# paper-forge daily report — 2026-09-29

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

## Reproductions

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

## Top candidates

### [Not All Objectives Are Born Equal: Priority-Constrained Descent for Hierarchical Multi-Objective Optimization](https://arxiv.org/abs/2606.29521)
`2606.29521` · score 68.4 · analysis: heuristic · feasibility 8/10

Deep learning problems rarely involve objectives that are equal in importance. A primary objective defines the goal, whilst secondary objectives, such as sparsity, compression, or robustness constrain the solution.

Rubric evidence: + cheap/closed-form ("closed-form"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("closed-form"); + formal statement ("convergence"); - system/framework (many moving parts) ("framework"); + code: https://github.com/DaraVaram/priority-constrained-descent; + 1 HF upvotes

### [Nonnegative DAG Learning via Concomitant Estimation](https://arxiv.org/abs/2609.34595)
`2609.34595` · score 61.6 · analysis: heuristic · feasibility 8/10

We study the problem of learning directed acyclic graphs (DAGs) with nonnegative edge weights from observational data. We propose the Nonnegative and Concomitant (NoCo) DAG estimator, which jointly recovers the weighted graph structure and the exogenous noise variances in the linear structural equation model for the observations.

Rubric evidence: + cheap/closed-form ("convex"); + toy/synthetic experiments ("Simulated"); + small/synthetic data ("Simulated"); + explicit algorithm/pseudocode ("Algorithm 1"); + concrete method ("closed-form"); no public code found

### [SMAT: Simple and Efficient Merge-Aware Training](https://arxiv.org/abs/2609.33437)
`2609.33437` · score 61.1 · analysis: heuristic · feasibility 6/10

Model merging integrates the capabilities of multiple experts without joint retraining, but standard expert training optimizes task loss alone and does not guarantee good performance after merging. Merge-aware training (MAT) aims to improve merged performance, but existing methods do not fully account for common merging operations and add training cost.

Rubric evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + formal statement ("bound"); + claims simplicity ("Simple"); - survey/position (no method) ("perspective"); + code: https://github.com/egangu/smat; + keywords: sampling, kernel; + 1 HF upvotes

### [RidgeRank: Efficient Visual Document Reranking via Score Fusion and a Shallow Linear Readout](https://arxiv.org/abs/2609.34192)
`2609.34192` · score 57.9 · analysis: heuristic · feasibility 7/10

Multimodal language models rerank visual document retrieval results accurately, but scoring every candidate page at full cost makes them slow. Some methods that compress these rerankers need relevance labels to regain accuracy, and they rank by the reranker score alone.

Claimed: NDCG@5 1.2 (abstract (regex extraction, unverified))
Rubric evidence: + cheap/closed-form ("closed-form"); + explicit algorithm/pseudocode ("Algorithm 1"); + concrete method ("closed-form"); + formal statement ("bound"); no public code found; + keywords: retrieval, ranking

### [SANTA++: Sampling Attention through Representative Keys](https://arxiv.org/abs/2609.35629)
`2609.35629` · score 55.8 · analysis: heuristic · feasibility 5/10

Attention often concentrates on a small subset of tokens in the context, but which subset matters changes from one query to the next. To exploit this changing structure, we introduce SANTA++, a training-free stochastic attention method that uses representative keys for memory-efficient selection without scanning the entire key-value (KV) cache.

Rubric evidence: + formal statement ("bound"); + code: https://github.com/OPUSLab/santapp-kernel-demo.git; + keywords: sampling, retrieval, kernel

### [When Privacy Moves ML-Mediated Decisions On Device: Information and Incentive Misalignment in Auctions](https://arxiv.org/abs/2609.33312)
`2609.33312` · score 54.2 · analysis: heuristic · feasibility 5/10

Moving ML-mediated decision making onto privacy-preserving clients decentralises the economic decision along with the inference. Shared budget constraints then depend on information that cannot be globally current, creating an information-structure failure that conventional pacing is not designed to solve.

Rubric evidence: + toy/synthetic experiments ("simulation"); + formal statement ("bound"); + code: https://github.com/sarkar-dipankar/on-device-auction-audit; + keywords: ranking

### [GUIDE-FBO: Guidance via Uncertainty Intervention and Distributional Exchange for Federated Bayesian Optimization](https://arxiv.org/abs/2609.35038)
`2609.35038` · score 53.2 · analysis: heuristic · feasibility 7/10

Federated Bayesian Optimization (FBO) enables distributed agents to collaboratively optimize expensive black-box objectives without sharing raw local observations. However, effective knowledge transfer remains challenging under communication constraints and task heterogeneity.

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

### [Calibrated Derivative-Process Sensitivity for Gaussian-Process Variable Selection](https://arxiv.org/abs/2609.33549)
`2609.33549` · score 52.6 · analysis: heuristic · feasibility 6/10

Automatic relevance determination (ARD), the default tool for variable selection in Gaussian-process (GP) regression, ranks inputs by inverse lengthscales -- which measure how fast a function varies, not how much an input contributes to prediction -- and offers no calibrated rule for deciding which inputs to keep. The prediction-centred alternative, the derivative sensitivity $ν_j = \mathbb{E}[(\partial f/\partial x_j)^2]$, is available in closed form from a fitted GP, but turning it into a selection rule is harder than it looks: at a null input the estimator is a degenerate quadratic form, so Wald and Bernstein-von Mises cutoffs are anti-conservative, and the natural residual bootstrap is mis-scaled.

Rubric evidence: + cheap/closed-form ("closed form"); + concrete method ("closed form"); + formal statement ("We prove"); no public code found; + keywords: kernel, ranking

### [Improving Test-Time Scaling with Adaptive Looped Transformers](https://arxiv.org/abs/2609.35748)
`2609.35748` · score 48.9 · analysis: heuristic · feasibility 4/10

Looped transformers have demonstrated promising parameter efficiency by reusing layers for latent computation. Prior studies compare looped and non-looped models at matched parameters or per-token FLOPs.

Claimed: accuracy 53.0% (abstract (regex extraction, unverified)); accuracy 3.4 (abstract (regex extraction, unverified))
Rubric evidence: + public benchmarks ("benchmarks"); + code: https://github.com/thu-nics/TaH; + 40 HF upvotes

### [Share-Borne AI Virus: Memory-Hopping Attacks Across LLM Agents](https://arxiv.org/abs/2609.35576)
`2609.35576` · score 48.9 · analysis: heuristic · feasibility 5/10

Large language models are increasingly deployed as stateful assistants that retain information across interactions and use tools to read, modify, and create persistent artifacts. As these artifacts are shared between users, they form an indirect communication channel between otherwise independent assistants.

Rubric evidence: - LLM-based ("LLM"); + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + code: https://github.com/psidharth567/Share-Borne-Virus

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 68.4 | [Not All Objectives Are Born Equal: Priority-Constrained Descent for Hierarchical Multi-Objective Optimization](https://arxiv.org/abs/2606.29521) `2606.29521` | 0.85 | 0.75 | 0.65 | 1.00 | 0.10 | heuristic | True |
| 61.6 | [Nonnegative DAG Learning via Concomitant Estimation](https://arxiv.org/abs/2609.34595) `2609.34595` | 0.85 | 0.75 | 0.90 | 0.00 | 0.00 | heuristic | True |
| 61.1 | [SMAT: Simple and Efficient Merge-Aware Training](https://arxiv.org/abs/2609.33437) `2609.33437` | 0.65 | 0.75 | 0.30 | 1.00 | 0.50 | heuristic | True |
| 57.9 | [RidgeRank: Efficient Visual Document Reranking via Score Fusion and a Shallow Linear Readout](https://arxiv.org/abs/2609.34192) `2609.34192` | 0.70 | 0.40 | 1.00 | 0.00 | 0.40 | heuristic | True |
| 55.8 | [SANTA++: Sampling Attention through Representative Keys](https://arxiv.org/abs/2609.35629) `2609.35629` | 0.50 | 0.40 | 0.55 | 1.00 | 0.60 | heuristic | True |
| 54.2 | [When Privacy Moves ML-Mediated Decisions On Device: Information and Incentive Misalignment in Auctions](https://arxiv.org/abs/2609.33312) `2609.33312` | 0.65 | 0.40 | 0.55 | 1.00 | 0.20 | heuristic | True |
| 53.2 | [GUIDE-FBO: Guidance via Uncertainty Intervention and Distributional Exchange for Federated Bayesian Optimization](https://arxiv.org/abs/2609.35038) `2609.35038` | 0.65 | 0.80 | 0.75 | 0.00 | 0.00 | heuristic | True |
| 52.6 | [Calibrated Derivative-Process Sensitivity for Gaussian-Process Variable Selection](https://arxiv.org/abs/2609.33549) `2609.33549` | 0.70 | 0.40 | 0.75 | 0.00 | 0.40 | heuristic | True |
| 48.9 | [Improving Test-Time Scaling with Adaptive Looped Transformers](https://arxiv.org/abs/2609.35748) `2609.35748` | 0.50 | 0.45 | 0.40 | 1.00 | 0.30 | heuristic | True |
| 48.9 | [Share-Borne AI Virus: Memory-Hopping Attacks Across LLM Agents](https://arxiv.org/abs/2609.35576) `2609.35576` | 0.45 | 0.75 | 0.40 | 1.00 | 0.00 | heuristic | False |
| 48.9 | [Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation](https://arxiv.org/abs/2609.35347) `2609.35347` | 0.50 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 47.9 | [Learning Conditional Expectation Operators via Functional Newton Updates](https://arxiv.org/abs/2609.35598) `2609.35598` | 0.65 | 0.75 | 0.40 | 0.00 | 0.20 | — | — |
| 47.9 | [Multi-Marginal Inverse Optimal Transport for Contrastive Learning Via Explicit Anchor-Positive-Negative Coupling](https://arxiv.org/abs/2609.33741) `2609.33741` | 0.65 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 47.9 | [Post-Training Leaves Behavioral Shadows on Unrelated Decisions](https://arxiv.org/abs/2609.29233) `2609.29233` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.4 | [EvE: An Alternate Optimizer to Adam](https://arxiv.org/abs/2609.35614) `2609.35614` | 0.50 | 0.80 | 0.40 | 0.00 | 0.40 | — | — |
| 46.8 | [Functional Autoencoders for Amplitude-Phase Representation Learning](https://arxiv.org/abs/2609.34207) `2609.34207` | 0.65 | 0.80 | 0.45 | 0.00 | 0.00 | — | — |
| 46.8 | [Draft-KV: Learning Useful Latent Communication Between Language Models](https://arxiv.org/abs/2609.34754) `2609.34754` | 0.50 | 0.40 | 0.40 | 1.00 | 0.23 | — | — |
| 46.3 | [Arbitrary-Accuracy Neural Approximation with Optimal Neuron Count and Near-Optimal Bit Complexity](https://arxiv.org/abs/2609.35628) `2609.35628` | 0.70 | 0.40 | 0.75 | 0.00 | 0.00 | — | — |
| 46.3 | [An Active-Bottleneck Mechanism for Weak-to-Strong Generalization](https://arxiv.org/abs/2609.33835) `2609.33835` | 0.70 | 0.40 | 0.60 | 0.00 | 0.20 | — | — |
| 46.3 | [Playing to Par: Reinforcement Learning for Provably Optimal Quadrilateral Block Decompositions](https://arxiv.org/abs/2609.32146) `2609.32146` | 0.50 | 0.40 | 0.55 | 1.00 | 0.00 | — | — |
| 45.8 | [Who Is Left of Whom? Tracing Spatial Evidence and Role Binding in Relative-Position Reasoning](https://arxiv.org/abs/2609.35486) `2609.35486` | 0.65 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 45.8 | [CLIMB: A Clinical Multimorbidity Benchmark for Diagnosing Co-occurring Conditions through Multiturn Conversations](https://arxiv.org/abs/2609.35462) `2609.35462` | 0.65 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 45.8 | [Weighted Spline-Expanded Networks with Distributional Balancing for Continuous Treatment Effects](https://arxiv.org/abs/2609.33740) `2609.33740` | 0.65 | 0.75 | 0.30 | 0.00 | 0.20 | — | — |
| 45.3 | [Provable Benefits of Regularization: Fast Rates for Adversarial Imitation Learning](https://arxiv.org/abs/2609.35698) `2609.35698` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 44.7 | [Simultaneous Translation between Sign Languages](https://arxiv.org/abs/2609.35608) `2609.35608` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.2 | [KVCMAS: Efficient KV cache Correction for Shared Context in Multi-Agent Systems](https://arxiv.org/abs/2609.34060) `2609.34060` | 0.50 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 43.7 | [CoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings](https://arxiv.org/abs/2609.35645) `2609.35645` | 0.65 | 0.80 | 0.30 | 0.00 | 0.00 | — | — |
| 43.7 | [On-Policy Self-Distillation for Multi-Turn Image Editing](https://arxiv.org/abs/2609.35611) `2609.35611` | 0.50 | 0.45 | 0.30 | 1.00 | 0.10 | — | — |
| 43.7 | [Calibrated Uncertainty for Informative Path Planning in Aquatic Environmental Monitoring](https://arxiv.org/abs/2609.34577) `2609.34577` | 0.65 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 43.2 | [Elicitation and Decision Geometry in Single-Index Bandits](https://arxiv.org/abs/2609.35622) `2609.35622` | 0.70 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 43.2 | [When Does Selection Replace Extraction? A Pre-Registered Test of Agent Memory with a Typed Decision Model](https://arxiv.org/abs/2609.34227) `2609.34227` | 0.30 | 0.40 | 0.55 | 1.00 | 0.20 | — | — |
| 42.6 | [Simplex Diffusion Models](https://arxiv.org/abs/2609.35553) `2609.35553` | 0.55 | 0.40 | 0.65 | 0.00 | 0.20 | — | — |
| 42.1 | [Self-Adapting Group of Experts for Multi-Agent Reasoning](https://arxiv.org/abs/2609.35412) `2609.35412` | 0.50 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 42.1 | [Jailbreaks for Black-Box Uncertainty Quantification in Large Reasoning Models](https://arxiv.org/abs/2609.35350) `2609.35350` | 0.50 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 42.1 | [Just-In-Time Agent Memory with Runtime Agentic Research](https://arxiv.org/abs/2609.34385) `2609.34385` | 0.50 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 42.1 | [Convex Optimization Is Free When Accuracy Is Expensive](https://arxiv.org/abs/2609.35418) `2609.35418` | 0.70 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 42.1 | [Conformal Coverage of Time Series: Validity and Inference](https://arxiv.org/abs/2609.33868) `2609.33868` | 0.50 | 0.40 | 0.70 | 0.00 | 0.20 | — | — |
| 41.6 | [ABC-Align: Prediction-Powered Alignment with Adaptive Bias Control](https://arxiv.org/abs/2609.34374) `2609.34374` | 0.45 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 41.1 | [From Input to Output: A Flexible Agent for Dual-End Interpretation of Sparse Autoencoder Features](https://arxiv.org/abs/2609.35367) `2609.35367` | 0.50 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 41.1 | [Rubric-Calibrated Preferences: Cross-Query Calibration of LLM Judgments via Item Response Theory](https://arxiv.org/abs/2609.35739) `2609.35739` | 0.30 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 41.1 | [Multi-Task Learning of Conditional Mean Operators: applications to dynamical systems and uncertainty quantification](https://arxiv.org/abs/2609.35429) `2609.35429` | 0.70 | 0.40 | 0.50 | 0.00 | 0.00 | — | — |
| 41.1 | [DepthBench: Measuring How Residual Connections Enable More Computational Depth](https://arxiv.org/abs/2609.32534) `2609.32534` | 0.25 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 40.5 | [The Statistical Cost of Causal Discovery with Feedback](https://arxiv.org/abs/2609.34050) `2609.34050` | 0.65 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 40.5 | [Two-Sample Testing for Inhomogeneous Random Graphs in Non-Integral $L_r$ Norms](https://arxiv.org/abs/2609.33968) `2609.33968` | 0.65 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 40.5 | [Change the Product, Keep the Parameters: Associative Algebra Layers for Transformers](https://arxiv.org/abs/2609.32814) `2609.32814` | 0.50 | 0.40 | 0.55 | 0.00 | 0.30 | — | — |
| 40.5 | [Measuring Collapse and Correction in Homogeneous-Panel LLM Debate](https://arxiv.org/abs/2609.35279) `2609.35279` | 0.30 | 0.40 | 0.40 | 1.00 | 0.23 | — | — |
| 40.0 | [DR-net-Mamba: Selective State-Space Modeling for Long-Range ECG Time-Series Denoising](https://arxiv.org/abs/2609.35634) `2609.35634` | 0.50 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 40.0 | [Let the Neurons Die: Exploiting ReLU-Induced Model Degradation](https://arxiv.org/abs/2609.35528) `2609.35528` | 0.50 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 40.0 | [Training and Inference Dynamics of PLDR-LLMs: Row-Map Collapse, Renormalization, and Predictive Reduction](https://arxiv.org/abs/2609.34130) `2609.34130` | 0.30 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 39.9 | [RenderRank: Learning to Rerank Text with Compressed Visual Tokens](https://arxiv.org/abs/2609.35069) `2609.35069` | 0.50 | 0.40 | 0.40 | 0.00 | 0.46 | — | — |
| 39.5 | [NanoForecast v0.5: Competitive Time Series Forecasting Through Training Pipeline Optimization](https://arxiv.org/abs/2609.31669) `2609.31669` | 0.40 | 0.40 | 0.30 | 1.00 | 0.10 | — | — |
| 38.9 | [Learned Preconditioning for a Primal-Dual Interior-Point Method](https://arxiv.org/abs/2609.35665) `2609.35665` | 0.70 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 38.9 | [5W1H+Which: Context-Valid Semantic Indexing with Progressive Ontology Binding](https://arxiv.org/abs/2609.35184) `2609.35184` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Beyond the Beam: Constructive Repair and Candidate Completion for Generative Recommendation](https://arxiv.org/abs/2609.33745) `2609.33745` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [Universality and Generalization of Causal Transformers Across Context Lengths](https://arxiv.org/abs/2609.35055) `2609.35055` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [On the Relation Between Interval Regret and Dynamic Regret](https://arxiv.org/abs/2609.34423) `2609.34423` | 0.70 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 38.9 | [Posterior Regimes and Latent Deception: Variational Bayesian Inference in Hidden Markov Models for Sequential Fraud Detection in Financial Transactions](https://arxiv.org/abs/2609.34031) `2609.34031` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [Valid and Efficient Split Conformal Regression for Time Series](https://arxiv.org/abs/2609.33866) `2609.33866` | 0.50 | 0.40 | 0.70 | 0.00 | 0.00 | — | — |
| 38.9 | [Benign Overfitting for General Norms and Distributions](https://arxiv.org/abs/2609.33675) `2609.33675` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.4 | [Probabilistic Geodesic Flow Matching on Location-Scale Families](https://arxiv.org/abs/2609.34613) `2609.34613` | 0.65 | 0.80 | 0.05 | 0.00 | 0.00 | — | — |
| 37.9 | [Isotonic surrogate modeling for computer experiments with many input variables](https://arxiv.org/abs/2609.35354) `2609.35354` | 0.50 | 0.40 | 0.50 | 0.00 | 0.20 | — | — |
| 37.9 | [TokenCast: Forecasting Token Consumption During LLM Agent Execution](https://arxiv.org/abs/2609.35760) `2609.35760` | 0.30 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 37.9 | [QuanReview: Offline, Auditable Reconciliation of Human and LLM Span Annotations](https://arxiv.org/abs/2609.35685) `2609.35685` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [VEX-Bench: Benchmarking Verification Complexity of LLM-Generated Misinformation](https://arxiv.org/abs/2609.35028) `2609.35028` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.4 | [Twist, Don't Tilt: Trajectory-Exact Constrained Decoding for Masked Diffusion Models](https://arxiv.org/abs/2609.35609) `2609.35609` | 0.35 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 37.4 | [Eval4DiRec: A Unified and Systematic Evaluation Framework for Diffusion-based Recommender Systems](https://arxiv.org/abs/2609.34404) `2609.34404` | 0.35 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 37.4 | [PReCache: Efficient KV Cache Sharing for Multi-LoRA Agents via Low-Rank Precomputation and Neutral Reconstruction](https://arxiv.org/abs/2609.34054) `2609.34054` | 0.25 | 0.45 | 0.30 | 1.00 | 0.20 | — | — |
| 37.4 | [Recursive Harness Distillation across Agents for Robot Manipulation](https://arxiv.org/abs/2609.33378) `2609.33378` | 0.50 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 36.8 | [Neural Harmonic Measure Operator](https://arxiv.org/abs/2609.35752) `2609.35752` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [A Unified Uncertainty Representation for Graph Neural Networks via Doubly-Spectral Stochastic Expansion](https://arxiv.org/abs/2609.35703) `2609.35703` | 0.50 | 0.45 | 0.55 | 0.00 | 0.00 | — | — |
| 36.8 | [Retrieving Biblical Intertextual References in Karen Blixen's Seven Gothic Tales](https://arxiv.org/abs/2609.35765) `2609.35765` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Beyond Fixed Features: Architecture-Dependent Sensitivity to Node Representations under Heterophily](https://arxiv.org/abs/2609.33764) `2609.33764` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Information-Theoretic Analysis of Next-Token Prediction under Markovian Data](https://arxiv.org/abs/2609.34731) `2609.34731` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.3 | [G$^2$PTQ: Improving LLM Post-Training Quantization with Generalized Gradient Compensation](https://arxiv.org/abs/2609.31009) `2609.31009` | 0.30 | 0.40 | 0.30 | 1.00 | 0.10 | — | — |
| 35.8 | [Bounding Retraining Equivalence and the Deletion Floor in Materials Machine Unlearning](https://arxiv.org/abs/2609.35635) `2609.35635` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Hardware-Aware Features for CUTLASS Kernel Selection](https://arxiv.org/abs/2609.35587) `2609.35587` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Beyond Token Scale: Chunk-Level Sparse Autoencoders for Reliable Semantic Feature Discovery](https://arxiv.org/abs/2609.35521) `2609.35521` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Singularities of Non-negative Matrix Factorization and their application to Bayesian inference](https://arxiv.org/abs/2609.34043) `2609.34043` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Oracle-Efficient Online Classification with Stochastic Inputs and Adversarial Outputs](https://arxiv.org/abs/2609.33760) `2609.33760` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Reliable Replay through Spatial Coherence in Online Continual Learning](https://arxiv.org/abs/2609.33725) `2609.33725` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [From Experience to Expertise: Adoption-Aware Memory Learning for Data-Scarce NPU Kernel Synthesis](https://arxiv.org/abs/2609.35568) `2609.35568` | 0.30 | 0.40 | 0.40 | 0.00 | 0.60 | — | — |
| 35.8 | [Scaling Long-Form Story Generation via Narrative State Tracking](https://arxiv.org/abs/2609.35759) `2609.35759` | 0.30 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 35.8 | [Epistemic Policy Divergence in Multi-Turn LLM Contamination: A Protocol-Gradient Investigation](https://arxiv.org/abs/2609.35308) `2609.35308` | 0.30 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 35.3 | [X-Reset: Scaling Object-Centric Reinforcement Learning via Cross-Embodiment Resets](https://arxiv.org/abs/2609.35715) `2609.35715` | 0.65 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 35.3 | [Concurrent Coded Signal-Multiplexing Ranging for Half-Duplex Asynchronous Networks](https://arxiv.org/abs/2609.33753) `2609.33753` | 0.65 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 35.3 | [QC-Stark: A Multi-Task Benchmark Revealing Capability Dissociations in LLMs Evaluated on Quantum Computing Tasks](https://arxiv.org/abs/2609.35581) `2609.35581` | 0.45 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 35.3 | [Relevance-Resolution Transfer via Scale-Decomposable Fractional Diffusion for Multi-Length Cross-Modal Hash Retrieval](https://arxiv.org/abs/2609.34393) `2609.34393` | 0.35 | 0.45 | 0.40 | 0.00 | 0.40 | — | — |
| 35.3 | [Unlocking Few-Step Diffusion for Faithful Previews](https://arxiv.org/abs/2609.34406) `2609.34406` | 0.35 | 0.45 | 0.40 | 0.00 | 0.40 | — | — |
| 34.7 | [Reinforcing Agentic Creativity in Scientific Ideation with Night Science](https://arxiv.org/abs/2609.35706) `2609.35706` | 0.30 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 34.7 | [A Hierarchy of Entropy-Shapley Games for Multivariate Predictive Uncertainty](https://arxiv.org/abs/2609.35217) `2609.35217` | 0.50 | 0.40 | 0.50 | 0.00 | 0.00 | — | — |
| 34.7 | [Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation](https://arxiv.org/abs/2609.16145) `2609.16145` | 0.15 | 0.45 | 0.40 | 1.00 | 0.10 | — | — |
| 34.6 | [An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning](https://arxiv.org/abs/2609.35505) `2609.35505` | 0.30 | 0.45 | 0.05 | 1.00 | 0.26 | — | — |
| 34.2 | [Conformal Prediction and Conditional Coverage for Tabular Foundation Models](https://arxiv.org/abs/2609.34887) `2609.34887` | 0.45 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 34.1 | [ColNanoVDR: Document-Free Query Distillation for Multi-Vector Visual Document Retrieval via Optimal Transport](https://arxiv.org/abs/2609.34899) `2609.34899` | 0.00 | 0.40 | 0.45 | 1.00 | 0.36 | — | — |
| 33.7 | [Statistical Learning of Contractive Dynamical Representations for Composite Adaptive Control](https://arxiv.org/abs/2609.35758) `2609.35758` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [Rethinking Circuit Evaluation: Do Circuits Explain Model Errors?](https://arxiv.org/abs/2609.35686) `2609.35686` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Shockingly Simple Self-retrospection Improves Agentic Models Without RL](https://arxiv.org/abs/2609.35741) `2609.35741` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [MS-GLA: Multi-Scale Gated Linear Attention for Addressing Representational Bottlenecks via Multi-Temporal Resolution](https://arxiv.org/abs/2609.35664) `2609.35664` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Almieyar: A Culturally Grounded Benchmark for Multi-Dialect Arabic Speech Recognition](https://arxiv.org/abs/2609.35564) `2609.35564` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [AwarenessBench: Assessing Cognitive Capabilities of Language Models](https://arxiv.org/abs/2609.35409) `2609.35409` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Do Coding Agents Reuse Existing Code or Reinvent the Wheel?](https://arxiv.org/abs/2609.35357) `2609.35357` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Mitigating Popularity Bias in Recommendation with Global Listwise Learning and Progressive Bi-Weighting](https://arxiv.org/abs/2609.35041) `2609.35041` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [Recommendation Ranking Off-Policy Evaluation under Ranking-Dependent Examination via Examination-Relevance Decomposition](https://arxiv.org/abs/2609.35034) `2609.35034` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Universal Approximation of Measure-to-Measure Operators by Pushforwards](https://arxiv.org/abs/2609.35483) `2609.35483` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [Fast Learning Rate Transfer in Shallow Linear Networks at Growing Training Horizons](https://arxiv.org/abs/2609.35029) `2609.35029` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [Deep Weighted Bellman Residual Minimization for $Q^*$ Estimation](https://arxiv.org/abs/2609.34593) `2609.34593` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [HARMONIA: Interpretable Graph Learning through Mixtures of Neural Bases](https://arxiv.org/abs/2609.33972) `2609.33972` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 32.6 | [The Hidden Perception Constraint in Task-Aware Compression](https://arxiv.org/abs/2609.35684) `2609.35684` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Which the Eye Fears: Writing with Read-Blindness Explains Massive Activations in Transformers](https://arxiv.org/abs/2609.35630) `2609.35630` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [GeoGAE: Scalable Graph-Level Autoencoding via Hyperball Cloud Representations](https://arxiv.org/abs/2609.35527) `2609.35527` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Towards Communication-Efficient Social Intelligence in Language Agents](https://arxiv.org/abs/2609.35749) `2609.35749` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Rubric Rewards from Item Response Theory](https://arxiv.org/abs/2609.35646) `2609.35646` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Language Models Act on Hidden Valence](https://arxiv.org/abs/2609.35591) `2609.35591` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Deep Learning Methods in Neuroscience: From Modeling Molecular Mechanisms to Classifying States of Consciousness](https://arxiv.org/abs/2609.35372) `2609.35372` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [PEAR: Progressive Evidence-Based AutoResearch for Industrial Search Systems](https://arxiv.org/abs/2609.35031) `2609.35031` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Propagate, Then Sharpen: Post-Hoc Refinement of Frozen Node Classifiers](https://arxiv.org/abs/2609.35080) `2609.35080` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Uniform Race: Parameter-Free Approximate Rejection Sampling](https://arxiv.org/abs/2609.34639) `2609.34639` | 0.30 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 32.1 | [Decide, Don't Generate: Competitive Dimensional ABSA with Jev's Typed Decisions](https://arxiv.org/abs/2609.35293) `2609.35293` | 0.15 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 32.1 | [QwenGyre: An Elastic Reinforcement Learning Framework for Training xLong-Horizon Agents](https://arxiv.org/abs/2609.33848) `2609.33848` | 0.30 | 0.40 | 0.45 | 0.00 | 0.30 | — | — |
| 31.6 | [Attention Graphons: A Graph Limit Perspective on Graph Transformers](https://arxiv.org/abs/2609.35620) `2609.35620` | 0.50 | 0.45 | 0.15 | 0.00 | 0.20 | — | — |
| 31.6 | [TCSAlgBench: Benchmarking Automated Proving for Research-Level Theoretical Computer Science](https://arxiv.org/abs/2609.35606) `2609.35606` | 0.30 | 0.45 | 0.45 | 0.00 | 0.20 | — | — |
| 31.6 | [EvoIn: Bridging Evolution and Internalization for Agent Fine-Tuning](https://arxiv.org/abs/2609.35290) `2609.35290` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [No Attention, No Problem: Rethinking Session-based Recommendation with Pure Convolution](https://arxiv.org/abs/2609.34802) `2609.34802` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.1 | [PDMD: Projected Distribution Matching Distillation for Video Diffusion Models](https://arxiv.org/abs/2609.35768) `2609.35768` | 0.35 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 31.1 | [Control-Geometry Straightening for Sampling-Based Latent Planning](https://arxiv.org/abs/2609.35603) `2609.35603` | 0.25 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 31.1 | [How Well Can LLMs Simulate Real Learner Evaluations of Educational Feedback?](https://arxiv.org/abs/2609.35376) `2609.35376` | 0.45 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 31.1 | [Knowing When Thinking Is Not Enough: Teaching Small Reasoning Models to Reason Beyond Their Parametric Knowledge](https://arxiv.org/abs/2609.34327) `2609.34327` | 0.00 | 0.45 | 0.30 | 1.00 | 0.30 | — | — |
| 30.5 | [Harness Learning Enables Generalizable Test-Time Adaptation](https://arxiv.org/abs/2609.35738) `2609.35738` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Graph World Models for Constrained Epidemic Policy Planning](https://arxiv.org/abs/2609.35545) `2609.35545` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Learning the Robustness Mechanism with Bilevel Optimization](https://arxiv.org/abs/2609.35541) `2609.35541` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Optimal Networks for Agentic Information Aggregation](https://arxiv.org/abs/2609.35537) `2609.35537` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Telescopic Language Models](https://arxiv.org/abs/2609.35769) `2609.35769` | 0.00 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 30.5 | [Tracing the Evolution of Oracle Bone Characters Across Three Millennia](https://arxiv.org/abs/2609.35674) `2609.35674` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Guided Uncertainty-Aware Robust Domain Transfer](https://arxiv.org/abs/2609.35654) `2609.35654` | 0.40 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 30.5 | [CoeF-SFL: Preserving Collaborative Server-Client Learning with Enhanced Communication Efficiency](https://arxiv.org/abs/2609.34360) `2609.34360` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [GT-PSSM: Unified Probabilistic Framework for Stochastic Dynamics Modeling and Dependency Learning in Multivariate Time Series Anomaly Detection](https://arxiv.org/abs/2609.34161) `2609.34161` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [How Does "English (US)" Become the Default? Triangulating Structural Bias Towards American English Across the LLM Pipeline](https://arxiv.org/abs/2604.04204) `2604.04204` | 0.05 | 0.40 | 0.40 | 1.00 | 0.10 | — | — |
| 29.5 | [Cartridges++: KV Cache Compression without Off-Context Derailment](https://arxiv.org/abs/2609.35621) `2609.35621` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [Output-aware Residual Stream Pruning for Large Language Models](https://arxiv.org/abs/2609.35579) `2609.35579` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [Late Attention Layers Alone Can Copy Entity Tokens, but Not Without Attending to Their Context](https://arxiv.org/abs/2609.35663) `2609.35663` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [Semantic Prefix Oracles for LLM Decoding: Contracts and Differential Validation](https://arxiv.org/abs/2609.35425) `2609.35425` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [Learning Multimodal Embeddings with Evidence-Aligned Readout](https://arxiv.org/abs/2609.33659) `2609.33659` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [Correct then Forecast: Observer State-Space Models for Time Series Forecasting](https://arxiv.org/abs/2609.33566) `2609.33566` | 0.50 | 0.45 | 0.20 | 0.00 | 0.00 | — | — |
| 29.5 | [When Do Model Internals Help? Exploring the Role of Representation Engineering in LLM Safety](https://arxiv.org/abs/2609.34771) `2609.34771` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 28.9 | [KV-streams for Efficient Compaction in Agentic Reinforcement Learning](https://arxiv.org/abs/2609.35750) `2609.35750` | 0.45 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 28.9 | [RIDE: Reference-Anchored Inference-Time Diffusion Editing for Scaffold Hopping](https://arxiv.org/abs/2609.35623) `2609.35623` | 0.35 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 28.9 | [Quantifying Behavioral Tails in Black-Box Language Models](https://arxiv.org/abs/2609.33638) `2609.33638` | 0.45 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 28.9 | [KernelZero: Co-Evolving Proposer and Coder for Continuously Improved GPU Kernel Generation](https://arxiv.org/abs/2609.33074) `2609.33074` | 0.30 | 0.40 | 0.30 | 0.00 | 0.30 | — | — |
| 28.4 | [MemoReason: Evaluating the Effect of Parametric Memory on Contextual Reasoning in LLMs](https://arxiv.org/abs/2609.35312) `2609.35312` | 0.30 | 0.45 | 0.45 | 0.00 | 0.00 | — | — |
| 28.4 | [Signal or Noise? Modality Contribution and Cooperation in Multimodal GraphRAG](https://arxiv.org/abs/2609.35304) `2609.35304` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 28.4 | [Correcting to Predict: Pseudo-Value Correction for Multimodal Attribute Value Extraction](https://arxiv.org/abs/2609.34383) `2609.34383` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 28.4 | [STITCH-RAG: Spatio-Temporal Influence Tracing over Topic Hypergraphs for Multi-Hop Retrieval-Augmented Generation](https://arxiv.org/abs/2609.34127) `2609.34127` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 27.9 | [Copy the Same, Distill the Difference: Initializing Linear Vision Transformers](https://arxiv.org/abs/2609.35745) `2609.35745` | 0.25 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 27.9 | [Behavioral Foundation Models for Quality Diversity](https://arxiv.org/abs/2609.35615) `2609.35615` | 0.25 | 0.45 | 0.50 | 0.00 | 0.00 | — | — |
| 27.9 | [When Harness Beats Scale, and When Reading Beats Both](https://arxiv.org/abs/2609.34366) `2609.34366` | 0.15 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 27.4 | [How to Loop MoE: Flatten the Experts, Untie the Attention](https://arxiv.org/abs/2609.35751) `2609.35751` | 0.00 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 27.4 | [ScAn-Bench: Evaluating Scaling Analysis Methodology](https://arxiv.org/abs/2609.35707) `2609.35707` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Beyond Energy: When Sustainability Dimensions Reshape LLM Serving Decisions](https://arxiv.org/abs/2609.35569) `2609.35569` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [EvoSkillRec: Skill-Genome Evolution for Recommender Architecture Discovery](https://arxiv.org/abs/2609.34552) `2609.34552` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [SPRINT: Single-Step Generative Recommendation via Average Probability Velocity](https://arxiv.org/abs/2609.34306) `2609.34306` | 0.50 | 0.40 | 0.15 | 0.00 | 0.00 | — | — |
| 27.4 | [Measuring and Mitigating Identity-Cue Preference Drift in LLM-based Recommender Systems](https://arxiv.org/abs/2609.34229) `2609.34229` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [Sparsity by Default: The Theory and Practice of ARD in Gaussian Process Regression for Variable Selection](https://arxiv.org/abs/2609.33550) `2609.33550` | 0.50 | 0.40 | 0.00 | 0.00 | 0.20 | — | — |
| 26.8 | [What masking geometry works best for EEG foundation models?](https://arxiv.org/abs/2609.33487) `2609.33487` | 0.05 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 26.3 | [Can LLMs Value the Right Evidence? Evidence-Value Misalignment in Dynamic Medical Diagnosis](https://arxiv.org/abs/2609.35627) `2609.35627` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [FactorEngram: Factorized N-gram Memory with Basis-Level Gating for Language Models](https://arxiv.org/abs/2609.35578) `2609.35578` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Representation Alignment as a Bottleneck in LLM-Based Retrosynthesis Planning](https://arxiv.org/abs/2609.35571) `2609.35571` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Frontier Learning: Training LLM Reasoners at the Edge of Capability](https://arxiv.org/abs/2609.35426) `2609.35426` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [TRACE: Single-Pass Decoding-Trace Risk Localization for Generation Calibration](https://arxiv.org/abs/2609.35387) `2609.35387` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Unbiased Top-$k$ Estimation for On-Policy Distillation](https://arxiv.org/abs/2609.34447) `2609.34447` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [AlphaPareto: Formulaic Alpha Discovery with LLM-Guided Multi-Objective Reinforcement Learning](https://arxiv.org/abs/2609.34188) `2609.34188` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [Transferable Mass Spectrum Prediction via Reference-Guided Test-time Specialization](https://arxiv.org/abs/2609.35649) `2609.35649` | 0.25 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 25.8 | [High-Level Text Preprocessing for Semantic Similarity Analysis of Discursive Texts: A Framework and Empirical Demonstration](https://arxiv.org/abs/2609.33983) `2609.33983` | 0.35 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 25.8 | [DISCO: Distributed Long Context Scaling with Grounding-Reasoning Disaggregation](https://arxiv.org/abs/2609.33485) `2609.33485` | 0.30 | 0.40 | 0.30 | 0.00 | 0.10 | — | — |
| 25.8 | [Groupwise Agentic Grading and Advantage Redistribution for Code Agent RL](https://arxiv.org/abs/2609.32577) `2609.32577` | 0.00 | 0.40 | 0.30 | 0.00 | 0.70 | — | — |
| 25.3 | [Unifying Distributional Training for One-Step Visual Generation](https://arxiv.org/abs/2609.35763) `2609.35763` | 0.50 | 0.00 | 0.30 | 0.00 | 0.20 | — | — |
| 25.3 | [SEABench: Benchmarking Endogenous Misalignment In Self-Evolving Agents](https://arxiv.org/abs/2609.35596) `2609.35596` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [AraDynFact: Dynamic Evaluation of Factual Knowledge in Arabic](https://arxiv.org/abs/2609.35461) `2609.35461` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 24.7 | [MeqMuon: Matrix-Equilibrating Muon for LLM Pretraining](https://arxiv.org/abs/2609.35701) `2609.35701` | 0.05 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 24.7 | [Statistical Benefits of Fine-Tuning from Pretrained Initialization in Diagonal Linear Networks](https://arxiv.org/abs/2609.34756) `2609.34756` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [Two-Timescale Fine-tuning Provably Learns New Features for Two-Layer ReLU Networks](https://arxiv.org/abs/2609.34667) `2609.34667` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [LLMs are General Asynchronous Agents](https://arxiv.org/abs/2609.35427) `2609.35427` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [Multilinguality in Hybrid Attention LLMs](https://arxiv.org/abs/2609.35378) `2609.35378` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [AX is the New AEO](https://arxiv.org/abs/2609.34951) `2609.34951` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 24.2 | [Understanding Generalization Requires Universal Induction](https://arxiv.org/abs/2609.34458) `2609.34458` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 22.6 | [Can Generative Retrievers Learn Semantic IDs Without Forgetting How to Speak?](https://arxiv.org/abs/2609.35430) `2609.35430` | 0.05 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 21.6 | [In-Flight KV Cache with Clean Anchors for Faster Autoregressive Video Diffusion](https://arxiv.org/abs/2609.32540) `2609.32540` | 0.00 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 21.1 | [Distillation Defenses Easily Break After Reinforcement Learning](https://arxiv.org/abs/2609.35699) `2609.35699` | 0.30 | 0.40 | 0.15 | 0.00 | 0.00 | — | — |
| 17.9 | [Rethinking Personalized Generation: Test-Time Alignment via Factorized Ranking Models](https://arxiv.org/abs/2609.35695) `2609.35695` | 0.00 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 17.9 | [A Statistical Perspective on Knowledge Distillation: Foundations, Classical Methods, and Large Language Model Extensions](https://arxiv.org/abs/2609.33727) `2609.33727` | 0.30 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 16.8 | [Less Sycophancy, Stronger Refusal? Lessons for AI Safety from Mechanistic Interpretability](https://arxiv.org/abs/2609.35544) `2609.35544` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [Spontaneous Context Restoration: How Language Models Recover from Corrupted Inputs](https://arxiv.org/abs/2609.35475) `2609.35475` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |

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