# paper-forge daily report — 2026-10-08

167 new papers screened; 10 analyzed in depth; 0 reproduction runs.

## Reproductions

- none today (reimplementation is opt-in: `forge daily --reimplement`)

## Top candidates

### [Identifiability of a dissipative knowledge-dynamics model: exact recovery under designed excitation, degeneration on observational data](https://arxiv.org/abs/2610.09889)
`2610.09889` · score 62.6 · analysis: heuristic · feasibility 8/10

Human learning is a dissipative dynamical process: mastery accumulates through practice, decays through forgetting, and propagates across interdependent concepts. We model it as a nonlinear dissipative system of ordinary differential equations whose parameters are mechanistically meaningful (a concept-transfer matrix encoding prerequisite coupling, per-concept forgetting rates, and a saturating practice-response gain), and we study when those parameters can actually be recovered from data.

Rubric evidence: + cheap/closed-form ("closed-form"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + concrete method ("closed-form"); + formal statement ("We prove"); + claims simplicity ("simple"); no public code found

### [Finite-Rank Logistic Gaussian Processes with Exact Likelihood for Conditional Density Estimation](https://arxiv.org/abs/2610.09452)
`2610.09452` · score 61.6 · analysis: heuristic · feasibility 8/10

Conditional density estimation describes how the entire distribution of a response changes with covariates, and in imaging studies also with location. Logistic Gaussian processes (LGP) give a flexible prior for such densities.

Rubric evidence: + cheap/closed-form ("closed form"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("closed form"); + formal statement ("bound"); no public code found; + keywords: sampling

### [A self-learning scientific agent for X-ray diffraction](https://arxiv.org/abs/2610.07862)
`2610.07862` · score 61.1 · analysis: heuristic · feasibility 6/10

A central challenge for scientific agents is to turn analytical experience into reusable expertise grounded in physical evidence. Here we introduce Gan Jiang, a self-learning agent for powder X-ray diffraction built on a diffraction-analysis ecosystem we developed: XMatcher, XQueryer, XDecomposer and WPEM.

Rubric evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + claims simplicity ("simple"); - system/framework (many moving parts) ("agents is to turn analytical experience into reusable expertise grounded in physical evidence. Here we introduce Gan Jiang, a self-learning agent for powder X-ray diffraction built on a diffraction-analysis ecosystem we developed: XMatcher, XQueryer, XDecomposer and WPEM. Together, these engines span phase identification, multiphase decomposition and physics-constrained whole-pattern modelling. Gan Jiang converts analytical experience into executable skills by diagnosing failures, revising skill instructions and code, and validating revisions before reuse, without retraining the language model or changing the underlying physical models. Skills selected using development data and frozen before held-out evaluation achieve higher refinement scores than the original expert-designed skills across FullProf, GSAS-II and PyWPEM. The agent resolves strongly overlapping reflections, quantifies a five-phase ancient Egyptian cosmetic, tracks lattice evolution in an operating battery and compares atomic configurations in a disordered oxide catalyst. On DeltaXRDbench, it leads the evaluated methods in single- and multiphase identification across simulated and experimental data. Without supplied composition, single-phase top-1 accuracies reach 96.30\%, 81.78\% and 40.83\% on MP500, RRUFF and opXRD, respectively, compared with 58.00\%, 58.47\% and 26.45\% for the strongest comparator. These results demonstrate how an integrated scientific tool"); + code: https://github.com/DeltaGanjiang/XDecomposer; + 7 HF upvotes

### [Transition Path Sampling Using Koopman Operators and Exit-Time Optimal Control](https://arxiv.org/abs/2610.10054)
`2610.10054` · score 57.4 · analysis: heuristic · feasibility 7/10

Sampling transitions between metastable states is a central problem in dynamical systems theory and molecular dynamics in particular. A key challenge is the existence of high free-energy barriers that separate the states, making transitions extremely rare.

Rubric evidence: + cheap/closed-form ("closed form"); + toy/synthetic experiments ("simulation"); + concrete method ("we propose a new"); + formal statement ("We prove"); no public code found; + keywords: sampling, kernel

### [Q-Learning with Scalar Adjoint Matching](https://arxiv.org/abs/2610.10437)
`2610.10437` · score 56.8 · analysis: heuristic · feasibility 6/10

Flow policies capture rich and diverse action distributions, and fine-tuning them with off-policy RL to improve beyond the demonstrations has drawn growing interest. However, fine-tuning a flow policy against a learned value function is not trivial, because the policy generates its action over many flow steps.

Rubric evidence: - pretraining ("pretrained"); + cheap/closed-form ("closed-form"); + explicit algorithm/pseudocode ("Algorithm 1"); + concrete method ("closed-form"); + code: https://github.com/yonghdong/sqam; + 10 HF upvotes

### [On KL-Regularized Policy Optimization](https://arxiv.org/abs/2610.08963)
`2610.08963` · score 56.3 · analysis: heuristic · feasibility 5/10

Asynchronous reinforcement learning (RL) for large language model (LLM) agents trains one policy on trajectories generated by another: rollouts come from stale checkpoints, and the inference engine's probabilities differ from the trainer's even at identical parameters. Standard remedies either clip importance ratios, which biases the update, or, as in GRPO, sample a group of responses per prompt, which is costly when episodes are long.

Rubric evidence: - LLM-based ("large language model"); + cheap/closed-form ("closed-form"); + concrete method ("closed-form"); + formal statement ("bound"); - system/framework (many moving parts) ("agents trains one policy on trajectories generated by another: rollouts come from stale checkpoints, and the inference engine's probabilities differ from the trainer's even at identical parameters. Standard remedies either clip importance ratios, which biases the update, or, as in GRPO, sample a group of responses per prompt, which is costly when episodes are long. We propose KL-Regularized Policy Optimization (KLPO), a framework that anchors the KL regularizer at the sampler. The regularized improvement step then has a closed-form Gibbs solution, and KLPO fits its log-ratio optimality condition by least squares on the sampler's own trajectories, so the sampler probability enters through a log-ratio and no importance weights are needed. Profiling out the regression intercept replaces the intractable log-partition function with the signal's sampler mean plus a sampler-to-trainer KL divergence. For token-level policy mirror descent targets, we show that the resulting gradient can be computed from terminal returns without a critic, via sampler-centered scores or a single trajectory residual, even under stochastic tool"); + code: https://github.com/yifanzhang-pro/KLPO; + keywords: monte carlo; + 18 HF upvotes

### [Dataset Pruning from First Principles: A Label-Free Linear Programming Approach](https://arxiv.org/abs/2610.10347)
`2610.10347` · score 55.8 · analysis: heuristic · feasibility 7/10

Dataset pruning reduces a large training set to a representative subset while preserving model performance. Existing geometry-based methods typically assume that nearby points in embedding space share similar properties.

Rubric evidence: + cheap/closed-form ("closed-form"); + small/synthetic data ("CIFAR-10"); + public benchmarks ("benchmarks"); + concrete method ("closed-form"); + formal statement ("convergence"); - system/framework (many moving parts) ("framework"); no public code found; + keywords: sampling

### [RobotWorld: Benchmarking Multimodal Agents for Robot Use Across Diverse Tasks and Embodiments](https://arxiv.org/abs/2610.10409)
`2610.10409` · score 53.7 · analysis: heuristic · feasibility 5/10

General-purpose agents increasingly write code, use tools, and complete complex digital tasks, raising the question of how far these capabilities carry into the physical world. To investigate this, we introduce RobotWorld, a challenging simulation testbed for robot use: turning instructions and observations into physical task execution through robot interfaces.

Rubric evidence: + toy/synthetic experiments ("simulation"); + formal statement ("bound"); - system/framework (many moving parts) ("agents increasingly write code, use tools"); + code: https://github.com/robotworldai/robotworld; + 25 HF upvotes

### [Conformal Prediction for Spatially Dependent Data via Sequential Whitening](https://arxiv.org/abs/2610.10168)
`2610.10168` · score 51.1 · analysis: heuristic · feasibility 6/10

Split conformal prediction uses prediction errors on held-out (calibration) data to determine how wide the prediction intervals should be. It guarantees distribution-free finite-sample coverage when these errors and the error at the target site are exchangeable.

Rubric evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + formal statement ("bound"); no public code found; + keywords: sampling

### [DSReg: Provably Recovering Individual World Latents without Reconstruction](https://arxiv.org/abs/2610.09457)
`2610.09457` · score 51.1 · analysis: heuristic · feasibility 6/10

Methods that recover individual latent variables of the world, from nonlinear ICA to dictionary learning and causal representation learning, anchor the latents to observations through reconstruction, auxiliary supervision, or distributional asymmetries such as non-Gaussianity. Methods without these anchors, including joint-embedding predictive architectures (JEPAs), identify the latent state only up to a linear transformation, so individual latents remain mixed.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + formal statement ("we prove"); no public code found; + keywords: regularization

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 62.6 | [Identifiability of a dissipative knowledge-dynamics model: exact recovery under designed excitation, degeneration on observational data](https://arxiv.org/abs/2610.09889) `2610.09889` | 0.85 | 0.80 | 0.90 | 0.00 | 0.00 | heuristic | True |
| 61.6 | [Finite-Rank Logistic Gaussian Processes with Exact Likelihood for Conditional Density Estimation](https://arxiv.org/abs/2610.09452) `2610.09452` | 0.85 | 0.75 | 0.75 | 0.00 | 0.20 | heuristic | True |
| 61.1 | [A self-learning scientific agent for X-ray diffraction](https://arxiv.org/abs/2610.07862) `2610.07862` | 0.65 | 0.75 | 0.45 | 1.00 | 0.30 | heuristic | True |
| 57.4 | [Transition Path Sampling Using Koopman Operators and Exit-Time Optimal Control](https://arxiv.org/abs/2610.10054) `2610.10054` | 0.85 | 0.40 | 0.75 | 0.00 | 0.40 | heuristic | True |
| 56.8 | [Q-Learning with Scalar Adjoint Matching](https://arxiv.org/abs/2610.10437) `2610.10437` | 0.45 | 0.40 | 0.90 | 1.00 | 0.30 | heuristic | False |
| 56.3 | [On KL-Regularized Policy Optimization](https://arxiv.org/abs/2610.08963) `2610.08963` | 0.50 | 0.40 | 0.65 | 1.00 | 0.50 | heuristic | True |
| 55.8 | [Dataset Pruning from First Principles: A Label-Free Linear Programming Approach](https://arxiv.org/abs/2610.10347) `2610.10347` | 0.70 | 0.80 | 0.65 | 0.00 | 0.20 | heuristic | True |
| 53.7 | [RobotWorld: Benchmarking Multimodal Agents for Robot Use Across Diverse Tasks and Embodiments](https://arxiv.org/abs/2610.10409) `2610.10409` | 0.65 | 0.40 | 0.45 | 1.00 | 0.30 | heuristic | True |
| 51.1 | [Conformal Prediction for Spatially Dependent Data via Sequential Whitening](https://arxiv.org/abs/2610.10168) `2610.10168` | 0.65 | 0.75 | 0.55 | 0.00 | 0.20 | heuristic | True |
| 51.1 | [DSReg: Provably Recovering Individual World Latents without Reconstruction](https://arxiv.org/abs/2610.09457) `2610.09457` | 0.65 | 0.75 | 0.55 | 0.00 | 0.20 | heuristic | True |
| 50.5 | [Sparsifying Stochasticity, Not Capacity: Partial Stochasticity via Deep Weight Factorization of Prior Scales](https://arxiv.org/abs/2610.09886) `2610.09886` | 0.70 | 0.80 | 0.40 | 0.00 | 0.20 | — | — |
| 50.5 | [TIDES: Implicit Time-Awareness in Selective State Space Models](https://arxiv.org/abs/2605.09742) `2605.09742` | 0.70 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 49.5 | [What Transfers from a VLM Teacher? Comparing Supervision Signals for Visual Document Retrieval](https://arxiv.org/abs/2610.09177) `2610.09177` | 0.50 | 0.40 | 0.40 | 1.00 | 0.40 | — | — |
| 47.9 | [Temporally Interpretable Differentiable Decision Trees](https://arxiv.org/abs/2610.10367) `2610.10367` | 0.65 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 47.9 | [m-Set Adversarial Bandits with Winner Feedback](https://arxiv.org/abs/2610.10128) `2610.10128` | 0.65 | 0.75 | 0.40 | 0.00 | 0.20 | — | — |
| 47.9 | [Possibilistic Radial Transport for Approximate IM Inference](https://arxiv.org/abs/2610.09956) `2610.09956` | 0.65 | 0.75 | 0.40 | 0.00 | 0.20 | — | — |
| 47.9 | [Outperformance Inverse Optimization: Learning Objective Functions that Outperform Agent Decisions](https://arxiv.org/abs/2610.09890) `2610.09890` | 0.65 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 47.9 | [On-Policy Distillation with Negative-Policy Rollouts](https://arxiv.org/abs/2610.07874) `2610.07874` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.9 | [Recurrent Looped Transformer](https://arxiv.org/abs/2610.07591) `2610.07591` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.4 | [SoccerNet-FoulRet: Retrieving Semantically Similar Soccer Foul Videos](https://arxiv.org/abs/2610.09742) `2610.09742` | 0.50 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 46.3 | [Which Rollout Taught It That? BehaviorTrace and the Limits of Training-Data Attribution in Online RL](https://arxiv.org/abs/2610.10422) `2610.10422` | 0.50 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 46.3 | [Cluster-Robust Prediction-Powered Inference](https://arxiv.org/abs/2610.09601) `2610.09601` | 0.70 | 0.40 | 0.60 | 0.00 | 0.20 | — | — |
| 46.3 | [DLoop: Looped Speculative Decoding](https://arxiv.org/abs/2610.07659) `2610.07659` | 0.45 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 45.8 | [Executing Causal Structure Learning with Linear-Attention Transformers](https://arxiv.org/abs/2610.10395) `2610.10395` | 0.65 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 45.3 | [Rubix: Global Correspondence-Free Point Set Alignment through Assignment Geometry](https://arxiv.org/abs/2610.10408) `2610.10408` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 45.3 | [Measurement-Efficient Differentiable Quantum Architecture Search for Combinatorial Optimization](https://arxiv.org/abs/2610.10351) `2610.10351` | 0.50 | 0.45 | 0.30 | 1.00 | 0.20 | — | — |
| 44.7 | [Safe Meta-Policy Design with Risk Control](https://arxiv.org/abs/2610.10393) `2610.10393` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Agentic RAG Evaluation: Budget Allocation Across Questions, Trajectories, and Reads](https://arxiv.org/abs/2610.05034) `2610.05034` | 0.50 | 0.40 | 0.30 | 0.00 | 0.90 | — | — |
| 44.2 | [SciExam for ENSO: Can AI Agents Build Climate Models?](https://arxiv.org/abs/2610.10513) `2610.10513` | 0.50 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 44.2 | [Universal Local Error and Realized Amplification for the First-Order EDM Predictor](https://arxiv.org/abs/2610.10190) `2610.10190` | 0.10 | 0.75 | 0.55 | 1.00 | 0.20 | — | — |
| 43.6 | [StepCAD: Mesh-to-CAD Code Generation via LLM Policy and Geometry-Guided Search](https://arxiv.org/abs/2610.03799) `2610.03799` | 0.30 | 0.45 | 0.40 | 1.00 | 0.36 | — | — |
| 43.2 | [Evolutionary Architecture Search for Chlorophyll-$a$ Prediction in Lakes using Sentinel-2](https://arxiv.org/abs/2610.10496) `2610.10496` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 43.2 | [ORDERS: An Empirical Study of Norm-Rank Aggregation for Personalized Federated Learning](https://arxiv.org/abs/2610.10361) `2610.10361` | 0.50 | 0.75 | 0.40 | 0.00 | 0.20 | — | — |
| 43.2 | [Broadly Applicable Approximate MCMC for Switching Stochastic Differential Equations Using Uniformization and Time-Conditioned Factorized Neural Likelihood Estimation](https://arxiv.org/abs/2610.10194) `2610.10194` | 0.50 | 0.75 | 0.40 | 0.00 | 0.20 | — | — |
| 42.6 | [PoreML: A Data-Driven Framework for Learning Multiphase Flow in Porous Media](https://arxiv.org/abs/2610.10314) `2610.10314` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.6 | [AdaPS-LiNGAM: Adaptive Predecessor Selection for Linear Non-Gaussian Acyclic Models under Small-Sample Settings](https://arxiv.org/abs/2610.09782) `2610.09782` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.6 | [WebFovea: When the Model Is Right but the Click Is Wrong -- Reliable Round Trips for Vision-Based Web Agents on Live Websites](https://arxiv.org/abs/2610.03036) `2610.03036` | 0.30 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 42.1 | [Decoupling Exploration from Optimization in RLVR](https://arxiv.org/abs/2610.10536) `2610.10536` | 0.50 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 42.1 | [Decentralized SGD under Heavy-Tailed Noise: Optimal Convergence Rates and the Role of Gradient Clipping](https://arxiv.org/abs/2610.10527) `2610.10527` | 0.70 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 42.1 | [Two-Level Softmax Sampling Done Right: Correcting Bias from Size Imbalance and Dispersion](https://arxiv.org/abs/2610.10483) `2610.10483` | 0.70 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 41.6 | [Seq-Flow: Efficient Probabilistic Forecasting with Self-Rollout Error Control](https://arxiv.org/abs/2610.10440) `2610.10440` | 0.35 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 41.6 | [Minimal Witness Reinforcement Learning](https://arxiv.org/abs/2610.07226) `2610.07226` | 0.30 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 41.6 | [ReSAIL: Mitigating Collapse in Iterative Agent Self-Distillation](https://arxiv.org/abs/2609.39306) `2609.39306` | 0.30 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 40.5 | [Derivative Gaussian Processes on a Two-Direction Budget](https://arxiv.org/abs/2610.10428) `2610.10428` | 0.65 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 40.5 | [DecepEval: A Benchmark for Evaluating Deception in LLM Agents](https://arxiv.org/abs/2610.07967) `2610.07967` | 0.30 | 0.45 | 0.30 | 1.00 | 0.30 | — | — |
| 40.0 | [HAN-Mamba: Hierarchical Selective State Space Networks for Multi-Scale Financial Volatility Forecasting](https://arxiv.org/abs/2610.10323) `2610.10323` | 0.70 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 40.0 | [Estimating Uncoded Crash Factors with Tabular Foundation and System One Models: Kumo Tabular and Jev](https://arxiv.org/abs/2610.10321) `2610.10321` | 0.30 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 40.0 | [Reasoning-Token Spikes Under Prompted Untruthful Responding in Large Language Models](https://arxiv.org/abs/2610.10405) `2610.10405` | 0.30 | 0.40 | 0.55 | 1.00 | 0.00 | — | — |
| 40.0 | [Expected Sample Complexity in Multi-Armed Bandits](https://arxiv.org/abs/2610.09929) `2610.09929` | 0.50 | 0.40 | 0.45 | 0.00 | 0.40 | — | — |
| 39.5 | [From Pareto to Preference: Personalized Test-Time Scaling via Amortized Agentic Policy Discovery](https://arxiv.org/abs/2610.09684) `2610.09684` | 0.30 | 0.40 | 0.30 | 1.00 | 0.30 | — | — |
| 39.5 | [Reading Position Is the Baseline to Beat: A Time-Ordered Evaluation of Personalised Highlight Prediction](https://arxiv.org/abs/2610.09262) `2610.09262` | 0.65 | 0.75 | 0.00 | 0.00 | 0.20 | — | — |
| 39.5 | [Gains and Collapse in On-Policy Distillation:A Reinforcement Learning Perspective](https://arxiv.org/abs/2610.03185) `2610.03185` | 0.50 | 0.40 | 0.00 | 1.00 | 0.30 | — | — |
| 39.3 | [EngramEdit: Decoupled Knowledge Updates in LLMs through Conditional Memory](https://arxiv.org/abs/2610.10533) `2610.10533` | 0.30 | 0.40 | 0.40 | 1.00 | 0.16 | — | — |
| 38.9 | [Best Arm Identification for Bandits with Shifting Means](https://arxiv.org/abs/2610.10488) `2610.10488` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Your Prompt Should Do More: Effects of Retrieval Instructions in Embedding Models](https://arxiv.org/abs/2610.10508) `2610.10508` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [From Prompts to Trees: Effective LLM-Guided Tree Generation for Few-Shot Tabular Classification](https://arxiv.org/abs/2610.10227) `2610.10227` | 0.30 | 0.40 | 0.50 | 1.00 | 0.00 | — | — |
| 38.9 | [Shared Gaussianization: What Gaussian Regularizers Certify About Contrastive Learning, and What They Miss](https://arxiv.org/abs/2610.10299) `2610.10299` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Neural Sampling with Reweighted Normalizing Flows via the Wasserstein--Fisher--Rao JKO Scheme](https://arxiv.org/abs/2610.10278) `2610.10278` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Global Exponential Convergence of Two-Layer Linear Network Training](https://arxiv.org/abs/2610.09356) `2610.09356` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.4 | [ExperienceIndex: Artifact-Grounded Memory](https://arxiv.org/abs/2610.10091) `2610.10091` | 0.65 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 38.4 | [Certified by Abstention: Distribution-Free Guarantees for Chain-of-Thought Verifiers at Small Calibration Budgets](https://arxiv.org/abs/2610.09541) `2610.09541` | 0.65 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 38.4 | [Questioning the Questions: Sustaining Self-Evolution in Reasoning Models](https://arxiv.org/abs/2610.04299) `2610.04299` | 0.50 | 0.45 | 0.40 | 0.00 | 0.30 | — | — |
| 37.9 | [YANchor-4B: Effective Long-Horizon Reasoning in O(N) Time with O(1) Memory](https://arxiv.org/abs/2610.10118) `2610.10118` | 0.20 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 37.9 | [The Long Road to the Same Answer: Cognitive Bias Under Escalating Reasoning Budgets in Large Language Models](https://arxiv.org/abs/2610.10049) `2610.10049` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [Finding the Right Balance: Relevance and Diversity in LLM Retrieval](https://arxiv.org/abs/2610.09412) `2610.09412` | 0.30 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 37.9 | [EntroPrefill: Renyi-Guided Context Pruning with Conditional Stability Guarantees for Retrieval-Augmented Generation](https://arxiv.org/abs/2610.09757) `2610.09757` | 0.50 | 0.45 | 0.45 | 0.00 | 0.20 | — | — |
| 37.4 | [NeuralBES: A Differentiable, Control-Aware Emulator for Scalable Building Energy Modeling](https://arxiv.org/abs/2610.10459) `2610.10459` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 37.4 | [Reflected Anchored Langevin Algorithms](https://arxiv.org/abs/2610.09522) `2610.09522` | 0.35 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 36.8 | [Oracle-Efficient and Parameter-Free Agnostic Smoothed Online Learning](https://arxiv.org/abs/2610.10499) `2610.10499` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.8 | [From Chunks to Functional Evidence: Function-Aware Retrieval for EDA Documentation QA](https://arxiv.org/abs/2610.09361) `2610.09361` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Towards Calibrated Probabilistic Forecasts for Events of Interest via Outcome-Conditional Recalibration](https://arxiv.org/abs/2610.10076) `2610.10076` | 0.50 | 0.45 | 0.55 | 0.00 | 0.00 | — | — |
| 36.8 | [Fluctuations of Nonlinear Observables in Mean Field Neural Network Training](https://arxiv.org/abs/2610.09768) `2610.09768` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.8 | [Scalable Logistic Gaussian Process Density Regression with Kinetic Langevin Sampling](https://arxiv.org/abs/2610.09591) `2610.09591` | 0.30 | 0.45 | 0.40 | 0.00 | 0.60 | — | — |
| 35.8 | [Why Forget-Only Unlearning Needs Memorization](https://arxiv.org/abs/2610.10519) `2610.10519` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [A Good Self-Teacher Meets the Student Where They Are: Joint On-Policy Learning and Teaching](https://arxiv.org/abs/2610.10447) `2610.10447` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Boosting and the Expressive Power of Simple Weak Learners via the $γ$-VC Dimension](https://arxiv.org/abs/2610.10383) `2610.10383` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Continual Learning without Continual Training](https://arxiv.org/abs/2610.10379) `2610.10379` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Data Reuse in Non-Stationary Learning](https://arxiv.org/abs/2610.10340) `2610.10340` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Average-Reward Reinforcement Learning for Multichain MDPs: A Hierarchical Decomposition Approach](https://arxiv.org/abs/2610.10326) `2610.10326` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Training with Missed Targets in Generative Recommendation: Separating Supervision from Probability Competition](https://arxiv.org/abs/2610.10124) `2610.10124` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Quantize by Drift: Label-Free Mixed-Precision Post-Training Quantization for Text Embedders](https://arxiv.org/abs/2610.09227) `2610.09227` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Trustworthy Domain-Specific AI for Structured Knowledge Retrieval and Reasoning](https://arxiv.org/abs/2610.08894) `2610.08894` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Finite-Sample Approximation of Hessian-Guided Perturbed Wasserstein Gradient Flows](https://arxiv.org/abs/2610.10218) `2610.10218` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [RoBART: Bayesian Additive Regression Trees with Tree-Specific Rotations](https://arxiv.org/abs/2610.10214) `2610.10214` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Controlling Dependence in Implicit Generative Models via Spread Mutual Information](https://arxiv.org/abs/2610.10021) `2610.10021` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Leaner Transformers Can Easily Learn to Cluster](https://arxiv.org/abs/2610.09760) `2610.09760` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Unbounded Characteristic and Universal Kernels](https://arxiv.org/abs/2610.09731) `2610.09731` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Gauss-Newton Accuracy and Indefinite Hessians: Uniform Coexistence in Low-Cost Sets](https://arxiv.org/abs/2610.09675) `2610.09675` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.7 | [CADFather: Autonomous CAD Reconstruction through Coordinated Tool Use](https://arxiv.org/abs/2610.09127) `2610.09127` | 0.25 | 0.40 | 0.30 | 1.00 | 0.16 | — | — |
| 35.3 | [Conditional Flow Matching for Generation of 3D Multi-variable Instantaneous Urban Microclimate Fields](https://arxiv.org/abs/2610.10430) `2610.10430` | 0.65 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 35.3 | [FastOPD: On-Policy Distillation for Lightweight VLA Deployment](https://arxiv.org/abs/2610.02832) `2610.02832` | 0.60 | 0.40 | 0.30 | 0.00 | 0.10 | — | — |
| 34.7 | [AutoAdapt: Automatic Domain Discovery Enables Low-Cost Extensibility](https://arxiv.org/abs/2610.10349) `2610.10349` | 0.50 | 0.45 | 0.45 | 0.00 | 0.00 | — | — |
| 34.7 | [Shaer: Controlled Arabic Poetry Generation with Meter Subform and Semantic Conditioning](https://arxiv.org/abs/2610.09756) `2610.09756` | 0.30 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 34.7 | [BEACON-SP: Ontology-Grounded GraphRAG Framework for Clinical Suicide Risk Assessment](https://arxiv.org/abs/2610.09026) `2610.09026` | 0.50 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 34.2 | [Improving Proactive AI Assistance with Hierarchical Procedural Understanding](https://arxiv.org/abs/2610.06505) `2610.06505` | 0.40 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 33.7 | [Composing What Each Teacher Learned: Multi-Teacher On-Policy Distillation through Teacher-Relative Shifts](https://arxiv.org/abs/2610.10460) `2610.10460` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [ResidualQuant: KV Cache Quantization for Looped Transformers with 2-Bit Residuals](https://arxiv.org/abs/2610.10381) `2610.10381` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [RSIGym: A Flexible Environment for Recursive Self-Improvement](https://arxiv.org/abs/2610.10310) `2610.10310` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Mechanics of Long-Context Hybrid Models Part 1.1: From Hybrid Attention to Hybrid Position](https://arxiv.org/abs/2610.10114) `2610.10114` | 0.05 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 33.7 | [Eigenvalues of the Hessian in Deep Learning: The Origin of Symmetry and Its Breaking](https://arxiv.org/abs/2610.09919) `2610.09919` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Task-Sufficient Contraction: Source Selection for Machine Information Interfaces](https://arxiv.org/abs/2610.08884) `2610.08884` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 32.6 | [Input-Blind Controls Produce Substantial Oracle Headroom for Layer Programs in Multiple-Choice Evaluation](https://arxiv.org/abs/2610.10368) `2610.10368` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Pathwise Information Certificates for Decentralized Adaptive Sensing](https://arxiv.org/abs/2610.10362) `2610.10362` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [How Do Transformers Learn to Represent Symmetries?](https://arxiv.org/abs/2610.10305) `2610.10305` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Learning to Act with Task Progress: Distilling Small Agents from Compact Teacher Supervision](https://arxiv.org/abs/2610.10332) `2610.10332` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Itgan at NADI 2026 shared task: Parameter-Efficient Whisper Adaptation for Robust, Mixed-Dialect and Code-Switched Arabic ASR](https://arxiv.org/abs/2610.09934) `2610.09934` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [CrossWeave: Bridging Perspectives Across Online Communities with a Dual-Pane Design](https://arxiv.org/abs/2610.10441) `2610.10441` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Building Navigable Graphs Without Search in Three Composable Stages](https://arxiv.org/abs/2610.09041) `2610.09041` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Revisiting Explainable AI through Model-Independent Concept Dictionaries](https://arxiv.org/abs/2610.10301) `2610.10301` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Computations of the slice genus and the unknotting number of links via machine learning](https://arxiv.org/abs/2610.10206) `2610.10206` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Gaussian Equivalence for Multi-Head Self-Attention](https://arxiv.org/abs/2610.10033) `2610.10033` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Extreme Binary Classification: Extreme Value Theory for Extreme Constraint on False Negative](https://arxiv.org/abs/2610.09984) `2610.09984` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Unpaired Canonical Correlation Analysis](https://arxiv.org/abs/2610.09530) `2610.09530` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Benign Overfitting under Heterogeneous Input Fusion](https://arxiv.org/abs/2610.09340) `2610.09340` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.1 | [PHRBench: A Behavioral Evaluation of Post-Hallucination Reasoning in LLMs](https://arxiv.org/abs/2610.10455) `2610.10455` | 0.45 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 31.6 | [Distilling Graph Geometry: Knowledge Gap from GNNs to MLPs](https://arxiv.org/abs/2610.10520) `2610.10520` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [Continual Graph Multi-Agent Reinforcement Learning](https://arxiv.org/abs/2610.10302) `2610.10302` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [Judging in Latent Space: Efficient Generative Reward Modeling via Semantics-Preserving Compression](https://arxiv.org/abs/2610.09788) `2610.09788` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.1 | [Decoupling Logic from Persona: Structural Immunity of Edge LLM Agents to Context Pollution](https://arxiv.org/abs/2610.09772) `2610.09772` | 0.45 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 30.5 | [CoTrace: Data Recipes for Training Terminal Agents with Harness-Model Co-Evolution](https://arxiv.org/abs/2610.10426) `2610.10426` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [SkillSandbox: Skill Verification via Dynamic Scenario Synthesis](https://arxiv.org/abs/2610.10088) `2610.10088` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [PARC-Loc: Text-to-Point-Cloud Localization with Partial Assignment and Relational Consistency](https://arxiv.org/abs/2610.09761) `2610.09761` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [The Silhouette Operator: Identifiability of Low-Rank Measures from One-Dimensional Projections](https://arxiv.org/abs/2610.09687) `2610.09687` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Adjoint-Based Calibration and Optimal Control of Stochastic Multiscale Bioprocess Digital Twins](https://arxiv.org/abs/2610.09505) `2610.09505` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [PatchBench: Measuring Collateral Damage in Activation Patching](https://arxiv.org/abs/2610.10276) `2610.10276` | 0.30 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 30.5 | [Inverting Multi-Vector Visual Document Indices](https://arxiv.org/abs/2610.09920) `2610.09920` | 0.25 | 0.45 | 0.40 | 0.00 | 0.30 | — | — |
| 30.0 | [RunningTab: Direct Workspace Interaction with Environment-Side Tabs](https://arxiv.org/abs/2610.10444) `2610.10444` | 0.30 | 0.45 | 0.30 | 0.00 | 0.30 | — | — |
| 29.5 | [Cross-Domain Pretraining for Steady-State Neural CFD Surrogates](https://arxiv.org/abs/2610.10398) `2610.10398` | 0.40 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 29.5 | [Kernel Autoresearch for Open-Ended Model Discovery](https://arxiv.org/abs/2610.10394) `2610.10394` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [LLM Persuasion Is in the Eye of the Evaluation](https://arxiv.org/abs/2610.10232) `2610.10232` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [Does Document Structure Help Dense Retrieval? A Placebo-Controlled Ablation of Four Mechanisms Across Two Corpora](https://arxiv.org/abs/2610.10170) `2610.10170` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [From High Recall to High Utility: Dataset-Adaptive Post-Processing of LLM-Generated Customer Intents](https://arxiv.org/abs/2610.09039) `2610.09039` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 28.4 | [Beyond Outcome Rewards: Constructing and Assigning Retrieval Credit for Search Agents](https://arxiv.org/abs/2610.10179) `2610.10179` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 27.9 | [Koopman Observers for Diffusion Acceleration: Correcting Feature Forecasts with Shallow Measurements](https://arxiv.org/abs/2610.10366) `2610.10366` | 0.35 | 0.35 | 0.30 | 0.00 | 0.20 | — | — |
| 27.8 | [Structuring MoE Expert Selection for Agentic Reinforcement Learning](https://arxiv.org/abs/2610.07332) `2610.07332` | 0.30 | 0.45 | 0.30 | 0.00 | 0.16 | — | — |
| 27.4 | [Steerspeech: Activation Steering For Emotion Control In Generated Speech](https://arxiv.org/abs/2610.10415) `2610.10415` | 0.40 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 27.4 | [LLM4Impact: Integrating Heterogeneous Information for Scientific Impact Prediction](https://arxiv.org/abs/2610.10138) `2610.10138` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [EASE: Entropy-Adaptive Distribution Shaping for Evading AI-generated Text Detectors](https://arxiv.org/abs/2610.09976) `2610.09976` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [InsClaimBench: Benchmarking Insurance Claim Adjudication Across the Decision Chain](https://arxiv.org/abs/2610.09671) `2610.09671` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Rephrase Before You Act: Characterizing and Mitigating Language Sensitivity in Vision-Language-Action Models](https://arxiv.org/abs/2610.10526) `2610.10526` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [SemanticFold: Latent Sequence Compression SeparatesLanguage Modeling, Decodability, and Reasoning](https://arxiv.org/abs/2610.10304) `2610.10304` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [HySPE: Positional Encoding via Symplectic Dual Shears](https://arxiv.org/abs/2610.10154) `2610.10154` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [I would rather quit NLP than read another paper like this: The rise of antithesis in NLP papers](https://arxiv.org/abs/2610.10092) `2610.10092` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Constrained-Action AI Remediation for SIEM/XDR via a NeMo-Guardrails Proxy](https://arxiv.org/abs/2610.09906) `2610.09906` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [From Expert-Guided Proof Search to Automated Open-Problem Solving](https://arxiv.org/abs/2610.09769) `2610.09769` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [The Impact of Backbone Evolution on LLM-Based Relevance Assessments](https://arxiv.org/abs/2610.09820) `2610.09820` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.3 | [Training Parallel Speculative Draft Models by Directly Minimizing Expected Decoding Rounds](https://arxiv.org/abs/2610.10411) `2610.10411` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [LiveMACE: Process-Aware Evaluation of LLM Agent Capabilities in Evolving Markets](https://arxiv.org/abs/2610.09872) `2610.09872` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [Training Advisors for LLM Agents from Task Outcomes](https://arxiv.org/abs/2610.09858) `2610.09858` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [MIRROR: From Imitation to Internalization in LLM Personalization](https://arxiv.org/abs/2610.09795) `2610.09795` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [OrBIT: Structure-Guided Embedding Compression](https://arxiv.org/abs/2610.10385) `2610.10385` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [Fault-tolerant foundation models](https://arxiv.org/abs/2610.10311) `2610.10311` | 0.00 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [Document-Level Text Simplification in Estonian Using Large Language Models](https://arxiv.org/abs/2610.10378) `2610.10378` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [InterView-C: A Synchronized Multimodal Corpus of VR Avatar-Mediated Survey Interviews](https://arxiv.org/abs/2610.10145) `2610.10145` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 23.7 | [Kinetic Langevin Meets Split Gibbs: Accelerated Posterior Sampling for Imaging Inverse Problems with Diffusion Priors](https://arxiv.org/abs/2610.10187) `2610.10187` | 0.35 | 0.00 | 0.45 | 0.00 | 0.20 | — | — |
| 23.2 | [A Deafening Silence: Catastrophic Forgetting Lives in the Output Embeddings of Tokens the Data Never Speaks](https://arxiv.org/abs/2610.09835) `2610.09835` | 0.00 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 22.6 | [SpikingVLA: Asynchronous Spiking Vision-Language-Action Models](https://arxiv.org/abs/2610.09710) `2610.09710` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.6 | [Pre-training of Bayesian Optimization Algorithm through Bayesian Optimization](https://arxiv.org/abs/2610.10186) `2610.10186` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 21.1 | [Towards Explaining Query Expansion Performance in Information Retrieval](https://arxiv.org/abs/2610.09724) `2610.09724` | 0.30 | 0.40 | 0.00 | 0.00 | 0.20 | — | — |
| 18.4 | [Thinking in Depth: Retrospective Inference for Tabular Foundation Models](https://arxiv.org/abs/2610.10317) `2610.10317` | 0.05 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 17.9 | [Cache the Encoder Within:Compact, Reusable Memory across LLM Queries](https://arxiv.org/abs/2610.10058) `2610.10058` | 0.00 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 17.9 | [Validity Without Ground Truth: What Stated-Preference Economics Offers the Evaluation of Language Models](https://arxiv.org/abs/2610.10506) `2610.10506` | 0.30 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 17.9 | [Sensitive-Topic Leakage Through LLM Routing Metadata: Measurement and Mitigation](https://arxiv.org/abs/2610.09981) `2610.09981` | 0.30 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 16.8 | [Nobody Truly Agrees on Sentiment: Humans, Bespoke Tools, and LLMs Struggle with Social Media Texts](https://arxiv.org/abs/2610.10318) `2610.10318` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [Bridge Routing Heads: Where Multilingual Multi-hop Reasoning Lives in LLMs](https://arxiv.org/abs/2610.09733) `2610.09733` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 14.7 | [SOTA: Stock Options Trading Agents Guided by Option-Implied Return Distributions](https://arxiv.org/abs/2610.10407) `2610.10407` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [GAGR-Lab: Evaluating Joint Spatial-Geometric and Analytic Function Reasoning](https://arxiv.org/abs/2610.10201) `2610.10201` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |

Scores: weighted mean of rubric criteria (0–1 each) × 100; see README for the rubric.