# paper-forge daily report — 2026-10-09

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

## Reproductions

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

## Top candidates

### [Transformed Samplers with Variance Reduction](https://arxiv.org/abs/2610.10870)
`2610.10870` · score 71.1 · analysis: heuristic · feasibility 8/10

Markov chain Monte Carlo (MCMC) methods are the standard tool for computing expectations under complex probability distributions. Control variates reduce the variance of the resulting estimates, but a good control variate requires solving the Poisson equation of the sampler, which rarely admits a closed-form solution.

Rubric evidence: + cheap/closed-form ("closed-form"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("closed-form"); + formal statement ("Theorem"); + claims simplicity ("simple"); no public code found; + keywords: sampling, kernel, monte carlo

### [Retrieval-Centric Deep Learning in Growing Nonparametric Neural Networks](https://arxiv.org/abs/2610.03858)
`2610.03858` · score 67.1 · analysis: heuristic · feasibility 6/10

We investigate a general-purpose layer for deep learning that, instead of compressing arbitrary-size training data into fixed-size weight matrices, stores a new pair of key-value representations for every data point during training, and retrieves and recombines these representations through an attention mechanism at inference time - resulting in a growing neural net (NN). While Irie et al.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + formal statement ("theorem"); + claims simplicity ("simple"); - survey/position (no method) ("perspective"); + code: https://github.com/paradigms-of-intelligence/rcdl; + keywords: optimizer, retrieval, kernel; + 6 HF upvotes

### [RIFT: Relative Isolation From Trees For Anomaly Detection](https://arxiv.org/abs/2610.12244)
`2610.12244` · score 64.7 · analysis: heuristic · feasibility 9/10

Isolation Forest (IF) is a widely used baseline for unsupervised anomaly detection. Recent studies provide a closed-form expression for the infinite-forest limit for one-dimensional data.

Rubric evidence: + cheap/closed-form ("closed-form"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmark"); + explicit algorithm/pseudocode ("Algorithm 1"); + concrete method ("closed-form"); + formal statement ("Theorem"); + claims simplicity ("simple"); no public code found

### [Training on the Future: A Delay-Aware Audit of Test-Time Adaptation for Time-Series Forecasting](https://arxiv.org/abs/2610.12232)
`2610.12232` · score 61.1 · analysis: heuristic · feasibility 7/10

Test-time adaptation (TTA) methods for time-series forecasting update a deployed model, or a small adapter around it, from incoming ground truth. But the label of an $H$-step forecast exists only $H$ steps later, and real data pipelines add further delay.

Rubric evidence: + cheap/closed-form ("closed-form"); + public benchmarks ("benchmarks"); + concrete method ("closed-form"); + formal statement ("bound"); + claims simplicity ("simple"); + code: https://github.com/Xodios/TRAINING-ON-THE-FUTURE-A-DELAY-AWARE-AUDIT-OF-TEST-TIME-ADAPTATION-FOR-TIME-SERIES-FORECASTING

### [Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching](https://arxiv.org/abs/2610.12421)
`2610.12421` · score 58.9 · analysis: heuristic · feasibility 6/10

Dense correspondence matching has historically been bounded by simplifying spatio-temporal priors, such as smooth motion and rigid geometry. While effective for classical tasks, these assumptions break down in image editing and reference-guided generation (IEG), where transformations can preserve visual identity while breaking physical continuity.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); - system/framework (many moving parts) ("framework"); + code: https://github.com/luping-liu/FreeMatching; + 28 HF upvotes

### [ReSPO: Reshaped Sequence Policy Optimization for Gradient Starvation in Off-Policy Learning](https://arxiv.org/abs/2609.35433)
`2609.35433` · score 58.4 · analysis: heuristic · feasibility 5/10

Reinforcement learning from verifiable rewards (RLVR) frequently reuses rollouts across multiple policy updates, increasing the mismatch between the current policy and the data-generating policy. We identify a sign-dependent gradient starvation problem in clipped policy optimization: clipping suppresses under-generated positive responses at the low-importance-weight tail while permitting severely over-generated negative responses to dominate the high-weight tail.

Rubric evidence: + public benchmarks ("benchmark"); + formal statement ("bound"); + code: https://github.com/yhangchen/ReSPO-code; + keywords: kernel, variational; + 7 HF upvotes

### [Prediction-Powered Data Fusion for Treatment Effect Estimation](https://arxiv.org/abs/2610.12332)
`2610.12332` · score 54.7 · analysis: heuristic · feasibility 6/10

Randomized controlled trials (RCTs) identify treatment effects without confounding but are often small, whereas observational studies (OBS) are large but may be confounded. Many estimators combining a small RCT with a large OBS have been developed for the average treatment effect (ATE) and the conditional ATE (CATE).

Rubric evidence: + cheap/closed-form ("closed-form"); + concrete method ("closed-form"); + formal statement ("bound"); - system/framework (many moving parts) ("framework"); + code: https://github.com/CausalDataScience/DataFusionPPI

### [From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment Simulation](https://arxiv.org/abs/2610.06100)
`2610.06100` · score 54.7 · analysis: heuristic · feasibility 6/10

Realistic environment replicas are increasingly valuable for training and evaluating LLM agents, yet the original systems may be inaccessible or impractical to reproduce. We explore agentic language world modeling: rather than rebuilding an executable environment, a world model agent serves as the environment for a task agent and supports faithful and stateful simulation.

Rubric evidence: - LLM-based ("LLM"); + toy/synthetic experiments ("Simulation"); + small/synthetic data ("simulated"); + claims simplicity ("simple"); - system/framework (many moving parts) ("Agentic"); + code: https://github.com/ruyue0001/trace2env; + 164 HF upvotes

### [A Closer Look at Agentic BBO: Benchmarking LLM Agents for Black-Box Optimization](https://arxiv.org/abs/2610.12183)
`2610.12183` · score 53.6 · analysis: heuristic · feasibility 5/10

Black-box optimization (BBO) arises in many scientific and engineering problems where objective evaluations are expensive and limited. Recent large language model (LLM) agents offer a new way to approach BBO by combining task semantics, computation, optimization tools, and feedback-driven decision making, showing great potential due to the integration with mathematically rigorous tools.

Rubric evidence: - LLM-based ("LLM"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmark"); - system/framework (many moving parts) ("Agentic"); + code: https://github.com/lamda-bbo/agentic-bbo; + keywords: optimizer; + 2 HF upvotes

### [LAIR-Net: Leaky Alignment-Impulse Residual Networks for Tabular Regression](https://arxiv.org/abs/2610.11538)
`2610.11538` · score 52.1 · analysis: heuristic · feasibility 7/10

Deep randomized models fix hidden-layer parameters through random initialization and learn only closed-form readouts, typically adding depth by stacking random trans formations without target-aware control of hidden-state evolution. We propose LAIR Net, the Leaky Alignment-Impulse Residual Network, which mixes a shallow learned anchor into each hidden state through a leaky residual transition.

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

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 71.1 | [Transformed Samplers with Variance Reduction](https://arxiv.org/abs/2610.10870) `2610.10870` | 0.85 | 0.75 | 0.90 | 0.00 | 0.60 | heuristic | True |
| 67.1 | [Retrieval-Centric Deep Learning in Growing Nonparametric Neural Networks](https://arxiv.org/abs/2610.03858) `2610.03858` | 0.65 | 0.75 | 0.30 | 1.00 | 0.88 | heuristic | True |
| 64.7 | [RIFT: Relative Isolation From Trees For Anomaly Detection](https://arxiv.org/abs/2610.12244) `2610.12244` | 0.85 | 0.80 | 1.00 | 0.00 | 0.00 | heuristic | True |
| 61.1 | [Training on the Future: A Delay-Aware Audit of Test-Time Adaptation for Time-Series Forecasting](https://arxiv.org/abs/2610.12232) `2610.12232` | 0.70 | 0.45 | 0.90 | 1.00 | 0.00 | heuristic | True |
| 58.9 | [Beyond Spatio-Temporal Priors: A Generalizable Approach for Dense Correspondence Matching](https://arxiv.org/abs/2610.12421) `2610.12421` | 0.65 | 0.80 | 0.30 | 1.00 | 0.30 | heuristic | True |
| 58.4 | [ReSPO: Reshaped Sequence Policy Optimization for Gradient Starvation in Off-Policy Learning](https://arxiv.org/abs/2609.35433) `2609.35433` | 0.50 | 0.45 | 0.55 | 1.00 | 0.70 | heuristic | True |
| 54.7 | [Prediction-Powered Data Fusion for Treatment Effect Estimation](https://arxiv.org/abs/2610.12332) `2610.12332` | 0.70 | 0.40 | 0.65 | 1.00 | 0.00 | heuristic | True |
| 54.7 | [From Traces to Agentic Worlds: Agentic Language World Models for Interactive Environment Simulation](https://arxiv.org/abs/2610.06100) `2610.06100` | 0.45 | 0.75 | 0.45 | 1.00 | 0.30 | heuristic | False |
| 53.6 | [A Closer Look at Agentic BBO: Benchmarking LLM Agents for Black-Box Optimization](https://arxiv.org/abs/2610.12183) `2610.12183` | 0.45 | 0.80 | 0.30 | 1.00 | 0.36 | heuristic | False |
| 52.1 | [LAIR-Net: Leaky Alignment-Impulse Residual Networks for Tabular Regression](https://arxiv.org/abs/2610.11538) `2610.11538` | 0.85 | 0.45 | 0.75 | 0.00 | 0.00 | heuristic | True |
| 51.1 | [A Balanced Data Diet: Addressing the Exploration Bottleneck in Mega-Scale RL for Robot Control](https://arxiv.org/abs/2610.12465) `2610.12465` | 0.65 | 0.75 | 0.55 | 0.00 | 0.20 | — | — |
| 51.1 | [Exploiting Gradients in Bayesian Inference of Expensive Simulators](https://arxiv.org/abs/2610.12076) `2610.12076` | 0.65 | 0.40 | 0.55 | 1.00 | 0.00 | — | — |
| 51.0 | [Chaos in the Text: Revealing the Modality Preference in Mixed-Modality Retrievers](https://arxiv.org/abs/2610.11816) `2610.11816` | 0.50 | 0.45 | 0.40 | 1.00 | 0.43 | — | — |
| 48.9 | [Efficient and Generalizable Archetypal Analysis for Discrete Data](https://arxiv.org/abs/2610.12035) `2610.12035` | 0.85 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 47.9 | [SplitJEPA: Learning Invariant and Variant Latent Worlds without Reconstruction](https://arxiv.org/abs/2610.12349) `2610.12349` | 0.65 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 47.9 | [Score-Based Learning of Cluster DAGs from Interventions](https://arxiv.org/abs/2610.11947) `2610.11947` | 0.65 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 47.9 | [From Geometry to Generalization: Why Row Normalization Can Beat Adam and Muon](https://arxiv.org/abs/2610.11309) `2610.11309` | 0.65 | 0.75 | 0.40 | 0.00 | 0.20 | — | — |
| 46.8 | [Unlocking the Regulatory Genome by ARGUS: An Evidence-Constrained Agentic Framework for Interpreting Single Nucleotide Variants](https://arxiv.org/abs/2610.12281) `2610.12281` | 0.45 | 0.75 | 0.30 | 1.00 | 0.00 | — | — |
| 46.5 | [SpaceCast-Bench: Evaluating Predictive Spatial Reasoning in Vision-Language Models](https://arxiv.org/abs/2610.12402) `2610.12402` | 0.50 | 0.45 | 0.30 | 1.00 | 0.28 | — | — |
| 46.3 | [$σ$Transfer: Uncertainty Transfer from Small to Large Networks under $μ\mathrm{P}$](https://arxiv.org/abs/2610.11668) `2610.11668` | 0.50 | 0.75 | 0.55 | 0.00 | 0.20 | — | — |
| 45.8 | [ISBO: Scalable Spatio-Temporal Bayesian Optimization with Log Gaussian Cox Process Models via the INLA-SPDE Approach](https://arxiv.org/abs/2610.12213) `2610.12213` | 0.65 | 0.75 | 0.45 | 0.00 | 0.00 | — | — |
| 45.3 | [Differentiable Systematic Resampling for Variational Sequential Monte Carlo](https://arxiv.org/abs/2610.12094) `2610.12094` | 0.50 | 0.40 | 0.55 | 0.00 | 0.60 | — | — |
| 45.3 | [Learning structured linear dynamical systems from missing observations](https://arxiv.org/abs/2610.11869) `2610.11869` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 44.7 | [Marformer: A Transformer for Predicting Missing Data Distributions](https://arxiv.org/abs/2610.12379) `2610.12379` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [InterviewPlayground: A Simulation Environment for Evaluating AI Interviewers](https://arxiv.org/abs/2610.12023) `2610.12023` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Embedding-Bias in Conditional Independence Testing](https://arxiv.org/abs/2610.11584) `2610.11584` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Calibrating Ambiguity Set via Diagnostic Transport for Distributionally Robust Optimization](https://arxiv.org/abs/2610.10793) `2610.10793` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [ReGain: Restoring Subject Fidelity in Personalization on Synthetic Images](https://arxiv.org/abs/2609.38680) `2609.38680` | 0.50 | 0.75 | 0.40 | 0.00 | 0.30 | — | — |
| 44.2 | [HRIL: Learning Multimodal Synergy via Higher-Order Tensor Modeling](https://arxiv.org/abs/2610.12393) `2610.12393` | 0.50 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 44.2 | [On-Policy Distillation Teaches New Skills but Not New Knowledge](https://arxiv.org/abs/2610.09639) `2610.09639` | 0.65 | 0.75 | 0.30 | 0.00 | 0.10 | — | — |
| 43.2 | [Toward Joint Optimization of Circuit Depth and Training Data Size in Adaptively Grown Quantum Classifiers](https://arxiv.org/abs/2610.12428) `2610.12428` | 0.50 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 43.2 | [Verification with Transfer: Exact Information Frontiers and Their Price in Calls](https://arxiv.org/abs/2610.12211) `2610.12211` | 0.70 | 0.40 | 0.60 | 0.00 | 0.00 | — | — |
| 43.2 | [Quickest Change Detection with Diffusion-Integrated Scores](https://arxiv.org/abs/2610.12200) `2610.12200` | 0.50 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 43.2 | [Easy to anticipate, hard to compute: boundary dependence finds the computed outputs that entropy patching misses](https://arxiv.org/abs/2610.11790) `2610.11790` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 42.6 | [PSI-SINDy: Post-Selection Inference for Sparse Identification of Nonlinear Dynamics](https://arxiv.org/abs/2610.11486) `2610.11486` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.6 | [U-Space: Uncovering When and Why Uncertainty Arises in Language Models](https://arxiv.org/abs/2610.09087) `2610.09087` | 0.30 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 42.6 | [SparseEngine: Sparse-First Inference Engine](https://arxiv.org/abs/2609.39068) `2609.39068` | 0.30 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 42.1 | [Subspace Uncertainty and Sharp Sampling Thresholds on the Boolean Cube](https://arxiv.org/abs/2610.12358) `2610.12358` | 0.50 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 42.1 | [DataSense-Bench: The First Step Toward an AI Scientist](https://arxiv.org/abs/2610.12190) `2610.12190` | 0.30 | 0.45 | 0.45 | 1.00 | 0.20 | — | — |
| 42.1 | [Optimal random quantisers for spherically symmetric distributions](https://arxiv.org/abs/2610.11772) `2610.11772` | 0.70 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 42.1 | [Beyond Distributional Fidelity: Causal-Penalized Diffusion for Synthetic Tabular Data](https://arxiv.org/abs/2610.11407) `2610.11407` | 0.50 | 0.80 | 0.30 | 0.00 | 0.20 | — | — |
| 42.1 | [A General $\widetildeΩ(\sqrt{T γ_T})$ Lower Bound for Kernel Bandits](https://arxiv.org/abs/2610.11082) `2610.11082` | 0.50 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 41.1 | [Syn-Omni: Structured Specialization and Progressive Collaboration for Omnimodal Embeddings](https://arxiv.org/abs/2610.12256) `2610.12256` | 0.50 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 41.1 | [Efficient quadratic entropy with distance sketches](https://arxiv.org/abs/2610.11976) `2610.11976` | 0.50 | 0.45 | 0.45 | 0.00 | 0.40 | — | — |
| 41.1 | [FastBench: Can Streaming VLMs Perceive High-Dynamic Real-World Streams?](https://arxiv.org/abs/2610.12427) `2610.12427` | 0.30 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 40.0 | [Composite Online-to-Nonconvex Conversion with Optimal Oracle Complexity](https://arxiv.org/abs/2610.12328) `2610.12328` | 0.70 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 40.0 | [SanSi: A Looped Typed Decision Model for System 1.5 Thinking](https://arxiv.org/abs/2610.07730) `2610.07730` | 0.25 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 39.5 | [Regularized Small Area Estimation with Graph Laplacian Benchmarking priors](https://arxiv.org/abs/2610.11266) `2610.11266` | 0.65 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 39.5 | [EDiS: Edge Disjoint Subgraph Sparsification Framework for Graph Neural Networks](https://arxiv.org/abs/2610.09059) `2610.09059` | 0.50 | 0.45 | 0.45 | 0.00 | 0.30 | — | — |
| 38.9 | [Closing the Horizon Gap in Policy Optimization for Adversarial MDPs](https://arxiv.org/abs/2610.12362) `2610.12362` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [NativeScope: Relation-Localized Retrieval over Native Topology with a Correct Anchor](https://arxiv.org/abs/2610.12243) `2610.12243` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [When KL Regularization Misfires in Group Policy Optimization](https://arxiv.org/abs/2610.12161) `2610.12161` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [Compact and Efficient Indexes for Learned Sparse Retrieval](https://arxiv.org/abs/2610.12300) `2610.12300` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [Accelerating Non-Smooth and Heavy-Tailed Sampling](https://arxiv.org/abs/2610.11139) `2610.11139` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Explaining the Saliency Map Sparsity of Adversarially-Trained Neural Networks](https://arxiv.org/abs/2610.10666) `2610.10666` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.4 | [RiCo: Neural Simulation of Rigid-Body Interactions via Local Contact Reasoning](https://arxiv.org/abs/2610.12333) `2610.12333` | 0.65 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 38.4 | [Is Real-World Training Data Necessary for Generalist Graph Anomaly Detection?](https://arxiv.org/abs/2610.12167) `2610.12167` | 0.45 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 37.9 | [A Unified Bellman Operator for Safety-Critical Reinforcement Learning](https://arxiv.org/abs/2610.12420) `2610.12420` | 0.50 | 0.40 | 0.65 | 0.00 | 0.00 | — | — |
| 37.9 | [Accurate but Not Humble: Evaluating Epistemic Humility in LLM Agents under Knowledge Conflict](https://arxiv.org/abs/2610.12360) `2610.12360` | 0.30 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 37.9 | [Examining Social Attribution in LLM Reasoning: A Theory-Guided Probing Methodology](https://arxiv.org/abs/2610.12022) `2610.12022` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [GRPODropout: Less is More for Online Reinforcement Learning Rollouts](https://arxiv.org/abs/2610.11854) `2610.11854` | 0.30 | 0.40 | 0.15 | 1.00 | 0.40 | — | — |
| 37.4 | [ViSkill: Reinforcing VLM Agents with Evolving Visual-Native Skills](https://arxiv.org/abs/2610.12403) `2610.12403` | 0.50 | 0.00 | 0.30 | 1.00 | 0.30 | — | — |
| 37.4 | [Minimax Gaussian Mechanisms for Continual Machine Unlearning](https://arxiv.org/abs/2610.11628) `2610.11628` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 36.8 | [Caught in the Act: Probes Effectively Detect Sabotage and Catch Unverbalized Deception](https://arxiv.org/abs/2610.12445) `2610.12445` | 0.30 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 36.8 | [Can AI Agents Learn Their Way to the Top? Evaluating Heuristic Learning in a Long-Running Game Agent Competition](https://arxiv.org/abs/2610.12341) `2610.12341` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Detecting Spin in Clinical Trials with Large Language Models](https://arxiv.org/abs/2610.11845) `2610.11845` | 0.30 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 36.8 | [Autoregressive Retriever: Improving Query Understanding from Item Feedback for Universal Multimodal Retrieval](https://arxiv.org/abs/2610.11666) `2610.11666` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Learning Multi-Step Query Rewriting via Corpus Feedback for Conversational Search](https://arxiv.org/abs/2610.10955) `2610.10955` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.3 | [Event-Centric Memory with Query-Aware Graph Augmentation for Long-Term Conversational Agents](https://arxiv.org/abs/2610.11920) `2610.11920` | 0.45 | 0.45 | 0.30 | 0.00 | 0.40 | — | — |
| 36.3 | [V-CoLA: Vision Token Compression with Linear Attention](https://arxiv.org/abs/2610.11251) `2610.11251` | 0.50 | 0.45 | 0.30 | 0.00 | 0.30 | — | — |
| 36.2 | [One Block, Multiple Depths: Recurrent Vision Transformers with Depth-Programmed Experts](https://arxiv.org/abs/2610.12448) `2610.12448` | 0.70 | 0.00 | 0.40 | 0.00 | 0.36 | — | — |
| 36.0 | [SpatialOPSD: Self-Distilling Spatial Intelligence from Verified Coding Agent Traces](https://arxiv.org/abs/2610.11366) `2610.11366` | 0.30 | 0.45 | 0.45 | 0.00 | 0.48 | — | — |
| 35.8 | [Prospective Prediction of OOD Degradation from Source-Side Training Dynamics](https://arxiv.org/abs/2610.12397) `2610.12397` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Learning to Plan by Looking Back: Hindsight Hierarchies for Training Reasoning Models](https://arxiv.org/abs/2610.12168) `2610.12168` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Language-Specific Effects of Tokenizer Choice in Multilingual Language Models](https://arxiv.org/abs/2610.12144) `2610.12144` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Rehearse Everything, Remember Nothing: Attic-KV Rehearses What Will Be Read](https://arxiv.org/abs/2610.12133) `2610.12133` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Language Models for Page-Level Layout Decisions in E-commerce Search](https://arxiv.org/abs/2610.10920) `2610.10920` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Sequential Conditional Independence Testing with Machine Learning Models](https://arxiv.org/abs/2610.11388) `2610.11388` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Tight Bounds for Equivalence Testing with Non-Adaptive Conditional Samples](https://arxiv.org/abs/2610.11145) `2610.11145` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Conformal Prediction under Partial Verification](https://arxiv.org/abs/2610.10829) `2610.10829` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Density Ratio Estimation with Stein Displacement Fields](https://arxiv.org/abs/2610.12437) `2610.12437` | 0.60 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 35.8 | [DataVista: Diagnosing Multimodal LLMs on Data Video Understanding](https://arxiv.org/abs/2610.11993) `2610.11993` | 0.30 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 35.3 | [Just Weather Scoring: Efficient End-to-end Nowcasting with Distributional Diffusion](https://arxiv.org/abs/2610.12189) `2610.12189` | 0.35 | 0.45 | 0.55 | 0.00 | 0.20 | — | — |
| 34.7 | [Learning to Steer, Steering to See: Unveiling the Geometry of RLVR in Large Language Models via Trainable Vectors](https://arxiv.org/abs/2609.34344) `2609.34344` | 0.30 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 34.2 | [Diffusion Removes Langevin's Conditioning Dependence: A Sharp Gaussian Analysis](https://arxiv.org/abs/2610.12052) `2610.12052` | 0.35 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 34.2 | [Feature Space Adaptation for Effortless Gaussian Process Flows](https://arxiv.org/abs/2610.11459) `2610.11459` | 0.35 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 34.2 | [CARE: Certifying Acceleration for Vision-Language-Action Inference](https://arxiv.org/abs/2610.08917) `2610.08917` | 0.50 | 0.40 | 0.40 | 0.00 | 0.10 | — | — |
| 33.7 | [Bi-FORK: Generative Modeling of High-Dimensional Bifurcating Systems](https://arxiv.org/abs/2610.12449) `2610.12449` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Bilevel optimization for data-driven learning of Koopman embeddings using kernel-based autoencoders](https://arxiv.org/abs/2610.12370) `2610.12370` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Latent Core Tokenizer: Compress, but Meaningfully](https://arxiv.org/abs/2610.12376) `2610.12376` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [SteerablePlex: Can We Steer Full-Duplex Models?](https://arxiv.org/abs/2610.12201) `2610.12201` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Conditional Kernel Stein Discrepancy](https://arxiv.org/abs/2610.11863) `2610.11863` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [SPD-MetaFormer is what you need for small-data brain decoding](https://arxiv.org/abs/2610.10952) `2610.10952` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Foundations of Large Language Models](https://arxiv.org/abs/2501.09223) `2501.09223` | 0.05 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 33.2 | [TokenRouter: Efficient Serving System for Token-Level LLM Routing](https://arxiv.org/abs/2610.12242) `2610.12242` | 0.30 | 0.40 | 0.00 | 1.00 | 0.30 | — | — |
| 33.2 | [Gated Memory: Admission-Controlled Memory Formation for Conversational AI](https://arxiv.org/abs/2610.11270) `2610.11270` | 0.45 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 32.6 | [asdex: Automatic Sparse Differentiation in JAX](https://arxiv.org/abs/2610.12336) `2610.12336` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Spatial Pattern Formation from Multi-Agent Learning in Public Goods Dilemmas](https://arxiv.org/abs/2610.12321) `2610.12321` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Scalable Hierarchical Graph Generation via Soft Community Structure](https://arxiv.org/abs/2610.12163) `2610.12163` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [On-Chain Archaeology of Bitcoin Oracles: Evidence of Use under Limited Observability](https://arxiv.org/abs/2610.11439) `2610.11439` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Testing Algebraic Complete Intersections](https://arxiv.org/abs/2610.12288) `2610.12288` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Credal Machine Learning for Risk-Averse Decision Making](https://arxiv.org/abs/2610.12115) `2610.12115` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [RobustLDS: Learning linear dynamical systems under adversarial corruptions](https://arxiv.org/abs/2610.11906) `2610.11906` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Softmax Attention on Gaussian Mixtures: Linear When It Can, Selective When It Must](https://arxiv.org/abs/2610.11798) `2610.11798` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.1 | [Beyond Resolution: Object-to-Image Ratio Mismatch in Instance Retrieval](https://arxiv.org/abs/2610.11489) `2610.11489` | 0.25 | 0.45 | 0.40 | 0.00 | 0.40 | — | — |
| 31.6 | [All Verdicts are Not Equal: Rethinking LLM Judge Reliability](https://arxiv.org/abs/2610.12083) `2610.12083` | 0.30 | 0.45 | 0.45 | 0.00 | 0.20 | — | — |
| 31.6 | [When History Helps and Hurts: Selective History Use across Multimodal Turns](https://arxiv.org/abs/2610.11948) `2610.11948` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.1 | [Memento 3: Model-Based Recursive Self-Improvement through Reflective Rulebooks](https://arxiv.org/abs/2610.11794) `2610.11794` | 0.30 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 30.5 | [FAITH: Feasibility-Aware Safety-Filtered RL for High-Dimensional Systems](https://arxiv.org/abs/2610.12432) `2610.12432` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Batch Before You Lift: Scalable Topological Deep Learning on Large Graphs](https://arxiv.org/abs/2610.12247) `2610.12247` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [From Sparse Representations to Behavioral Insights for Multimodal Depression Assessment](https://arxiv.org/abs/2610.11787) `2610.11787` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [EVIE: Evidence-Vector-Informed Embeddings for Visual Document Retrieval](https://arxiv.org/abs/2610.11553) `2610.11553` | 0.50 | 0.40 | 0.00 | 0.00 | 0.40 | — | — |
| 30.5 | [Compactness and Consistency: A Conjoint Framework for Deep Graph Clustering](https://arxiv.org/abs/2610.11506) `2610.11506` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Automated Detection of Match Phases in Football from Spatio-Temporal Tracking Data Using Graph Neural Networks](https://arxiv.org/abs/2610.11571) `2610.11571` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Decision-Sufficient Posterior Approximation](https://arxiv.org/abs/2610.11038) `2610.11038` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [MARGIN: Runtime Confidence Calibration for Multi-Agent Foundation Model Coordination](https://arxiv.org/abs/2605.22949) `2605.22949` | 0.30 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 30.0 | [When Should Agents Think? Adaptive Reasoning via Cross-Turn Estimation](https://arxiv.org/abs/2610.12061) `2610.12061` | 0.45 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 30.0 | [Investigating the Role of Reasoning-Language Alignment in Monolingual Retrieval-Augmented Generation](https://arxiv.org/abs/2610.03136) `2610.03136` | 0.30 | 0.45 | 0.30 | 0.00 | 0.30 | — | — |
| 29.5 | [VFold: Symmetry-Aware Cross-Layer Value Cache Compression](https://arxiv.org/abs/2610.12338) `2610.12338` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [ILM: An AI-Powered Storytelling Educational Tool](https://arxiv.org/abs/2610.12064) `2610.12064` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [H2CE: Modeling Geo-Semantic Interactions for POI Reranking with Heterogeneous Two-Stage Cross-Encoders](https://arxiv.org/abs/2610.11277) `2610.11277` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 28.9 | [Ambient Discrete Diffusion: Using the Wrong Data at the Right Time for Data Efficient Learning](https://arxiv.org/abs/2610.12340) `2610.12340` | 0.35 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 28.9 | [SciTBERT: A family of chronologically consistent language models for scientific and technological language processing](https://arxiv.org/abs/2610.12207) `2610.12207` | 0.25 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 28.8 | [REMORY: Learning Residual Memory for Context Compaction](https://arxiv.org/abs/2610.11287) `2610.11287` | 0.00 | 0.45 | 0.30 | 1.00 | 0.16 | — | — |
| 28.4 | [RIT-RAG: Navigating Document Corpora with Retrieval-Induced Trees](https://arxiv.org/abs/2610.11370) `2610.11370` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 27.9 | [SkillContrast: Difference-Guided Text Selection for Agent Skill Reranking](https://arxiv.org/abs/2610.11650) `2610.11650` | 0.25 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 27.9 | [Rare Gate Disagreements Can Limit Plasticity: When Gradient Flow Mispredicts Finite-Batch SGD](https://arxiv.org/abs/2610.11475) `2610.11475` | 0.25 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 27.4 | [Predicting Alignment Generalization with Value Representations](https://arxiv.org/abs/2610.12410) `2610.12410` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Specialized Decision Models vs. General-Purpose LLMs: Benchmarking Jev Across Knowledge, Reasoning, and Multilingual Tasks](https://arxiv.org/abs/2610.11978) `2610.11978` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [MindFlow: Mind Supernet Powered Thinking Flows for Research Idea Innovation](https://arxiv.org/abs/2610.11966) `2610.11966` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 26.8 | [SparseDecoding: Decoding-Aware Pruning for Accurate and Efficient LLM Inference](https://arxiv.org/abs/2610.12327) `2610.12327` | 0.00 | 0.45 | 0.45 | 0.00 | 0.50 | — | — |
| 26.3 | [Prior or Feedback? What an LLM Uses When Adapting Neural Operators](https://arxiv.org/abs/2610.12325) `2610.12325` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Which Skill to Distill? SGUID: Selecting a Compact Skill Bank for Model-Skill Co-Evolution](https://arxiv.org/abs/2610.12367) `2610.12367` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Verdict Without the Rule: Diagnosing and Auditing Regulatory Rule Sensitivity in LLM Compliance Systems](https://arxiv.org/abs/2610.12313) `2610.12313` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [A persistent accuracy ceiling in automated verbal deception detection](https://arxiv.org/abs/2610.12118) `2610.12118` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Can Decision Models Understand Stance? Evaluating Jev Against General-Purpose LLMs](https://arxiv.org/abs/2610.11901) `2610.11901` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Overview of the NTCIR-19 Automatic Evaluation of LLMs 2 (AEOLLM-2) Task](https://arxiv.org/abs/2610.11598) `2610.11598` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [Overcoming Prior Barriers: Supervised Fine-Tuning under Long-Tail Distribution](https://arxiv.org/abs/2610.12345) `2610.12345` | 0.05 | 0.40 | 0.75 | 0.00 | 0.00 | — | — |
| 25.8 | [DiffuPlex: Accelerating Full-Duplex Spoken Dialog Models via Rolling Masked Diffusion](https://arxiv.org/abs/2610.12214) `2610.12214` | 0.35 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 25.8 | [Project Greenhouse: Progress Toward Fully Open and Sovereign Agentic Search](https://arxiv.org/abs/2610.11922) `2610.11922` | 0.25 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 25.3 | [Agentic-TTT: Training test-time policy for test-time training](https://arxiv.org/abs/2610.12002) `2610.12002` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [Not Every Change Is Necessary: Recoverable Drift in Large Language Model Unlearning](https://arxiv.org/abs/2610.11915) `2610.11915` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 24.7 | [CSF: Contextual Safety Filtering for Motion Generators](https://arxiv.org/abs/2610.12467) `2610.12467` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [VioLA: Learning Generalist Humanoid Control Policies from Human Data](https://arxiv.org/abs/2610.12435) `2610.12435` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [What can linear attention learn from nonlinear teachers in-context?](https://arxiv.org/abs/2610.10761) `2610.10761` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [Cited but Not Consulted: A Counterfactual Audit of Legal Chain-of-Thought Faithfulness](https://arxiv.org/abs/2610.12361) `2610.12361` | 0.00 | 0.45 | 0.40 | 0.00 | 0.40 | — | — |
| 24.2 | [OnTrack: Real-Time Monitoring and Intervention in LLM Agent Trajectories via Streaming Structure-Aware Optimal Transport](https://arxiv.org/abs/2610.12375) `2610.12375` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [EgoVoice: Proactive Spoken Assistance from Egocentric Multimodal Streams](https://arxiv.org/abs/2610.12248) `2610.12248` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [When Lower Reconstruction Loss Hurts: Distributionally Robust Refinement for Low-Bit LLM Quantization](https://arxiv.org/abs/2610.11226) `2610.11226` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.6 | [Learning Kilometer-Scale Weather Prediction with Global-Regional Alignment](https://arxiv.org/abs/2610.12401) `2610.12401` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.1 | [AdaptLSTM: Efficient Adaptive Online Learning for Cloud Workload Forecasting under Distribution Drift](https://arxiv.org/abs/2610.12265) `2610.12265` | 0.50 | 0.00 | 0.30 | 0.00 | 0.00 | — | — |
| 21.6 | [Real Long-Term Memory for AI: A 50-Million-Token Window That Is Faster and Cheaper Than Recompute](https://arxiv.org/abs/2610.10845) `2610.10845` | 0.00 | 0.45 | 0.40 | 0.00 | 0.23 | — | — |
| 21.1 | [MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement](https://arxiv.org/abs/2610.11959) `2610.11959` | 0.05 | 0.40 | 0.30 | 0.00 | 0.30 | — | — |
| 21.1 | [Forms of LLM-Integrated Applications from LLM-Chats to Autonomous AI Agent System](https://arxiv.org/abs/2610.11899) `2610.11899` | 0.30 | 0.40 | 0.00 | 0.00 | 0.20 | — | — |
| 20.0 | [Language Models as AI Research World Models](https://arxiv.org/abs/2610.12235) `2610.12235` | 0.00 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 19.5 | [Rounding in Preconditioner Space: Redesigning 4-bit AdamW Optimizer-State Quantization](https://arxiv.org/abs/2610.12444) `2610.12444` | 0.25 | 0.40 | 0.00 | 0.00 | 0.20 | — | — |
| 17.9 | [WOVEN: Weaving Visual World Modeling into Multimodal LLMs](https://arxiv.org/abs/2610.12417) `2610.12417` | 0.00 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 17.9 | [Natural Language to First-Order Logic LLM-based Autoformalization](https://arxiv.org/abs/2610.12030) `2610.12030` | 0.30 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 17.4 | [AdaCast: Conditional Parameter Generation for Adaptive Time Series Forecasting](https://arxiv.org/abs/2610.12240) `2610.12240` | 0.05 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 15.8 | [Long Text to Predictive Features: LLM-Guided Blockwise Feature Engineering via Executable Program Search](https://arxiv.org/abs/2610.12390) `2610.12390` | 0.30 | 0.00 | 0.30 | 0.00 | 0.00 | — | — |
| 15.8 | [HarnessSQL: Harness-Native Training for SQL Agents in Realistic Database Environments](https://arxiv.org/abs/2610.12274) `2610.12274` | 0.00 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [SAIL: Scientific Agentic Intelligence via a Science-Aware Loop](https://arxiv.org/abs/2610.11451) `2610.11451` | 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.