70-80% and 3-5x) while KV-only drops to ~1% language accuracy. State-swap confirms this causally: the answer takes its value from the KV side and its language from the recurrent side. Recurrent-only generation also accepts words that were never in the context but share meaning or parts with seen items. Attention provides a lookup over what was said; the recurrent state shapes how the model says it next.
Read the full abstract

All authors Kirill Afendulev, Alexey Dontsov, Elena Tutubalina, Anton Korznikov

Computation and Language

A Systematic Evaluation of Cross-Lingual Consistency Enhancement Methods in Multilingual Language Models

Jirui Qi, Mingyang Wang, Hinrich Schütze et al. (+2 more)

Multilingual language models often produce inconsistent answers to semantically equivalent questions across languages, motivating methods to improve cross-lingual consistency (CLC). However, existing methods are typically evaluated using different models, tasks, and protocols, leaving their relative strengths unclear. In this work, we present a unified evaluation of representative CLC-enhancement methods for question answering, spanning inference-time interventions and post-training approaches across three model families and three closed-form benchmarks. The results show that post-training methods are generally more reliable, with direct distribution alignment consistently improving CLC across all model-dataset combinations, while other methods are more sensitive to answer format and the breadth of language coverage. Notably, cross-domain transfer is limited unless source and target tasks share similar output formats. We further investigate whether CLC enhancement hurts models' ability to respond differently *when needed*, that is, when asked culture-dependent questions. Across two benchmarks of culturally diverse question answering, we find no systematic degradation in controlled closed-form evaluation, whereas open-ended generation reveals occasional accuracy reductions, particularly for non-English responses. Our work highlights the need to evaluate CLC enhancement for both cross-domain robustness and culturally appropriate variation, informing future work in post-training and benchmark development.
Read the full abstract

All authors Jirui Qi, Mingyang Wang, Hinrich Schütze, Raquel Fernández, Arianna Bisazza

Computation and Language

The Anatomy of an ASR Hallucination

Hamees Sayed, Apoorv Singh, Kumar Aman et al. (+1 more)

ASR systems sometimes produce fluent text that is unrelated to the speech they receive. We view these hallucinations as one possible consequence of a broader grounding failure, in which the transcript is no longer adequately guided by the audio. To understand where this failure becomes possible, we study two independently trained Conformer-Large recognizers - one CTC and one RNN-T - under environmental degradation and speaker-background shift. In both models, the final encoder stage emerges as a critical boundary: bypassing the final block causes divergence on nearly every utterance, whereas bypassing middle blocks has little effect. At this same stage, the representations become more compact, text becomes readable by the trained decoder, and grapheme information becomes explicit. Importantly, the intervention produces garbled or repetitive output rather than fluent fabrication. Our result therefore identifies a mechanistic precondition for hallucination - the failure to produce adequately grounded output - not the complete origin of naturally occurring hallucinations. Together, the results reveal a consistent terminal-stage dependency for grounded recognition across two decoder families and multiple distribution shifts.
Read the full abstract

All authors Hamees Sayed, Apoorv Singh, Kumar Aman, Akshat Mandloi

Computation and Language

Evaluation of Phonetic Encoding Algorithms on Transcription Datasets

Can Özbey, Emre Kaplan and Berkin Deniz Kahya

In this work, a novel evaluation scheme built on a generalized variant of the Rand Index measure, namely, the Hüllermeier-Rifqi Index, is proposed in order to assess how well phonetic encoding algorithms conform to word-based transcriptions in IPA (International Phonetic Alphabet) notation. For this objective, the discordance score is obtained by calculating the absolute difference between the pairwise similarity values of ground-truth transcriptions and those of corresponding phonetic encodings, which are computed using normalized edit distance as a permutation dependent string metric. The resulting score is subsequently adjusted with respect to that of a random string generator incorporating the same alphabet as the encoder under consideration. A wide range of phonetic encoders were evaluated as such on multi-lingual transcription datasets along with their recall capabilities based on the collision rate. The validity of the proposed scheme is further supported by its applicability in measuring the orthographic transparency of a language when the writing system is viewed as an inherent phonetic representation.
Read the full abstract

Computation and Language

You Really Didn't Get That? Benchmarking Social Pragmatic Inference for Indirect and Playful Chinese Online Comments

Shiwei Hong, Junjie Ma, Emma Jiren Wang et al. (+5 more)

Chinese online comments often convey social meaning through indirect and playful language that is hard to interpret without context. Existing evaluations largely organize items around predefined phenomena or controlled pragmatic categories, leaving open whether models can distinguish plausible readings of what a naturally occurring comment is doing in a particular exchange. We introduce a benchmark for evaluating whether LLMs can recover such situated pragmatic meanings. From more than 200,000 public Chinese social media interaction records, we construct 4,735 human-validated diagnostic items, each pairing a target comment with reconstructed preceding context and plausible misreadings. We evaluate eight LLMs as both question writers and solvers in a cross-writer setting. The task is challenging: the strongest model achieves 81.42% leave-writer-out accuracy. Across all eight models, the mean leave-writer-out accuracy is 68.70% while human accuracy was 90.8%. Case analysis shows that models often recognize broad irony or playfulness while misidentifying the mechanism or interactional move.
Read the full abstract

All authors Shiwei Hong, Junjie Ma, Emma Jiren Wang, Ethan Z. Rong, Siying Hu, Haichang Li, Ziying Wang, Zhicong Lu

Computation and Language

VERGE: Verification-Enhanced Refinement for Grounded Extraction of Early-Onset Colorectal Cancer Symptoms in Clinical Notes

Nikkie Hooman, Monarch Nigam, Amy E. Hughes et al. (+2 more)

Early-onset colorectal cancer is increasing among younger adults, yet red-flag symptoms in this age group have no evidence-based guidelines for follow-up testing, and structured encounter data do not capture the detail needed to support early detection and inform follow-up, including symptom duration, context, and fam- ily history, an established colorectal-cancer risk factor. This study aimed to develop and evaluate an automated method for extracting six red-flag symptoms and family-history risk status from free-text clinical notes. We developed VERGE, an agentic workflow in which an initial label and evidence are proposed using retrieval-augmented generation, then passed through a bounded verification- refinement cycle that checks textual grounding and clinical validity, corrects and rechecks a claim until resolved or a limit is reached, and escalates unresolved claims for human review. VERGE was evaluated on 4,033 clinician-labeled note-finding pairs against a single-agent baseline, a rule-based clinical language-processing baseline, and an alternative underlying language model. Compared with the single-agent baseline, VERGE reduced false positive find- ings, improving precision from 0.764 to 0.849 and MCC from 0.681 to 0.730, a balanced gain across the precision-recall trade-off, and resolved most flagged errors autonomously, with human review required for only 1.5 percent of claims. These results indicate that a bounded, verification-based workflow can reduce unnecessary positive findings without sacrificing the ability to detect true ones. This approach offers a path toward more reliable and trustworthy clinical language-processing tools to support colorectal cancer risk assessment in younger patients.
Read the full abstract

All authors Nikkie Hooman, Monarch Nigam, Amy E. Hughes, Rasmi G. Nair, Mehak Gupta

Computation and Language · Cross-listed from cs.SD

Knowing When Not to Answer: Pseudo-Ensembles for Abstention in Music Audio-Language Models

Aanya Maheshwari and Vatsal Raina

Music audio-language models are evaluated almost entirely by accuracy on multiple-choice questions. This protocol forces the model to commit to an option, so a lucky guess looks the same as real musical understanding. What is missing is a way to tell when the model does not know the answer, so that it can abstain instead of guessing. The usual solution, an ensemble of independently trained models, is far too expensive here, which leaves the entropy of a single predictive distribution as the only available confidence signal. We instead build pseudo-ensembles from one pretrained model by perturbing its input in ways that cannot change the correct answer, then averaging the resulting distributions over the options. Our main construction simply shuffles the order in which the candidate answers are presented; we also study ensembles built from corrupted audio and from swapped option labels. A pseudo-ensemble gives several predictive distributions per question, so it supports the full family of ensemble-based uncertainty measures (entropy of the expected distribution, expected entropy, and their difference, the mutual information) rather than entropy alone. Evaluating TinyMU on MuChoMusic, we find that averaging over four option orderings raises accuracy from 55.7% to 59.2%, and that the resulting uncertainty measures rank the model's errors better than the single-pass entropy baseline, reducing the area under the error retention curve from 0.293 to 0.261. All of this costs a few extra forward passes and no retraining, which makes abstention practical for compact music audio-language models.
Read the full abstract

Computation and Language

Adapting from Downturns: Prediction of Long-Term Conversational-Skill Development in Mental-Health Crisis Counselors

Vivian Nguyen, Lillian Lee, Elizabeth A. Olson et al. (+1 more)

How do people learn to become better conversationalists? This question is especially important in the context of mental-health counseling, where conversational skills are essential, yet volunteer counselors often have limited access to supervision and structured feedback. Understanding how counselors develop their ability to steer conversations toward positive outcomes – and identifying early which counselors are (not) on track to improve – can help prioritize support for the counselors who need it most. In this work, we introduce the task of predicting, early in a conversationalist's career, whether they will eventually improve at steering conversations toward positive outcomes, and demonstrate the feasibility of this task in the case of volunteer mental-health crisis counselors. Our central insight is that people may struggle with particular kinds of moments in a conversation, and that what is especially revealing of their likelihood of future improvement is how they learn to handle those moments over time. We operationalize this insight by designing a method that identifies the types of moments a counselor initially struggles with, captures how they adapt their response when they re-encounter similar moments in subsequent conversations, and learns which early adaptations predict improvement months or even years later. While this future-prediction task is challenging, our counselor-adaptation approach yields better results than baselines that learn directly from the conversation transcript.
Read the full abstract

All authors Vivian Nguyen, Lillian Lee, Elizabeth A. Olson, Cristian Danescu-Niculescu-Mizil

Computation and Language · Cross-listed from cs.LG

SharedSAE: One Feature Dictionary Across Language Models

Daniil Ognev, Célian Vasson, Lijie Hu et al. (+2 more)

Sparse autoencoders (SAEs) are widely used to interpret language model activations, but SAE training and latent labelling are typically repeated for every model. Here, we show that a single shared SAE can replace a collection of dedicated per-model SAEs. Our method, SharedSAE, combines a shared dictionary with model-specific encoder-decoder pairs. Unlike the closest prior method, which discards activation magnitudes and requires all models at inference, SharedSAE instead normalizes only selection scores, preserving magnitudes, and uses model dropout for single-model inference. We train SharedSAE on four 1B-scale base language models spanning distinct families and tokenizers. Despite sharing its latents across models, SharedSAE retains 96.6% of dedicated SAEs' mean explained variance; its latent activations exhibit cross-model correlations 1.8 times as high as separate SAEs aligned post-hoc, and its latent descriptions transfer across models. After the dictionary is frozen, new models can be efficiently adapted to it, achieving near-dedicated-SAE reconstruction quality while reusing the shared latent descriptions.
Read the full abstract

All authors Daniil Ognev, Célian Vasson, Lijie Hu, Kentaro Inui, Benjamin Heinzerling

Computation and Language · Cross-listed from cs.AI

A Removal Based Approach to Improve LLM Faithfulness at Test-Time

Qinglan Luo, S M A Nahian, John Guttag et al. (+2 more)

Large language models (LLMs) are increasingly used for consequential decisions, making their explanations an important tool for auditing model behavior. Unfortunately, these explanations can be unfaithful, failing to reflect the actual reasoning underlying the model's decisions. We consider a setting in which an LLM provides both an answer and an explanation in response to a question. We identify two distinct dimensions of unfaithful explanations: incompleteness, meaning that the explanation omits factors that influence the answer, and unsoundness, meaning that the explanation cites factors that did not influence the model's answer. Existing approaches to improving LLM faithfulness include training-time methods, which require access to model weights and extensive computational resources, and test-time methods that largely focus on addressing unsoundness. We introduce a test-time approach that directly targets incompleteness. We remove from the input the concepts not credited in the model's explanation and re-query the model on the reduced input. This eliminates unmentioned influences while preserving the influence of mentioned concepts. Across two datasets, multiple model families, and two independent faithfulness metrics, our approach improves explanation faithfulness compared to both standard prompting and prompting to encourage faithfulness. Our method is model-agnostic and can be applied at inference time without modifying model parameters, providing a flexible mechanism for reducing hidden influences and improving the reliability and safety of LLM-assisted decision making.
Read the full abstract

All authors Qinglan Luo, S M A Nahian, John Guttag, S. Mazdak Abulnaga, Katie Matton

Computation and Language

MedProb: Probing Internal Representations of Vision-Language Models for Medical Question Answering

Erfan Nourbakhsh, Ke Yang and Anthony Rios

Medical visual question answering (Med-VQA) is often assumed to require medical fine-tuning, large models, or complex multi-agent pipelines. We revisit this assumption with MedProb, a lightweight probing framework that predicts multiple-choice Med-VQA answers from frozen VLM representations without free-text generation. Across PATH-VQA, SLAKE, and VQA-RAD, MedProb recovers substantially more answer-relevant signal than prompting and performs stronger than medical VLMs and agentic systems. Probing also reduces the apparent gap between small and large models compared to prompting, suggesting that smaller VLMs contain more recoverable Med-VQA signal than generation-based evaluation reveals. Across 14 matched general-purpose and medical VLM pairs, medical adaptation does not consistently improve this linear decodability. Finally, free-text generation exhibits an answer-position bias of up to 10 percentage points, whereas MedProb also has positional bias, however, it is impacted differently than prompting. Our main results target the multiple-choice/multiclass Med-VQA setting; we additionally show the probe can be extended to open-ended generation via a rejection-sampling scoring procedure.
Read the full abstract

Computation and Language · Cross-listed from cs.MA

Abstraction Agent

Boning Li and Longbo Huang

Information abstraction, which groups strategically similar private states into a tractable number of buckets, is essential for scaling game-solving algorithms to large imperfect-information games. Constructing effective abstractions, however, has traditionally required domain-specific evaluators such as hand-strength calculators or equity estimators, which demand expert knowledge and engineering effort and are unavailable for most less-studied games. We propose the Abstraction Agent, a zero-shot pipeline that uses a large language model (LLM) to discover continuous strategic features from a natural-language game description, score private states on these features, and cluster them into abstraction buckets, without any game-specific evaluator, training data, or game-tree traversal during abstraction construction. The pipeline runs in four phases: feature discovery with calibration anchors, batched private-state scoring, correlation-based feature selection, and $k$-means clustering. The resulting abstractions reduce lifted-strategy exploitability by up to 62% relative to an expected-hand-strength baseline on heads-up no-limit Texas hold'em (HUNL) turn endgames, and beat a scalar rank baseline at every granularity on ROVER Trials, an original game absent from any pretraining corpus. Beyond these quantitative benchmarks, the pipeline transfers with unchanged prompts to four-card Pot-Limit Omaha, HUNL preflop and flop, and Riichi Mahjong, where the discovered features track each game's recognized strategic concepts. This is structured knowledge elicitation: converting implicit strategic knowledge in LLM parameters into explicit numerical features for downstream algorithmic computation. The code is available at https://github.com/lbn187/AbstractionAgent.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Harbor Adapters and Harbor-Index: Infrastructure and a Curated Meta-Dataset for Large-Scale Agentic Evaluation

Lin Shi, Haowei Lin, Zixuan Zhu et al. (+122 more)

Evaluating agents on the growing number of agentic benchmarks is challenging because they often require complex environments and agent integrations. We introduce Harbor Adapters, a unified evaluation infrastructure for agentic benchmarks. Our work makes three contributions. First, we develop benchmark adapters that port more than 80 benchmarks to evaluate arbitrary agents, and validate them through rigorous code review and parity experiments. Second, we conduct a large-scale evaluation of 8 models spanning capability tiers across 54 benchmarks; every model is run with Terminus-2 and with one of 3 native harnesses. This enables a broader analysis of agent capabilities and failure modes than was previously possible. Third, we introduce Harbor-Index, a curated set of 82 difficult, diverse, and high-quality tasks spanning 29 benchmarks, refined from the adapted suite through difficulty filtering, AI and human audit, and an audit-and-fix loop. Harbor-Index preserves the challenge and breadth of large-scale agentic evaluations while being affordable to run; no evaluated model-harness configuration exceeds 30% pass rate, and the strongest (GPT-5.5 with Codex) reaches 28.0%. We release the adapters, evaluation results, in-depth analysis, and Harbor-Index as open-source artifacts to support more reliable and comprehensive evaluation of language-model agents.
Read the full abstract

All authors Lin Shi, Haowei Lin, Zixuan Zhu, Xiaoyue Zhou, Xiang Li, Xiangning Lin, Yaxuan Deng, Han Xu, Yuangang Li, Shanda Li, Zizhao Chen, Hanwen Xing, Harsh Raj, Bo Chen, Quan Shi, Steven Dillmann, Yipeng Gao, Puneesh Khanna, Ruofan Lu, Chao Beyond Zhou, Michael Yang, Robert Zhang, Siyuan Chai, Jiayu Chang, Yizhao Chen, Xiaokun Chen, Yiwei Dai, Wenting Yang, Hange Liu, Minghao Liu, Zihan Wang, Adnan El Assadi, Benedikt Stroebl, E. Kelly Buchanan, Han Meng, Junwei He, Longxuan Yu, Radin Shayanfar, Yukyung Lee, Zhikang Dong, Allen G Hart, Anjiang Wei, Anurag Kashyap, Arpandeep Khatua, Audrey Jixin Zheng, Chengrui Ma, David Heineman, Dubing Chen, Hai-Anh Trinh, Haishuo Fang, Hefan Zhang, Hui Shen, Issa Sugiura, Jiankai Sun, Jiechao Gao, Junhong Lin, Junnan Li, Kai Yang, Lei Hsiung, Maoyu Wang, Mengze Tang, Nabil Omi, Negin Raoof, Nicholas Edwards, Octavia Guo, Orfeas Menis Mastromichalakis, Pengliang Ji, Przemysław Hejman, Qi Qi, Qunshu Lin, Richard Zhuang, Rui Yang, Ruichen Zheng, Ryan Marten, Shaghayegh Fazliani, Shizheng Hou, Sicong Jiang, Sijie Li, Boqin Yuan, Michael Glass, Song Bian, Terry Yue Zhuo, Tianqing Wu, Tom Tang, Wanjia Zhao, Weihao Xuan, Wenhua Liang, Xian Liu, Xin Lan, Xuan Zhang, Xuandong Zhao, Yanchuan Tang, Yifan Jiang, Yijiang Li, Yitong Guan, Yizhi Li, Yonghui Liu, Yuheng Tang, Yujun Mao, Yunfei Zhao, Yuxin Wang, Yuxuan Tang, Zhenheng Tang, Zhifei Li, Ziruo Wang, Ziyu She, Kaiyuan Liu, Iheb Chaabane, Yuxin Tang, Xiangyi Li, Satya Sai Srinath Namburi GNVV, Xinyue Zheng, Andy Konwinski, Boxuan Li, Leon Liangyu Chen, Alex Dimakis, Nicholas Carlini, Soroush Vosoughi, Sanmi Koyejo, Di He, Etash Guha, Benjamin Feuer, Mike Merrill, Ludwig Schmidt, Alex Shaw

Computation and Language

Evidence Integration in Large Language Models

Sebastien Kawada and Manolis Kellis

Despite increasing reliance on LLMs that reason with external evidence supplied by tools, retrieval-augmented generation, other agents, and users, how LLMs integrate such evidence into decisions they have already begun to form remains largely unclear. We present a distributional theory in which evidence shifts the receiver's distribution of initial answers, driven by a receiver prior weight and a candidate evidence tilt, leading to three predictions. First, candidates more probable to the receiver are more persuasive. Second, receivers more readily integrate characteristic errors of their own than foreign errors from different sources. Third, identical evidence can improve weaker models and harm stronger ones. We confirm these over ten million trials, twelve LLMs from four families, and eight domains, four of them scientific discovery tasks in the physical and life sciences: quantum mechanics, physics, genetics, and molecular biology. The law also yields a receiver-relative reliability frontier: receiver-congruent errors depress performance more steeply than random errors of the same rate. LLMs also integrate candidates even after internally verifying their invalidity (93-100% with propositional constraints; up to 99.4% on held-out physical and life-sciences reasoning), demonstrating evidence integration is a receiver-specific control policy over existing distributions, determined by receiver properties rather than scalar trust in the evidence source. Causal interventions show candidate integration is implemented late in the network, as a structured sequence of steps admitting external candidate answers, promoting them, and transporting them into the answer state. Representations of verification are decodable but have little causal impact on answers. A J-lens decomposition shows the state underlying verbalized verification is fully dissociable from that underlying candidate integration.
Read the full abstract

Computation and Language

Memory as transformation: LETHE, a self-referential gan-inspired architecture

Francesco Vitucci, Anthony Di Furia and Francesco Scagliola

LETHE (Latent-parameter Evolution with Temporal Hierarchical quasi-Equilibrium) is a self-referential sonic-oblivion system implemented in SuperCollider. It adopts the formal vocabulary of Generative Adversarial Networks in a closed configuration without external datasets or supervision after initialization. Audio is processed by a 3 x 3 mixing matrix built around two delay lines; its nine coefficients and two delay times evolve through the interaction of a five-feature linear discriminator and a random-perturbation optimizer analogous to single-sample REINFORCE. The discriminator compares current energy behavior with an archive of the initial state and guides parameter updates. Circular, fixed, and live sources can be mixed independently. Across fixed and circular sessions with an ablation control, the active generator is necessary for parametric evolution ($\Delta c_{22}=0.000$ in all 15 ablation sessions). Situated in the tradition of self-referential electroacoustic music, LETHE delegates the sonic outcome to an adaptive closed loop whose parametric space is defined by the composer.
Read the full abstract

Computation and Language · Cross-listed from cs.MA

EvoHarnessBench: Can Your Agents Keep Pace with an Evolving Harness?

Zixuan Ke, Vaidehi Patil, Haizhou Shi et al. (+9 more)

Modern LLM-based agents operate through a harness of tools, reusable skills, and specialist agents that shapes what they observe and what they can do. In practice, this harness continually evolves as new capabilities are added. We introduce EvoHarnessBench, a benchmark for evaluating agents under controlled harness evolution across three axes (tools, skills, and agents). Unlike existing continual-learning benchmarks for agents, which typically place non-stationarity (i.e., what changes over time) in the task stream while keeping the harness fixed, EvoHarnessBench places non-stationarity in the externally supplied harness itself. It contains 17 multi-stage harness streams constructed deterministically from verifier-based benchmarks, comprising 802 tasks, 520 tools, 42 skills, and 62 agents. We evaluate two complementary settings corresponding to the central challenges of harness evolution: deployment evaluation, which isolates retention of previously accessible competence as the harness expands, and self-evolving adaptation evaluation, which tests whether accumulated experience remains useful as new capabilities are introduced. Our results reveal three persistent gaps. First, harness expansion alone can degrade performance on previously solved tasks, producing harness-induced forgetting. Second, gains from self-evolving adaptation remain inconsistent across stages of harness evolution, capability axes, and environments. Third, retention and adaptation can pull in different directions: preserving earlier competence does not necessarily improve adaptation to newly introduced capabilities, and vice versa. These results establish harness evolution as a distinct challenge for building agents that can keep pace with an evolving harness while preserving previously effective behavior.
Read the full abstract

All authors Zixuan Ke, Vaidehi Patil, Haizhou Shi, Yang Li, Ye Liu, Sarath Shekkizhar, Anurag Koul, Jiayu Wang, Xuan Phi Nguyen, Semih Yavuz, Mohit Bansal, Shafiq Joty

Computation and Language

Compile by Training: Turning Natural-Language Specifications into Local Neural Functions

Yuntian Deng, Pengyu Nie and Stuart Shieber

Many recurring text functions are easy to describe but difficult to implement with rules, while calling a large remote model for every input introduces repeated cost, latency, and dependency on a provider. We present compile by training, which turns a natural-language specification into a reusable neural function. At compile time, teacher models generate task-specific examples that are used to train a small adapter for a compact interpreter. The resulting function runs without the teachers and can be stored, versioned, and composed like ordinary software. On FuzzyBench-Hard, a subset on which the Program-as-Weights fast compiler produced no exact matches, compile by training reaches 83.6% semantic accuracy. This higher accuracy comes with a higher compile-time cost: roughly a minute rather than seconds for the fast compiler. We deploy the compiler in a public interactive service and demonstrate compiled functions in a multi-site website helper, a language-controlled 3D avatar, and a bidirectional English-Claudish translator.
Read the full abstract

Computation and Language

ESPO: Error-Structured Prompt Optimization via Diagnose, Diversify, and Stabilize

Lihao Liu, Peng Tang, Kunwar Yashraj Singh et al. (+1 more)

Evolutionary prompt optimizers such as GEPA suffer from prompt bloat: each iteration appends rules and caveats, producing prompts up to 3$\times$ longer yet no more accurate. We trace this to three deficiencies - incomplete error observation, limited search diversity, and unreliable selection - and propose ESPO (Error-Structured Prompt Optimization), which decomposes prompt optimization into three phases: Diagnose clusters all training errors into structural patterns in one round; Propose generates candidates via four complementary strategies with independent biases; Select applies bootstrap stability selection. On seven public NLP benchmarks - Tweet, MMLU, GSM8K, HotpotQA, ScoNe, HoVer, and PUPA - ESPO improves average accuracy by $+$3.76 pp over the state-of-the-art (74.67% vs 70.91% for GEPA), matching or exceeding GEPA on every dataset while producing prompts 47% shorter (1,004 vs 1,878 chars) and faster at inference. Cross-model experiments across four additional student models (Gemma 3 12B, Mistral 14B, Qwen3 32B, Claude Haiku 4.5) show ESPO yields the best average accuracy on every model tested, with the largest gap on Qwen3 GSM8K (15.00% $\to$ 91.40%). A generalization bound (Appendix) grounds each phase in a corresponding term of the test-time gap, and the ablation confirms a key prediction: adding diversity without bootstrap selection actually hurts performance ($-$1.20%).
Read the full abstract

All authors Lihao Liu, Peng Tang, Kunwar Yashraj Singh, Shabnam Ghadar

Computation and Language

Legibility is Not Interpretability: Comparing Judged and Actual Importance in Chain-Of-Thought Reasoning

Kevin Du, Alexander Hoyle, Laura Ruis et al. (+1 more)

Reasoning traces from chain-of-thought models appear to offer a legible window into how a model arrives at its answer. A growing body of work treats them as such, using LLM judges to diagnose errors, evaluate faithfulness, and provide step-level supervision via process reward models and generative critics. These practices rely on the text of a reasoning step carrying information about its functional role. But does the text actually encode information about which reasoning steps matter? We operationalize the importance of a reasoning step as its advantage: the change in expected reward, e.g., producing the correct final answer, from including that step, estimated via Monte Carlo rollouts. Basing ground truth on these estimates, we evaluate whether LLM judges can identify high-advantage steps and find that sufficiently capable LLMs can outperform a prevalence baseline but fall well short of a noise ceiling. Fine-tuning a model as a step-level critic yields strong improvement for incorrect responses but remains distant from ceiling for correct responses, suggesting that step importance is only partially recoverable from the text of the reasoning trace. Our findings contribute to a growing body of chain-of-thought faithfulness work that cautions against treating the legibility of reasoning traces as interpretability, especially with implications for process reward modeling.
Read the full abstract

All authors Kevin Du, Alexander Hoyle, Laura Ruis, Acyr Locatelli

Computation and Language

Knowledge Acquisition During Pre-training? Large Language Models Learn Better With Auxiliary Views

Joseph Lee, Yidi Huang, Dokyoon Kim et al. (+2 more)

Gaps remain in our understanding of how large language models (LLMs) acquire knowledge during pre-training. We posit that auxiliary views, reformulations of knowledge, are causally helpful for learning. We design controlled experiments to isolate this. First, we confirm that repetition is necessary for acquisition and clarify that paraphrasing helps only at smaller batch sizes. Second, holding the token budget fixed, allocating tokens from document repetition to auxiliary views improves learning, counterintuitively, even for factual recall. Third, the effectiveness of auxiliary views is not contingent on the strength of the teacher model that generates them. Fourth, we identify forms of knowledge, contextual and foundational, that aid learning in the presence of prior knowledge gaps. Finally, we examine how these effects manifest mechanistically via layer-wise biases and compression. Together, our findings suggest that auxiliary representations of knowledge, which arise naturally in large pre-training corpora, are a key factor in the success of pre-training and offer a plausible explanation for why data diversity matters.
Read the full abstract

All authors Joseph Lee, Yidi Huang, Dokyoon Kim, Shu Yang, Li Shen

Computation and Language

Last Translation Benchmark

Vilém Zouhar, Niyati Bafna, Mukund Choudhary et al. (+257 more)

For scientific progress, we need benchmarks that test the limits of state-of-the-art models, and evaluation methods that inform us about failure cases. As models get stronger, standard benchmarks for machine translation are approaching saturation. Further, automatic translation metrics are unreliable, opaque, and vulnerable to reward-hacking. Even gold human evaluation is not problem-free, because it often lacks reproducibility, objectivity, and scalability. Overall, this prevents us from tracking progress in the field and identifying pathways for improvement. We introduce the Last Translation Benchmark, a collection of human-authored and peer-reviewed examples (texts, images, audio, videos) that break leading machine translation models. We also present a new evaluation approach: each example comes with handcrafted verification rules describing concrete failure cases on that example, therefore allowing reliable and actionable future evaluation. The Last Translation Benchmark is a live dataset that accepts ongoing contributions. The latest version is LTBv1, containing accepted contributions prior to September 1st 2026, with future releases planned as new data is continuously collected.
Read the full abstract

All authors Vilém Zouhar, Niyati Bafna, Mukund Choudhary, Maike Züfle, Sara Rajaee, Pinzhen Chen, Jannis Vamvas, Sara Papi, Ona de Gibert, Bhavitvya Malik, Eliya Habba, Orfeas Menis Mastromichalakis, Patrícia Schmidtová, Michelle Wastl, Sheriff Issaka, Leshem Choshen, Stella Biderman, Antonis Anastasopoulos, Jan Niehues, Rico Sennrich, Mrinmaya Sachan, Ondřej Bojar, Kenton Murray, Jörg Tiedemann, Alham Fikri Aji, Philipp Koehn, Christof Monz, Alexandra Birch, Sowmya Vajjala, Chalamalasetti Kranti, Cristina España-Bonet, Nobin Sarwar, David Kaczér, Sourajit Saha, Jonathan Tonglet, Shunta Asano, Malik Marmonier, Daban Q. Jaff, Vaisakhi Mishra, Hend Al-Khalifa, Gabriele Sarti, Nils Rehlinger, Juan Daniel Cuervo Villa, Dominik Macháček, Saugata Purkayastha, Jagannathan Ramanujam, Shubhashis Roy Dipta, Aviral Nigam, Shuaib Shuaib Yusuf, Heejin Do, Jonathan Yahav, Johannes-Rudolf David, Maria Carmen Staiano, Zuzana Nadova, Fred Philippy, Ron Keinan, Maria Lymperaiou, Silvia Casola, Fabian Retkowski, Andrés Jerez, Hanna Yukhy- menko, Sangwon Ryu, Avantica Vempati, Sukannya Purkayastha, Adrian Cosma, Erivan Inan, Vitalii Babenko, Wafa Aissa, Valentin Scourneau, Fatima Haouari, Venkata Prasanth Kumar Gummadi, Mehdi Jafarzadeh, Manon Reusens, Kaiser Sun, Lukas Edman, Shaomu Tan, Giuseppe Gallipoli, Pawan Sasanka Ammanamanchi, Manar Ali, Mohammad Sadegh Gholizadeh, Dipankar Srirag, Marek Šuppa, Javier García Gilabert, Ruta Binkyte, Ana-Maria Bucur, Sabry E. Farrag, Youssef Saber, Yihong Liu, Theresia Veronika Rampisela, Christian Hoang, Nicoleta Cojocaru, Jan Kocoń, Jean Maillard, Xiaochuang Yuan, Sina Ahmadi, Daryna Dementieva, Philipp Mondorf, Kaustubh Dhole, Valmik Nahata, Roman Wixinger, Amir Hossein Yari, Shenbin Qian, Manuel Tuor, Fida Mohammad Thoker, Sergey Troshin, Lance Calvin Lim Gamboa, Amir Arsalan Rezapour, Kätriin Kukk, Koel Dutta Chowdhury, Shaswati Saha, Seth Aycock, Bo Chen, Linh Vu, Vatsal Venkatkrishna, Shayan Bali, Arafat Ahsan, Luan Thanh Nguyen, Hassan Soliman, Ngoc Quynh Tram Do, Azmine Toushik Wasi, R. Damanhuri, Marius Huber, Kazuki Egashira, Jimson Paulo Layacan, David Africa, Vladislav Poritski, Mike Zhang, Deep Shah, Abdulaziz Nura Kani, Luis Frentzen Salim, Paul Gavrikov, Bello Umar Bello, Ayush Sunil Munot, Anumit Garg, Yolanda Xavier, Qiaoyuan Zheng, Kawsar Ahmed, Debanshu Das, Zimu Wang, Gengyu Rao, Kamile Dementaviciute, Farhan Farsi, L D M S Sai Teja, Dawei Zhu, Yi Fan, Wei Liu, Marco Gaido, Elias Herranen, Sankalan Pal Chowdhury, Karen Sanchez, Guy Kaplan, Farzad Shami, Ashok Urlana, Amir Hossein Kargaran, Sofie Goethals, Priyaranjan Pattnayak, Oksana Volchek, Marii Ojastu, Hongbin Na, Emilian Radoi, Chenyi Zhao, Carlos Hinojosa, Andrei Niculae, Andrea Gregor de Varda, Zaid Alyafeai, Tomasz Limisiewicz, Reem Alzahrani, Pouya Sadeghi, Nehal Kathrotia, Mateusz Lango, Enzo Doyen, Alex Flückiger, Ulysses Sekai Tully Carr, Samuel Simko, Ritwik Tiwari, Rishit Dagli, Isaac R Caswell, Bowen Yi, Aicha Chorana, Selja Keränen, Sadiksha Chitrakar, Muhammad Ravi Shulthan Habibi, Joy Olusanya, Bishal Shrestha, Zhengxiang Wang, Vivek Harsha Lakkamaneni, Sophia Conrad, Panayiotis Panayiotou, Nazia Tasnim, Marta Punsola Munárriz, Marko Culjak, Luis Lara, Jenny Chim, Jannatul Nayem, Fidel Rodríguez Velásquez, Eran Yahav, Blanka Kövér, Beatrice Savoldi, Anmol Goel, Aishik Mandal, Tosin Adewumi, Raoyuan Zhao, Mykola Haltiuk, Antonia Karamolegkou, Yuxing Lu, Thura Aung, Naser Almousa, Tommaso Cerruti, Raia Abu Ahmad, Beni Egressy, Alireza Pakniat, Stéphane J.P.S. Thunus, Rachel Bawden, Lena Libon, Samridhya Biswas, Prakhar Gupta, Nusrat Jahan Lia, Nguyen Tai, Natchapon Jongwiriyanurak, Minh Ngoc Do, Ivan Barać, Dzmitry Kuzmin, Badal Nyalang, Antoine Taroni, Andy Catruna, Rushikesh Zawar, Roland Aydin, Pavel Stepachev, Ilai Yaron Levy, Andreas Simons, Rayyan Merchant, Ziyi Yang, Samuel Frontull, Kenneth Enevoldsen, Harris Abdul Majid, Tim Graf, Tatiana Bielakova, Sharifa Djurabaeva, Shaoxiong Ji, Jirui Qi, Ayla Rigouts Terryn, Yurii Paniv, Xiyan Fu, Sunisth Kumar, Shree Harsha Bokkahalli Satish, Papa Abdou Karim Karou Diallo, Mengyu Ye, Maximilian Vieweg, Matija Akrap, Kristýna Onderková, Joseph Attieh, Ivan Yuri De Leon, Ibrahim Baroud, Esrael Teferi Tensay, Elisabeth Fittschen, David Dukić, Benoît Sagot, Jingwei Ni, Yu Fan, Juri Opitz

Computation and Language · Cross-listed from cs.AI

Rethinking On-Policy Distillation of Large Language Models II: One Training Example

Zixuan Fu, Bingxiang He, Yuxin Zuo et al. (+10 more)

On-policy distillation (OPD) combines student-generated rollouts with dense token-level supervision from a teacher. Existing work has mainly studied its algorithmic behavior, leaving the role of training data unclear. We examine this role at the data-minimal limit by training on a single query. One-shot OPD keeps improving for hundreds of steps and recovers most of full-data OPD's gain across task domains and model families. We explain this result through the states visited during training and the rate at which the student aligns with the teacher. We measure state coverage, the fraction of the states full-data OPD visits that a query set's rollouts reach. A single query already reaches \(71.5\%\), most of it within the first 100 steps. Adding semantically distinct queries raises coverage and validation accuracy together, until 16 queries reach \(98.9\%\) and match full-data training. Yet alignment slows at a similar pace whether OPD trains on one query or the whole dataset, and even a fixed set of states takes hundreds of steps to absorb. OPD is therefore data-overfed but algorithm-starved. Its rollouts quickly expose broad supervision, while the student absorbs that supervision increasingly slowly. The state-coverage result extends to multi-teacher OPD, where 16 semantically diverse queries per domain match full-data MOPD. As a further stress test, content-light templates and off-domain WildChat queries also approach the real-query baseline. Task content and induced state coverage can therefore come apart. We hope these findings direct future work toward the step efficiency of OPD, and prompt a re-examination of the data and the mechanisms behind its recent successes in frontier post-training.
Read the full abstract

All authors Zixuan Fu, Bingxiang He, Yuxin Zuo, Haohuan Huang, Jinqian Zhang, Ruhang Xiao, Cheng Qian, Qinyu Luo, Huan-ang Gao, Yudong Wang, Zhiyuan Liu, Ning Ding, Chaojun Xiao

Computation and Language · Cross-listed from cs.AI

Terminal-Universe: Turning Agent Trajectories into Scalable Terminal Environments

Jie Wu, Zhenru Zhang, Beichen Zhang et al. (+11 more)

As terminal-based code agents become prevalent, agent trajectories have accumulated at scale, while realistic, executable environments remain scarce. However, environments are what agent post-training actually requires: each can be re-queried into many verifiable tasks and provides execution feedback, whereas a trajectory is a single frozen demonstration. Rather than generating environments from scratch, we observe that the tool-execution history in existing trajectories exposes the structure and contents of the environments in which they ran, making it possible to reconstruct those environments from the trajectories themselves. Thus, we introduce Terminal-Universe, a framework which turns each trajectory into a reusable environment and explores it for synthesizing new tasks and continued interactions. Specifically, Terminal-Universe replays the file operations recorded in a trajectory to restore each file before the agent modified it, yielding a partial workspace; a completion agent then supplies the missing files and dependencies. On this recovered workspace, we both reconstruct the original intent task and synthesize entirely new ones. Besides, we also scale the tasks along two complementary axes: breadth and depth. For breadth, we mine directional dependency relations between related environments and synthesize cross-workspace queries spanning multiple codebases, as developers routinely do in real-world development. For depth, we extend the initial single-turn query into a multi-round session that captures iterative user feedback and requirement refinement via a user agent. Applied to public terminal agent trajectories, Terminal-Universe produces 37.3k task-sufficient environments. Supervised fine-tuning of Qwen3.5-27B on this corpus improves single-round performance on Terminal-Bench 2.1 by 11.9 points and multi-round performance on EvoCode-Bench v2 MT@4 by 13.8 points.
Read the full abstract

All authors Jie Wu, Zhenru Zhang, Beichen Zhang, Xuwu Wang, Yuhui Su, Mouxiang Chen, Peng Wang, Zhihai Wang, Que Shen, Hao Zhou, An Yang, Fei Huang, Yujiu Yang, Dayiheng Liu

Computation and Language

Sequential Beats Joint: On the Interplay between On-Policy Distillation and RLVR

Boyan Li, Bingsen Chen, Chenghao Yang et al. (+3 more)

Reinforcement learning with verifiable rewards (RLVR) and on-policy distillation (OPD) have emerged as two dominant methods for post-training reasoning LLMs. Prior work uses OPD's dense token-level supervision to complement the sparse RL reward, fusing the two signals within a single step: either as a weighted-additive combination or a teacher-modulated rescaling of the RL advantage. In this paper, we show that a simple two-stage scheme, OPD-then-RL, consistently outperforms pure OPD, pure RLVR, and all such joint baselines across logic and math reasoning benchmarks. Beyond the empirical results, we further provide a systematic understanding of this through pass@$k$ behavior, learning dynamics, and parameter updates, yielding a consistent explanation: OPD expands the student's coverage of teacher-supported solutions and RL sharpens within that support, while jointly optimizing the two signals causes them to interfere. To provide a practical recipe, we find that the OPD validation score is the key signal for when to switch to RL, and that OPD is a better cold start for RL than SFT. Together, our results establish OPD-then-RL as a simple yet strong way to combine the two methods, turning two entangled signals into complementary stages.
Read the full abstract

All authors Boyan Li, Bingsen Chen, Chenghao Yang, Ping Nie, Chen Zhao, Xi Ye

Computation and Language · Cross-listed from cs.CV

CORE: Improving Compositional Reasoning in MLLM Embedding via Reranker Distillation

Tingyu Song, Mingxin Li, Yanzhao Zhang et al. (+5 more)

MLLM-based embedding models remain limited in compositional retrieval, often failing to distinguish scenes containing the same concepts but different attribute-object bindings. Yet the same backbone can resolve such distinctions when used as a cross-attentive reranker, motivating us to distill its compositional judgments into the embedding model. We propose CORE, which synthesizes candidate lists spanning five compositional matching levels and introduces a Rank-KL objective that trains the embedding model to reproduce the reranker's fine-grained ranking. We further introduce a graded evaluation protocol and compare contrastive learning, pairwise CoSENT, and listwise Rank-KL under the same data and tuning budget. Our comparison shows that both CoSENT and Rank-KL use the multi-level supervision more effectively than contrastive learning, with Rank-KL achieving the strongest overall performance. Across three compositional reasoning benchmarks (COLA, SUGARCREPE++, NEGBENCH), CORE-RERANKER-8B achieves an 82.7% total average, outperforming Jina-Reranker by 10.7 points, while CORE-EMBED-8B achieves the best total average (0.666) among all evaluated embedding models. The improvements transfer to the MCMR benchmark without sacrificing retrieval performance on COCO and Flickr30K.
Read the full abstract

All authors Tingyu Song, Mingxin Li, Yanzhao Zhang, Dingkun Long, Chu Liu, Pengjun Xie, Yilun Zhao, Shu Wu

Computation and Language · Cross-listed from cs.SE

When Models Edit Too Much: On the Fidelity of Minimal Code Edits

Tongyao Zhu, Wei Hern Lim and Min-Yen Kan

Large language models (LLMs) are increasingly used to edit existing code, but correctness alone is not enough: useful repairs should also be minimal, reviewable, and faithful to the original implementation. We study over-editing, the tendency of a model to rewrite code beyond what is required to fix a bug. We construct an evaluation framework from 400 BigCodeBench problems by injecting controlled AST-level corruptions into reference solutions, giving each repair task a known minimal patch. Across frontier LLMs, over-editing is widespread even among strong models like GPT-5.5: high Pass@1 can coexist with unnecessarily large edits and added cognitive complexity. A preservation instruction substantially reduces this behavior, lowering average excess Levenshtein distance from 0.195 to 0.131, reducing added cognitive complexity by 26.6%, and increasing Pass@1 by 2.3 points. However, these gains do not simply follow from a larger reasoning budget or larger models. We next ask whether minimal editing can be learned directly during post-training. We observe that supervised fine-tuning overfits to seen corruption patterns, whereas reinforcement learning gives the best out-of-domain edit-fidelity and performance-retention trade-off. These results position edit fidelity as a distinct axis of code-repair quality and show that it can be measured and learned.
Read the full abstract

Computation and Language

Translation as a Decision Space: A Multi-Agent Perspective on Low-Resource Dialect Generation

Hasan Alkhder, Mohammad Abboush, Igor Tchappi et al. (+2 more)

Neural machine translation (NMT) systems typically produce a single output per input, obscuring the alternative decision trajectories implicitly available within multilingual decoding. This opacity becomes particularly problematic in low-resource dialect settings, where multiple linguistically valid realizations may differ in lexical authenticity, register, and structural stability. We propose reframing translation as a structured decision space explored by autonomous translation agents. Instead of analyzing a single output, we model distinct translation pathways as agents operating over a shared multilingual backbone. Inter-agent divergence is treated not as error but as an interpretable behavioral signal. We conduct an empirical study on Turkish–Syrian Arabic translation using three agents: (1) zero-shot direct translation, (2) dialect-stabilized translation via lightweight fine-tuning, and (3) pivot translation through English. Evaluation is performed on 5,000 dialogue sentences, while stabilization is trained on 5,000 additional Turkish–Syrian sentence pairs drawn from television dialogue and MADAR-Turk resources. Rather than optimizing for conventional performance metrics, we quantify structured behavioral displacement using dialect marker frequency, lexical proximity to standardized Arabic, and structural variance. Lightweight stabilization nearly doubles dialect marker usage, increasing it from 0.2266 to 0.4988, while significantly reducing structural instability. Pivot mediation introduces normalization pressure and measurable compression effects, whereas zero-shot translation exhibits the highest decision variance. We argue that translation divergence across agents reveals latent decision flexibility within multilingual models and we provide a principled interpretability framework for low-resource dialect generation.
Read the full abstract

All authors Hasan Alkhder, Mohammad Abboush, Igor Tchappi, Ahmet Zengin, Amro Najjar

Computation and Language · Cross-listed from cs.IR

The Dice Roll Method: A Standardized Protocol for Repeated-Query Auditing of Large Language Model Brand Recommendations

Dmitrij Żatuchin

Background: Researchers increasingly use repeated identical prompts to audit stochastic variation in large language model (LLM) brand recommendations, yet no standardized protocol exists for setting iteration counts, selecting stability metrics, or establishing reliability thresholds. Objective: We formalize the Dice Roll Method as a reusable protocol for repeated-query auditing of LLM brand recommendations, grounded in a generative model of temperature-scaled nucleus sampling. Methods: Total response variance is decomposed into sampling, prompt-phrasing, run-to-run, and model-version components. The stack: a negative-binomial mixed model with iterations as repeated measures; Cliff's delta as the distribution-free effect size; dependence-preserving bootstrap; simulation-based power; a generalizability-theory decomposition; drift diagnostics on pinned snapshots. We reanalyse five brand-recommendation auditing studies: approximately 190,000 observations, 270+ brands, 6 languages, iteration counts 5 to 40. Results: The recomputed D-study supports no fixed tiers. Across 250 distinct questions (S5), G is 0.96 at five runs and 1 to 2 runs reach 0.80, also within each model. For five versions of one question (S1 Valentine's subset, Gemini) it is 0.27 (95% CI 0.00 to 0.52), which fixes no run count. Runs follow from a pilot: n = 4 x residual / cell variance. Simulated single-cell power for a large Cliff's delta is 0.15 to 0.19 at five runs and 0.8 at 15 to 25. The four metric families (count, set, embedding, PASOR) are complementary. A pre-registered external validation on three independent corpora reproduces the D-study prediction in 37 of 39 cells with no failures; fixed tiers do not transfer. Conclusion: The protocol gives repeated-query auditing of LLM brand recommendations a statistically principled footing under the conditional, non-Gaussian structure of real autoregressive generation.
Read the full abstract