arXiv:2604.00860v5 Announce Type: replace
Abstract: Reinforcement learning has become a central post-training paradigm for improving LLM and agent capabilities. Yet existing RL post-training methods share a common blind spot: they construct local learning signals from sampled trajectories, rewards, or feedback-conditioned targets, then update the policy without explicitly verifying whether the resulting policy outperforms its predecessor. Optimizing these local signals does not necessarily produce a better policy, while finite sampling, generation stochasticity and feedback noise can further widen this gap. We argue that the missing ingredient is policy improvement feedback: the ability to measure progress across policy iterations. We introduce Policy Improvement Reinforcement Learning (PIRL), which formulates inter-iteration performance gain as an explicit objective structurally aligned with final task performance. Building on PIRL, we propose Policy Improvement Policy Optimization (PIPO), a plug-in closed-loop framework that verifies the previous update against a sliding-window historical performance anchor. PIPO uses this improvement feedback to modulate the local learning signal of the base policy optimization algorithm, reinforcing updates associated with measured progress and suppressing those associated with performance drops. We provide theoretical evidence that PIPO locally aligns policy updates with the PIRL improvement objective. Experiments on mathematical reasoning, code, tool-use, and self-distillation settings show that PIPO yields consistent gains across PPO, group-relative, and self-distillation policy optimization families.
Science Journals
arXiv:2508.20735v2 Announce Type: replace
Abstract: We study parallel algorithms for the minimisation and equivalence checking of Deterministic Finite Automata (DFAs). Regarding DFA minimisation, we implement four different massively parallel algorithms on Graphics Processing Units~(GPUs). Our results confirm the expectations that the algorithm with the theoretically best time complexity is not practically suitable to run on GPUs due to the large amount of resources needed. We empirically verify that parallel partition refinement algorithms from the literature perform better in practice, even though their time complexity is worse. Furthermore, we introduce a novel algorithm based on partition refinement with an extra parallel partial transitive closure step and show that on specific benchmarks it has better run-time complexity and performs better in practice.
In addition, we address checking the language equivalence and inclusion of two DFAs. We consider the Hopcroft-Karp algorithm, and explain how a variant of it can be parallelised for GPUs. We note that these problems can be encoded for the GPU-accelerated model checker \GPUexplore, allowing the use its lockless hash table and fine-grained parallel work distribution mechanism.
arXiv:2601.21738v2 Announce Type: replace
Abstract: Evaluation of Image Quality Assessment (IQA) models has long been dominated by global correlation metrics, such as Pearson Linear Correlation Coefficient (PLCC) and Spearman Rank-Order Correlation Coefficient (SRCC). While widely adopted, these metrics reduce performance to a single scalar, failing to capture how ranking consistency varies across the local quality spectrum. For example, two IQA models may achieve identical SRCC values, yet one ranks high-quality images (related to high Mean Opinion Score, MOS) more reliably, while the other better discriminates image pairs with small quality/MOS differences (related to $|\Delta$MOS$|$). Such complementary behaviors are invisible under global metrics. Moreover, SRCC and PLCC are sensitive to test-sample quality distributions, yielding unstable comparisons across test sets. To address these limitations, we propose \textbf{Granularity-Modulated Correlation (GMC)}, which provides a structured, fine-grained analysis of IQA performance. GMC includes: (1) a \textbf{Granularity Modulator} that applies Gaussian-weighted correlations conditioned on absolute MOS values and pairwise MOS differences ($|\Delta$MOS$|$) to examine local performance variations, and (2) a \textbf{Distribution Regulator} that regularizes correlations to mitigate biases from non-uniform quality distributions. The resulting \textbf{correlation surface} maps correlation values as a joint function of MOS and $|\Delta$MOS$|$, providing a 3D representation of IQA performance. Experiments on standard benchmarks show that GMC reveals performance characteristics invisible to scalar metrics, offering a more informative and reliable paradigm for analyzing, comparing, and deploying IQA models. Codes are available at https://github.com/Dniaaa/GMC.
arXiv:2607.06522v1 Announce Type: new
Abstract: Vision-language models (VLMs) struggle to generalize in interactive physical reasoning, particularly under unseen tasks and environments. Two key failure modes are prominent: hallucinated chain-of-thought (CoT) reasoning that contradicts physical reality, and misalignment between the model's reasoning and actions. We present VAORA (Visual Action Outcome Reasoning Alignment), a novel reward design that directly addresses both issues. VAORA introduces two complementary rewards: Visual Alignment Reward, which anchors VLM reasoning to the visual context independent of the agent action itself, and Visual-Action Alignment Reward, which grounds reasoning in the visual outcome induced by the model's action. Together, these rewards suppress hallucinated CoT and reduce the gap between reasoning and behavior. To improve training stability, we further employ smooth, dense rewards by estimating success probabilities using a pre-trained in-domain expert agent. Experiments on PHYRE and Virtual Tool support our performances across novel-task and unseen-environment settings, confirming that grounded and generalizable physical intelligence can be induced through VAORA.
arXiv:2607.06171v1 Announce Type: cross
Abstract: Let $H$ be an inclusion-free hypergraph on $n$ vertices. A weight assignment $w:[n]\to[d]$ is isolating if there is a unique edge $e$ whose weight $w(e) = \sum_{i \in e} w(i)$ is minimum. We show that the number of isolating weight assignments is at least $$ n\sum_{j=0}^{d-1} j^{n-1}, $$ a bound which is attained with equality by the hypergraph consisting of the $n$ singleton edges. This proves the conjecture stated in Faber & Harris (2018).
We also prove the bound for a more general class of edge-weight objectives, including arbitrary edge offsets.
arXiv:2503.09460v2 Announce Type: replace
Abstract: The draft candidate European Cybersecurity Certification Scheme for Cloud Services (EUCS) defines security controls that must be associated with measurable metrics to assess compliance. This association process is currently manual, time-consuming, and prone to inconsistencies. In this paper, we propose an automated approach based on Sentence Transformers to associate cloud security controls with quantifiable metrics by leveraging semantic similarity between their textual descriptions. We evaluate our method on a dataset of 70 controls derived from the EUCS framework. The proposed approach outperforms a FastText-based baseline, achieving a conditional Normalized Discounted Cumulative Gain at rank 10 score of 0.640 (+0.146) and improving the standard nDCG@10 score from 0.275 to 0.504. These results demonstrate that contextual embedding models significantly enhance both the likelihood of retrieving relevant metrics and their ranking quality. Our findings highlight the potential of transformer-based methods to support automated, scalable, and more reliable compliance processes in cloud cybersecurity certification.
arXiv:2603.04024v2 Announce Type: replace
Abstract: Ambiguous 3D medical image segmentation often involves boundaries where different expert delineations are non-identical yet clinically plausible. Modeling such inter-observer variability requires a careful balance between diversity and anatomical fidelity: deterministic models preserve coherent volumetric structures but collapse expert disagreement into a single mask, while stochastic generative models can produce diverse samples but may introduce disconnected components or slice-to-slice inconsistency when generating full 3D masks from unstructured noise. We propose Volumetric Directional Diffusion (VDD), a prior-anchored diffusion framework that shifts stochastic generation from full-mask synthesis to residual boundary exploration. VDD uses a coarse consensus prediction as an anatomical anchor and learns a directional diffusion process to generate plausible boundary variations around ambiguous regions while preserving stable volumetric topology. Experiments on three multi-rater datasets, including LIDC-IDRI, KiTS21, and ISBI 2015, show that VDD improves uncertainty distribution alignment while maintaining competitive segmentation accuracy and 3D structural consistency. These results suggest that prior-anchored residual diffusion can model clinically plausible expert disagreement without sacrificing anatomical fidelity.
arXiv:2604.01800v3 Announce Type: replace
Abstract: In incompressible flow simulations, non-uniform grids efficiently capture localized flow features; however, their spatially varying resolutions severely exacerbate computational complexity. The pressure Poisson equation (PPE) formulated on these grids yields highly complex linear systems, forming the primary computational bottleneck in fractional step method. To address this, we develop an extended hybrid framework tailored for non-uniform Cartesian grids, integrating deep learning with classical iterative solvers to accelerate PPE solutions. Specifically, the framework employs a deep operator network with a U-Net-based branch network. To effectively capture spatially varying resolutions, we propose a multi-level distance vector map construction strategy that computes discrete grid-spacing information corresponding to each hierarchical level of the U-Net. This grid-spacing information is explicitly fused into feature maps prior to convolution operations. Empowered by this grid-spacing-aware architecture, the framework seamlessly extends to simulate flows interacting with solid structures using a decoupled immersed boundary projection method. By training exclusively on fabricated linear systems rather than conventional flow-dependent datasets, the model generalizes effortlessly across diverse immersed obstacle geometries with fixed network weights. Benchmark results demonstrate that the framework significantly outperforms standalone preconditioned conjugate gradient methods and its standard convolution counterpart, underscoring its exceptional potential for real-world computational fluid dynamics applications.
arXiv:2604.03148v2 Announce Type: replace
Abstract: The absorption of light is one of the most fundamental processes in condensed-matter physics and optics. Here we investigate under which conditions laser light is absorbed by a crystalline material as an electromagnetic wave with delocalized properties or rather as photons that cause discrete, localized, nanometer-sized consequences. We excite the first-order phase transition of vanadium dioxide with laser pulses of sufficient frequency to overcome the band gap but with insufficient pulse energy to overcome the latent heat. According to Maxwell's equations and Bloch theory, no transition should occur, because nowhere in the material is enough energy. Nevertheless, we observe with ultrafast electron diffraction a disordered crystal geometry with nanometer-sized spots of switched material that grow and diminish with time. The amount of localized spots matches approximately to the number of photons in the absorbed laser wave. Two optical experiments substantiate this phenomenon, and simulations reproduce all measurements results. We discuss whether crystals defects, temperature, or a genuine wavefunction collapse can explain the discovered phenomenon. Practically, the reported absorption mechanism enables local consequences at substantially higher energy than average and provides insight into symmetry breaks and non-thermal fluctuations within complex materials.
arXiv:2604.03316v3 Announce Type: replace
Abstract: Attention sinks are defined as tokens that attract disproportionate attention. While these have been studied in single modality transformers, their cross-modal impact in Large Vision-Language Models (LVLM) remains largely unexplored: are they redundant artifacts or essential global priors? This paper first categorizes visual sinks into two distinct categories: ViT-emerged sinks (V-sinks), which propagate from the vision encoder, and LLM-emerged sinks (L-sinks), which arise within deep LLM layers. Based on the new definition, our analysis reveals a fundamental performance trade-off: while sinks effectively encode global scene-level priors, their dominance can suppress the fine-grained visual evidence required for local perception. Furthermore, we identify specific functional layers where modulating these sinks most significantly impacts downstream performance. To leverage these insights, we propose Layer-wise Sink Gating (LSG), a lightweight, plug-and-play module that dynamically scales the attention contributions of V-sink and the rest visual tokens. LSG is trained via standard next-token prediction, requiring no task-specific supervision while keeping the LVLM backbone frozen. In most layers, LSG yields improvements on representative multimodal benchmarks, effectively balancing global reasoning and precise local evidence.
arXiv:2604.00786v2 Announce Type: replace
Abstract: The L infinity star discrepancy is a measure for how uniformly a point set is distributed in a given space. Point sets of low star discrepancy are used as designs of experiments, as initial designs for Bayesian optimization algorithms, for quasi-Monte Carlo integration methods, and many other applications. Recent work has shown that classical constructions such as Sobol', Halton, or Hammersley sequences can be outperformed by large margins when considering point sets of fixed sizes rather than their convergence behavior. These results, highly relevant to the aforementioned applications, raise the question of how much existing constructions can be improved through size-specific optimization. In this work, we study this question for the so-called Kronecker construction. Focusing on the 3-dimensional setting, we show that optimizing the two configurable parameters of its construction yields point sets outperforming the state-of-the-art value for sets of at least 500 points. Using the algorithm configuration technique irace, we then derive parameters that yield new state-of-the-art discrepancy values for whole ranges of set sizes.
arXiv:2604.04703v2 Announce Type: replace
Abstract: Large language models (LLMs) are bringing richer dialogue and social behavior into games, but they also expose a control problem that existing game interfaces do not directly address: how should LLM characters participate in live multiplayer interaction while remaining executable in the shared game world, socially coherent with other active characters, and steerable by players when needed? We frame this problem as bounded autonomy, a control architecture for live multiplayer games that organizes LLM character control around three interfaces: agent-agent interaction, agent-world action execution, and player-agent steering. We instantiate bounded autonomy with probabilistic reply-chain decay, an embedding-based action grounding pipeline with fallback, and whisper, a lightweight soft-steering technique that lets players influence a character's next move without fully overriding autonomy. We deploy this architecture in a live multiplayer social game and study its behavior through analyses of interaction stability, grounding quality, whisper intervention success, and formative interviews. Our results show how bounded autonomy makes LLM character interaction workable in practice, frames controllability as a distinct runtime control problem for LLM characters in live multiplayer games, and provides a concrete exemplar for future games built around this interaction paradigm.
arXiv:2604.07607v2 Announce Type: replace
Abstract: Robot learning increasingly depends on large and diverse data, yet robot data collection remains expensive and difficult to scale. Egocentric human data offer a promising alternative by capturing rich manipulation behavior across everyday environments. However, existing human datasets are often limited in scope, difficult to extend, and fragmented across institutions. We introduce EgoVerse, a collaborative platform for human data-driven robot learning that unifies data collection, processing, and access under a shared framework, enabling contributions from individual researchers, academic labs, and industry partners. The current release includes 1,362 hours (80k episodes) of human demonstrations spanning 1,965 tasks, 240 scenes, and 2,087 unique demonstrators, with standardized formats, manipulation-relevant annotations, and tooling for downstream learning. Beyond the dataset, we conduct a large-scale study of human-to-robot transfer with experiments replicated across multiple labs, tasks, and robot embodiments under shared protocols. We find that policy performance generally improves with increased human data, but that effective scaling depends on alignment between human data and robot learning objectives. Together, the dataset, platform, and study establish a foundation for reproducible progress in human data-driven robot learning. Videos and additional information can be found at https://egoverse.ai/
arXiv:2604.13030v2 Announce Type: replace
Abstract: While diffusion models dominate the field of visual generation, they are computationally inefficient, applying a uniform computational effort regardless of different complexity. In contrast, autoregressive (AR) models are inherently complexity-aware, as evidenced by their variable likelihoods, but are often hindered by lossy discrete tokenization and error accumulation. In this work, we introduce Generative Refinement Networks (GRN), a next-generation visual synthesis paradigm that addresses these issues. At its core, GRN addresses the discrete tokenization bottleneck through a theoretically near-lossless Hierarchical Binary Quantization (HBQ), achieving a reconstruction quality comparable to continuous counterparts. Built upon HBQ's latent space, GRN fundamentally upgrades AR generation with a global refinement mechanism that progressively perfects and corrects artworks -- like a human artist painting. Besides, GRN integrates an entropy-guided sampling strategy, enabling complexity-aware, adaptive-step generation without compromising visual quality. On the ImageNet benchmark, GRN establishes new records in image reconstruction (0.56 rFID) and class-conditional image generation (1.81 gFID). We also scale GRN to more challenging text-to-image and text-to-video generation, delivering superior performance on an equivalent scale. We release all models and code to foster further research on GRN.
arXiv:2604.14104v2 Announce Type: replace
Abstract: We present observations from two consecutive TRACERS-2 orbits through the northern low-altitude cusp. During the first crossing, TRACERS-2 observed reversed cusp ion dispersion and sunward convection, consistent with magnetopause reconnection tailward of the cusp during this northward IMF interval. Simultaneous THEMIS-D observations at the equatorial magnetopause show heated magnetosheath plasma captured on closed field lines, with similar particle spectra as in in the low-altitude cusp, indicating that reconnection indeed occurred tailward of the cusp and in both hemispheres. When TRACERS-2 traversed the northern cusp again, 95 minutes later, the IMF was dominated by a negative BX component. Despite the different IMF conditions, TRACERS-2 recorded nearly the same cusp signatures as before, i.e., reversed ion dispersion and sunward convection. The observations indicate that tailward-of-cusp reconnection can occur for both northward and BX-dominated IMF and that these distinct IMF geometries can produce remarkably similar plasma and field signatures in the low-altitude cusp.
arXiv:2604.20394v2 Announce Type: replace
Abstract: We establish nearly optimal upper and lower bounds for approximating decision tree splits in data streams. For regression with labels in the range $\{0,1,\ldots,M\}$, we give a one-pass algorithm using $\tilde{O}(M^2/\epsilon)$ space that outputs a split within additive $\epsilon$ error of the optimal split, improving upon the two-pass algorithm of Pham et al. (ISIT 2025). Furthermore, we provide a matching one-pass lower bound showing that $\Omega(M^2/\epsilon)$ space is indeed necessary.
For classification, we also obtain a one-pass algorithm using $\tilde{O}(1/\epsilon)$ space for approximating the optimal Gini split, improving upon the previous $\tilde{O}(1/\epsilon^2)$-space algorithm. We complement these results with matching space lower bounds: $\Omega(1/\epsilon)$ for Gini impurity and $\Omega(1/\epsilon)$ for misclassification (which matches the upper bound obtained by sampling).
Our algorithms exploit the Lipschitz property of the loss functions and use reservoir sampling along with Count--Min sketches with range queries. Our lower bounds follow from careful reductions from the INDEX problem.
arXiv:2604.20557v2 Announce Type: replace
Abstract: Shared Control methods often use impedance control to track target poses in a robotic manipulator. The guidance behavior of such controllers is shaped by the used stiffness gains, which can be varying over time to achieve an adaptive guiding. When multiple target poses are tracked at the same time with varying importance, the corresponding output wrenches have to be arbitrated with weightings changing over time. In this work, we study the stabilization of both variable stiffness in impedance control as well as the arbitration of different controllers through a scaled addition of their output wrenches, reformulating both into a holistic framework. We identify passivity violations in the closed loop system and provide methods to passivate the system. The resulting approach can be used to stabilize standard impedance controllers, allowing for the development of novel and flexible shared control methods. We do not constrain the design of stiffness matrices or arbitration factors; both can be matrix-valued including off-diagonal elements and change arbitrarily over time. The proposed methods are furthermore validated in simulation as well as in real robot experiments on different systems, proving their effectiveness and showcasing different behaviors which can be utilized depending on the requirements of the shared control approach.
arXiv:2607.06092v1 Announce Type: new
Abstract: G.\ D.\ Baker formulated a forcing method to interpret integer optimisation problem into $2$-adic linear regression, and proved the NP-hardness of $2$-adic linear regression. We generalise the forcing method to a wider class of $p$-adic optimisation for the case where $p$ is not necessarily $2$, and prove the NP-hardness of $p$-adic linear regression, the NP-hardness of $2$-adic dynamic neural network by S.\ Albeverio, A.\ Khrennikov, and B.\ Tirrozi, and the NP-hardness of a partial generalisation of the $p$-adic optimisation problem associated to van der Put neural network by G.\ L.\ R.\ N'guessan.
arXiv:2607.05418v1 Announce Type: cross
Abstract: In this paper, we study the group randomness of 0-1 real sequences derived from a binary linear code by investigating the spectral behaviour of a suitable normalization of the Gram matrix of a $p \times n$ random matrix whose rows are uniformly drawn from those 0-1 real sequences, where $y=p/n \in (0,1)$ is fixed. We show that as $n \to \infty$, its empirical spectral distribution converges to the Marchenko-Pastur law at a rate at least of the order $n^{-1/4}$ with high probability, and the fluctuation of its largest eigenvalue is asymptotically Gaussian with mean $p+1+y$ and variance $4y$, provided that the dual distance of the code is at least 5.
arXiv:2607.05459v1 Announce Type: cross
Abstract: Tractatus Quanticum (arXiv:2512.06034 [quant-ph]) is described by its authors as 'a re-editing, which takes quantum mechanics into account, of Wittgenstein's famous Tractatus.' The original Tractatus appeared with an introduction by Bertrand Russell. For Tractatus Quanticum, that role fell to us. This is the result.
arXiv:2507.22758v2 Announce Type: replace
Abstract: Recent advancements in financial problem-solving have leveraged LLMs and agent-based systems, with a primary focus on trading and financial modeling. However, credit assessment remains an underexplored challenge, traditionally dependent on rule-based methods and statistical models. In this paper, we introduce MASCA, an LLM-driven multi-agent system designed to enhance credit evaluation by mirroring real-world decision-making processes. The framework employs a layered architecture where specialized LLM-based agents collaboratively tackle sub-tasks. Additionally, we integrate contrastive learning for risk and reward assessment to optimize decision-making. We further present a signaling game theory perspective on hierarchical multi-agent systems, offering theoretical insights into their structure and interactions. Our paper also includes a detailed bias analysis in credit assessment, addressing fairness concerns. Experimental results demonstrate that MASCA outperforms baseline approaches, highlighting the effectiveness of hierarchical LLM-based multi-agent systems in financial applications, particularly in credit scoring.
arXiv:2605.11692v2 Announce Type: replace
Abstract: We study a triplet majority-rule opinion-dynamics model with collective reversal on quenched networks. Interactions occur on local triplets composed of one agent and two of its neighbors, while collective reversal acts only on unanimous triplets. This rule separates local conformity from external perturbations that disrupt local agreement. We show that quenched network topology shifts the order--disorder critical point away from the well-mixed value. For Barab\'asi--Albert, Erd\H{o}s--R\'enyi, random regular, and Watts--Strogatz networks, the estimated critical exponents remain close to the mean-field values, suggesting mean-field-like universal behavior within the system sizes studied. The strongest shift of the critical point occurs for Watts--Strogatz networks, where clustering and local correlations make the ordered phase less stable. A rewiring analysis of Watts--Strogatz networks further shows that the ordered phase becomes more stable as the network becomes more random. These results indicate that quenched topology primarily controls the location of the transition, while the collective-reversal mechanism largely preserves mean-field-like critical behavior.
arXiv:2605.17437v2 Announce Type: replace
Abstract: Context. Metamorphic Testing addresses the test-oracle problem in scientific computing, but classical Mutation Score operates on syntactic AST mutations and misses domain semantics.
Objective. We propose the Semantic Mutation Score (SMS), built on five domain-semantic operators (Conservation Erosion, Operator Substitution, Hyperparameter, Trajectory Flip, Structural Injection). SMS degenerates almost everywhere to MS in a characterised limit, so any SMS-based conclusion remains consistent with prior mutation-testing literature in the classical regime.
Method. A 12-PUT x 5-MP design over four single-output float-to-float classes (numeric, probabilistic, surrogate, machine-learning) is paired with a three-layer attribution classifier separating true semantic faults from tolerance, OOD, statistical, and artefact categories. A same-source / cross-source ablation under an identical prompt isolates the LLM-source-diversity contribution. LLM-generated mutants are compared against a default-configuration cosmic-ray syntactic pool at the AST-normalised level.
Results. The pre-registered large-effect threshold for Cliff's delta is not met under the point-estimate criterion; the observed effect lies in the medium-effect range. Cross-source pooling under an identical prompt does not appreciably shift delta, indicating that LLM identity is not the lever within this design. AST-level overlap between LLM-generated and default cosmic-ray syntactic mutants is small; the Hyperparameter, Structural Injection, and Trajectory Flip classes are unreachable under default first-order syntactic configurations.
Conclusion. SMS is a backward-compatible adequacy metric for domain-semantic metamorphic-relation sets in scientific computing. The first-order unreachability evidence is independent of the effect-size question.
arXiv:2605.18419v2 Announce Type: replace
Abstract: Vision-language models (VLMs) can couple visual perception with open-ended clinical reasoning, making them attractive for computational histopathology. However, fine-tuning billions of parameters on scarce, expert-annotated pathology data is prohibitive, while in-context learning (ICL), which conditions the VLM on demonstrative image-text pairs without parameter updates, suffers from high sensitivity to which examples are selected and how the query is phrased, producing unreliable diagnostics. Existing selection strategies rely on query-dependent nearest-neighbour retrieval that ignores global data structure, require costly parameter updates, or disregard the joint vision-text embedding geometry of VLMs. We propose GAUC, a training-free coreset selection method operating directly in the pre-trained multimodal embedding space. GAUC jointly optimises three objectives: (1) a Maximum Mean Discrepancy term enforcing distributional fidelity between coreset and full dataset, (2) an Effective Mutual Information Difference regulariser bounding performance degradation under prompt paraphrases by exploiting the VLM's joint vision-text alignment, and (3) a predictive-uncertainty (entropy) penalty suppressing ambivalent, hallucination-prone outputs. On CRC-100K and MHIST across multiple open-source VLM architectures, GAUC \emph{matches} the accuracy of the strongest ICL selection and dataset-distillation baselines while substantially improving calibration, prompt robustness, and hallucination rates, all without a single gradient update.
arXiv:2607.05545v1 Announce Type: new
Abstract: LLM conformity is often used to describe cases where a model changes a correct answer toward a peer or group response. We show that most of this apparent conformity survives even after the peer is removed. The reason is a confound: standard conformity prompts mix two cues at once, the presence of a speaker and the repeated wrong answer itself. Existing benchmarks vary these cues together, so they cannot tell how much of the revision actually depends on the speaker. We introduce a no-source condition: the same asserted answer with the explicit speaker removed. Across six open-weight LLMs and seven QA and reasoning datasets, this condition alone causes harmful revision in $66.5\%$ of initially correct cases, compared with $10.3\%$ under a plain re-ask. The effect also remains when the repeated answer is paraphrased and when answer options are hidden in an open-ended setting. Source framing mainly modulates this floor: expert-panel framing raises it, while minimal person labels do not reliably raise it. When models flip, they are usually confidently wrong, and simple recalibration does not recover the original answer. Source attribution still matters, but it should be measured as an increment above this speaker-free floor. The methodological lesson is that conformity benchmarks should first measure what remains after the speaker is removed; without this step, benchmarks may mistake repeated text for social influence.