# paper-forge daily report — 2026-10-06

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

## Reproductions

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

## Top candidates

### [LinearPFN: Amortized Variable Selection for Linear Models with Interactions](https://arxiv.org/abs/2610.06580)
`2610.06580` · score 61.1 · analysis: heuristic · feasibility 6/10

Spike-and-slab regression is a standard Bayesian formulation of variable selection: it returns a posterior distribution over which candidate effects are active rather than a single selected subset, so that every candidate effect carries an inclusion probability. Its cost grows exponentially with the number of candidate effects, so the posterior can be enumerated exactly only when the number of predictors is small.

Claimed: AUC 1.0 (abstract (regex extraction, unverified))
Rubric evidence: - pretraining ("pretrained"); + cheap/closed-form ("closed form"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("closed form"); + code: https://github.com/schiekiera/LinearPFN; + keywords: monte carlo

### [Learning to Learn a Language](https://arxiv.org/abs/2610.05879)
`2610.05879` · score 60.4 · analysis: heuristic · feasibility 7/10

We present the Prior-Fitted Language Model (PFLM), a 300M-parameter byte-level transformer pretrained only on samples from a synthetic non-linguistic prior. Given a prefix of real text, it learns to predict the language in context with frozen weights, having never seen a word of any real language.

Rubric evidence: - pretraining ("pretrained"); + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("update rule"); + formal statement ("convergence"); + claims simplicity ("simple"); + code: https://github.com/cbl/prior-fitted-language-model; + 2 HF upvotes

### [The Blind Spot Paradox: When Adaptive Classifiers Defeat Drift Detectors](https://arxiv.org/abs/2610.05853)
`2610.05853` · score 59.5 · analysis: heuristic · feasibility 7/10

Monitoring concept drift from an adaptive classifier's error stream creates an operational conflict with the model's own update loop. When internal adaptation outpaces evidence accumulation, accuracy recovers before cumulative detectors (CUSUM, Page-Hinkley) can reach threshold.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + formal statement ("bound"); + code: https://github.com/TheBlindSpotParadox/TheBlindSpotParadox-Experiments

### [Closing the Context Gap: Activation Alignment for Tabular In-Context Learning](https://arxiv.org/abs/2610.06679)
`2610.06679` · score 57.9 · analysis: heuristic · feasibility 6/10

Tabular foundation models perform in-context learning (ICL) by conditioning predictions on labeled training examples provided as context. Unlike traditional models that separate training from inference, these models must process all training examples in every forward pass, making each prediction expensive.

Rubric evidence: - LLM-based ("foundation model"); + toy/synthetic experiments ("synthetic"); + CPU-friendly ("lightweight"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmark"); + formal statement ("bound"); + code: https://github.com/yoel-zeldes/tabalign

### [HeuFouFT: Task-Guided Metaheuristic Coordinate Search for Fourier Fine-Tuning](https://arxiv.org/abs/2610.06437)
`2610.06437` · score 56.8 · analysis: heuristic · feasibility 7/10

We introduce Heuristic-Guided Fourier Fine-Tuning (HeuFouFT), a task-guided framework for selecting trainable frequency coordinates in Fourier fine-tuning. Existing uniform and Gaussian band-pass schemes allocate a limited spectral budget through fixed, task-agnostic rules.

Rubric evidence: + toy/synthetic experiments ("Simulated"); + CPU-friendly ("lightweight"); + small/synthetic data ("Simulated"); + formal statement ("convergence"); - system/framework (many moving parts) ("framework"); no public code found; + keywords: optimizer, sampling

### [Steering by Influence: Curvature Aware Data Weighting for Activation Steering](https://arxiv.org/abs/2610.06383)
`2610.06383` · score 56.8 · analysis: heuristic · feasibility 6/10

Inference-time steering offers cheap, fine-grained control over a language model's outputs by estimating a concept's representation in activation space and shifting activations towards it. Existing methods build these representations from activation averages over contrastive datasets.

Rubric evidence: + concrete method ("closed-form"); + formal statement ("theorem"); + claims simplicity ("simple"); + code: https://github.com/JDIXON-2/Concept_Activation_Transport; + keywords: ranking

### [Sharpen Without Search: On-Policy Distillation of Sequence-Level Power Distribution](https://arxiv.org/abs/2610.06804)
`2610.06804` · score 52.6 · analysis: heuristic · feasibility 5/10

A language model can give a correct answer more probability than any single incorrect answer and still usually sample an incorrect one, because the incorrect answers together hold more probability. The power distribution raises each complete answer's probability to a power above one and renormalizes, shifting probability toward answers the model finds most likely (sharpening).

Claimed: accuracy 23.0 (abstract (regex extraction, unverified)); accuracy 5.3 (abstract (regex extraction, unverified))
Rubric evidence: + formal statement ("Theorem"); + code: https://github.com/ArminAzizi98/OPPD; + keywords: sampling, monte carlo

### [BazaarBench: Delegation Safety in Decentralized C2C Marketplaces Run by LLM Agents](https://arxiv.org/abs/2610.06748)
`2610.06748` · score 50.0 · analysis: heuristic · feasibility 6/10

In decentralized consumer-to-consumer (C2C) marketplaces, people list goods, negotiate with strangers, and rate one another, so trust rests on reputation. Large language model (LLM) agents now act for users, raising risks to their money, privacy, and reputation.

Rubric evidence: - LLM-based ("LLM"); + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); + public benchmarks ("benchmark"); + code: https://github.com/ziyan-wang98/BazaarBench

### [Global error estimators for parametric monotone nonlinearities and neural approximations](https://arxiv.org/abs/2610.05767)
`2610.05767` · score 49.5 · analysis: heuristic · feasibility 6/10

We construct computable error estimators, which double as loss functions for neural networks, for a class of parametric nonlinear partial differential equations with a monotonicity property, and prove that they are globally reliable and efficient. The value of such a loss function is bounded above and below by the squared error in the natural trial norm, for every trial function, not merely for those near the exact solution; this global property rests on monotonicity.

Rubric evidence: + cheap/closed-form ("closed form"); + concrete method ("closed form"); + formal statement ("bound"); no public code found; + keywords: variational

### [ufakzeka-karar: An Open Turkish Typed-Decision Model with Order-Invariant Option Scoring](https://arxiv.org/abs/2610.06744)
`2610.06744` · score 48.9 · analysis: heuristic · feasibility 5/10

ufakzeka-karar is an open Turkish decision model with 182,494,466 parameters. Given a Turkish text and questions of a fixed answer type (a choice, a level on an ordered scale, or yes or no), it returns a temperature-scaled probability for every option and an expected error that serves as a "not sure" signal, without generating text and in one CPU forward pass for up to ten options.

Rubric evidence: + CPU-friendly ("CPU"); + public benchmarks ("benchmark"); + code: https://github.com/ufakai/ufakzeka-karar

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 61.1 | [LinearPFN: Amortized Variable Selection for Linear Models with Interactions](https://arxiv.org/abs/2610.06580) `2610.06580` | 0.60 | 0.75 | 0.60 | 1.00 | 0.20 | heuristic | True |
| 60.4 | [Learning to Learn a Language](https://arxiv.org/abs/2610.05879) `2610.05879` | 0.40 | 0.75 | 0.90 | 1.00 | 0.16 | heuristic | False |
| 59.5 | [The Blind Spot Paradox: When Adaptive Classifiers Defeat Drift Detectors](https://arxiv.org/abs/2610.05853) `2610.05853` | 0.65 | 0.80 | 0.55 | 1.00 | 0.00 | heuristic | True |
| 57.9 | [Closing the Context Gap: Activation Alignment for Tabular In-Context Learning](https://arxiv.org/abs/2610.06679) `2610.06679` | 0.60 | 0.80 | 0.55 | 1.00 | 0.00 | heuristic | True |
| 56.8 | [HeuFouFT: Task-Guided Metaheuristic Coordinate Search for Fourier Fine-Tuning](https://arxiv.org/abs/2610.06437) `2610.06437` | 0.80 | 0.75 | 0.45 | 0.00 | 0.40 | heuristic | True |
| 56.8 | [Steering by Influence: Curvature Aware Data Weighting for Activation Steering](https://arxiv.org/abs/2610.06383) `2610.06383` | 0.50 | 0.40 | 0.90 | 1.00 | 0.20 | heuristic | True |
| 52.6 | [Sharpen Without Search: On-Policy Distillation of Sequence-Level Power Distribution](https://arxiv.org/abs/2610.06804) `2610.06804` | 0.50 | 0.40 | 0.55 | 1.00 | 0.40 | heuristic | True |
| 50.0 | [BazaarBench: Delegation Safety in Decentralized C2C Marketplaces Run by LLM Agents](https://arxiv.org/abs/2610.06748) `2610.06748` | 0.45 | 0.80 | 0.40 | 1.00 | 0.00 | heuristic | False |
| 49.5 | [Global error estimators for parametric monotone nonlinearities and neural approximations](https://arxiv.org/abs/2610.05767) `2610.05767` | 0.70 | 0.40 | 0.75 | 0.00 | 0.20 | heuristic | True |
| 48.9 | [ufakzeka-karar: An Open Turkish Typed-Decision Model with Order-Invariant Option Scoring](https://arxiv.org/abs/2610.06744) `2610.06744` | 0.65 | 0.45 | 0.40 | 1.00 | 0.00 | heuristic | True |
| 48.9 | [The Surrogate Is Not the Reward: Post-Surrogate Primary-Outcome Acquisition in Contextual Bandits](https://arxiv.org/abs/2610.06610) `2610.06610` | 0.65 | 0.80 | 0.40 | 0.00 | 0.20 | — | — |
| 48.9 | [Isotropic Gaussian Processes Improve Vanilla Bayesian Optimization in High Dimensions](https://arxiv.org/abs/2610.05780) `2610.05780` | 0.65 | 0.80 | 0.40 | 0.00 | 0.20 | — | — |
| 48.9 | [Two-Sample Testing via Path-based Inference](https://arxiv.org/abs/2610.05684) `2610.05684` | 0.65 | 0.80 | 0.55 | 0.00 | 0.00 | — | — |
| 48.9 | [SearchJev: A Fast and Calibrated System-1 Model for Search Agents](https://arxiv.org/abs/2610.05107) `2610.05107` | 0.50 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 47.9 | [Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text-to-Speech Model](https://arxiv.org/abs/2610.06817) `2610.06817` | 0.65 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 47.9 | [Inverse Cross-spectral Neural Networks for Multivariate Time Series](https://arxiv.org/abs/2610.06630) `2610.06630` | 0.65 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 47.9 | [Watermarking: from Impossibility to Auditable Compliance](https://arxiv.org/abs/2610.06317) `2610.06317` | 0.65 | 0.75 | 0.55 | 0.00 | 0.00 | — | — |
| 47.9 | [Towards Looped Models Done Right, Part II: Rethinking at Fixed Points](https://arxiv.org/abs/2610.06833) `2610.06833` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.9 | [Self-Generated Feedback Destabilizes Test-Time Training: A Causal Decomposition of Long-Horizon Adaptation](https://arxiv.org/abs/2610.05076) `2610.05076` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.4 | [ALoDLM: Adaptively Looped Diffusion Language Models](https://arxiv.org/abs/2610.04198) `2610.04198` | 0.35 | 0.45 | 0.55 | 1.00 | 0.30 | — | — |
| 46.8 | [Decoupling Time and Space: A Temporally Conditioned Refinement for EEG Source Imaging](https://arxiv.org/abs/2610.06726) `2610.06726` | 0.65 | 0.75 | 0.50 | 0.00 | 0.00 | — | — |
| 46.3 | [SPRIG: Semantic-ID-enhanced Paths for Knowledge Graph-based Generative Recommendation](https://arxiv.org/abs/2610.06590) `2610.06590` | 0.50 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 46.3 | [Noise Out, Bias In: Targeted Bias Injection in Diffusion Language Models via Closed-Loop Activation Steering](https://arxiv.org/abs/2610.05894) `2610.05894` | 0.35 | 0.40 | 0.55 | 1.00 | 0.30 | — | — |
| 45.8 | [Round-Trip KNN Clustering: multiscale hierarchical cluster detection on directed nearest-neighbour graphs](https://arxiv.org/abs/2610.06795) `2610.06795` | 0.65 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 45.8 | [PlotGround: Grounding Plot Digitization in Real Scientific Figures and Their Source Data](https://arxiv.org/abs/2610.06825) `2610.06825` | 0.65 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 45.8 | [Retrieval-Based In-Context Learning: A Domain Adaptation Framework](https://arxiv.org/abs/2610.05717) `2610.05717` | 0.65 | 0.75 | 0.30 | 0.00 | 0.20 | — | — |
| 44.7 | [H-JEPA: End-to-End Learning of Hierarchical World Models for Visual Planning](https://arxiv.org/abs/2610.06805) `2610.06805` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.2 | [IdeaLens: Detecting AI Ideas in Long-form Writing](https://arxiv.org/abs/2610.06778) `2610.06778` | 0.50 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 43.7 | [Valid Stopping in Adaptive Generator-Verifier Loops](https://arxiv.org/abs/2610.06432) `2610.06432` | 0.65 | 0.80 | 0.30 | 0.00 | 0.00 | — | — |
| 42.1 | [Finding Gaussian Structure in Bosonic States](https://arxiv.org/abs/2610.06810) `2610.06810` | 0.70 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 42.1 | [Constraint-Aware Conversational Job Recommendation in Code-Mixed Low-Resource Settings](https://arxiv.org/abs/2610.05787) `2610.05787` | 0.30 | 0.45 | 0.30 | 1.00 | 0.40 | — | — |
| 42.1 | [Quantifying the Stability of Multi-Step Reasoning via Error Amplification](https://arxiv.org/abs/2610.06404) `2610.06404` | 0.50 | 0.40 | 0.70 | 0.00 | 0.20 | — | — |
| 41.6 | [Revisiting Label-Free Speaker Embedding Enhancement with vMF Profile Likelihood](https://arxiv.org/abs/2610.06691) `2610.06691` | 0.55 | 0.40 | 0.75 | 0.00 | 0.00 | — | — |
| 41.6 | [Learning Steadily: Accumulating Relative Point Margin Scores for Face Image Quality Assessment](https://arxiv.org/abs/2609.31662) `2609.31662` | 0.50 | 0.45 | 0.55 | 0.00 | 0.30 | — | — |
| 41.4 | [TextReg: Mitigating Prompt Distributional Overfitting via Regularized Text-Space Optimization](https://arxiv.org/abs/2605.21318) `2605.21318` | 0.30 | 0.45 | 0.30 | 1.00 | 0.36 | — | — |
| 41.1 | [Balancing Memory Pathways: Analyzing and Improving Memory Utilization in Hybrid LMs](https://arxiv.org/abs/2610.06750) `2610.06750` | 0.50 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 41.1 | [COMPASS 2.0: psychometric representational similarity analysis distinguishes symptom structure from personal signal](https://arxiv.org/abs/2610.06615) `2610.06615` | 0.50 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 41.1 | [OVAL: Output-Aware Local Page Bases for KV Cache Retrieval](https://arxiv.org/abs/2610.06686) `2610.06686` | 0.30 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 41.1 | [SpatialChain: A Benchmark for Auditing Spatial Reasoning Faithfulness in VLMs](https://arxiv.org/abs/2610.06413) `2610.06413` | 0.30 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 40.0 | [OntoInk: Interactive Ontology Visualization, Validation, and Reasoning](https://arxiv.org/abs/2610.05945) `2610.05945` | 0.50 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 40.0 | [On the Sequentially Semiseparable Structure of SDE-Induced Kernels and Efficient Algorithms for Kernel-Based Estimation Problems](https://arxiv.org/abs/2610.05856) `2610.05856` | 0.70 | 0.40 | 0.00 | 0.00 | 0.60 | — | — |
| 39.5 | [OmniConfess: Eliciting Token Confessions to Mitigate Omni-Modal Hallucination](https://arxiv.org/abs/2610.02999) `2610.02999` | 0.30 | 0.45 | 0.40 | 1.00 | 0.10 | — | — |
| 38.9 | [Singular parameters and missing limits in neural PDE solvers](https://arxiv.org/abs/2610.06770) `2610.06770` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [A Solvable Model of Adaptive Learning Rate Rescaling: Acceleration, Stability & Scaling](https://arxiv.org/abs/2610.06701) `2610.06701` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Learning What to Imitate: Entropy-Aware Distribution Mixing](https://arxiv.org/abs/2610.06671) `2610.06671` | 0.70 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 38.9 | [KESurv: A Kernel Ensemble Method for Patient-Specific Survival Prediction](https://arxiv.org/abs/2610.06434) `2610.06434` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [Bradley-Terry model under general comparison graphs](https://arxiv.org/abs/2610.06237) `2610.06237` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [PaLoRA: Paced Low-Rank Adaptation for Continual Learning](https://arxiv.org/abs/2610.04226) `2610.04226` | 0.50 | 0.05 | 0.40 | 1.00 | 0.20 | — | — |
| 38.4 | [Beyond States: Investigating the Effects of Context on User Modeling with Feature-Conditioned Markov Models](https://arxiv.org/abs/2610.06060) `2610.06060` | 0.65 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 38.4 | [Word-Level Text Unmixing via Evidence-Preserving Ownership Routing with Language Models](https://arxiv.org/abs/2610.06603) `2610.06603` | 0.45 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 37.9 | [A Response Theory Probe for Learned Stochastic AI Simulators, Tested on Lorenz-63](https://arxiv.org/abs/2610.06798) `2610.06798` | 0.50 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 37.9 | [Beyond Semantic Similarity: Performance and Costs of Agentic Retrieval for Complex Tasks](https://arxiv.org/abs/2610.05750) `2610.05750` | 0.30 | 0.40 | 0.30 | 1.00 | 0.20 | — | — |
| 36.8 | [Protocol-Sensitive Evaluation of Log Anomaly Detection: Component Costs and Target-Access Sensitivity on HDFS and BGL](https://arxiv.org/abs/2610.05807) `2610.05807` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [OpticalRec: Unified Optical Vision-Language Representation for Multimodal Recommendation](https://arxiv.org/abs/2610.05432) `2610.05432` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Latent Similarity Gaussian Processes: A Theory-Grounded Approach to Personalized Suicide-Risk Forecasting for Clinical Decision-Support](https://arxiv.org/abs/2610.06355) `2610.06355` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [Taylor Representations for Model-Free RL in Networked MDPs](https://arxiv.org/abs/2610.05456) `2610.05456` | 0.50 | 0.45 | 0.55 | 0.00 | 0.00 | — | — |
| 36.3 | [What Gradients Add to Text Leakage in Split Language Models, Counted per Token and per Document](https://arxiv.org/abs/2610.04128) `2610.04128` | 0.50 | 0.40 | 0.40 | 0.00 | 0.23 | — | — |
| 35.8 | [Private online learning and prediction for Littlestone classes](https://arxiv.org/abs/2610.06822) `2610.06822` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [BRANCH-MoE: Balance-Aware Tree Routing for Large Embedding Models](https://arxiv.org/abs/2610.06725) `2610.06725` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Improved Convergence of Large Stepsize Gradient Descent for Logistic Regression](https://arxiv.org/abs/2610.06675) `2610.06675` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [The Birkhoff Geometry of Manifold-Constrained Hyper-Connections: Two Channels, Vertex Viscosity, and Sinkhorn as a Retraction](https://arxiv.org/abs/2610.06653) `2610.06653` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [On the Cardinality of Optimal Representations in the Binary-Source Information Bottleneck](https://arxiv.org/abs/2610.06627) `2610.06627` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Molecules of a Story: Community Detection in PMI-weighted Narrative Networks](https://arxiv.org/abs/2610.06600) `2610.06600` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Do Speech Representations Preserve Regional Accent Across Read and Spontaneous Speech?](https://arxiv.org/abs/2610.06430) `2610.06430` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Optimal compression with quantum retrieval](https://arxiv.org/abs/2610.06702) `2610.06702` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [On the Comparison of Optimizers for Imbalanced Learning](https://arxiv.org/abs/2610.06421) `2610.06421` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Finite-Sample Distribution Theory and Efficient Large-Scale Inference for Online Quantile Regression](https://arxiv.org/abs/2610.05869) `2610.05869` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Gaussian Limits for SGD Without Stationary Moments](https://arxiv.org/abs/2610.05599) `2610.05599` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Moment-Accurate Gaussian Mixtures for Constant-Step Stochastic Approximation](https://arxiv.org/abs/2610.05595) `2610.05595` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [MemPilot: Orchestrating On-Demand Multimodal Memory Curation for LLM Agents](https://arxiv.org/abs/2610.06830) `2610.06830` | 0.30 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 35.3 | [Direct Intermediate Initialization for Tilted Diffusion Samplers](https://arxiv.org/abs/2610.06834) `2610.06834` | 0.35 | 0.45 | 0.40 | 0.00 | 0.40 | — | — |
| 35.2 | [Representation-Space MMD for Diffusion Language Models](https://arxiv.org/abs/2610.06648) `2610.06648` | 0.00 | 0.45 | 0.40 | 1.00 | 0.43 | — | — |
| 34.7 | [Detecting Nighttime Anomalies from NASA Black Marble Using a Generalized Spatio-Temporally Robust Framework of Machine Leaning Ensembles](https://arxiv.org/abs/2610.06674) `2610.06674` | 0.50 | 0.40 | 0.50 | 0.00 | 0.00 | — | — |
| 34.7 | [Reward Stealing Attack on Large Language Models](https://arxiv.org/abs/2610.06670) `2610.06670` | 0.30 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 33.7 | [CLIFT: Conformal Self-Verification for Web Agent Training and Test-Time Scaling](https://arxiv.org/abs/2610.06829) `2610.06829` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Adapting prior-data fitted networks for tabular anomaly detection](https://arxiv.org/abs/2610.06693) `2610.06693` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [The Pushback Paradox: A Two-Probe Diagnostic for Language Model Compliance](https://arxiv.org/abs/2610.06673) `2610.06673` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [What Did the Agent Actually Do? Evidence-Grounded Oversight for Long-Horizon Agents](https://arxiv.org/abs/2610.06406) `2610.06406` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.2 | [RealtimeWAM: One-Step Asynchronous World Action Models](https://arxiv.org/abs/2610.06617) `2610.06617` | 0.00 | 0.45 | 0.40 | 1.00 | 0.30 | — | — |
| 32.9 | [Dynamic Harness Search: Building Multi-Agent Systems Per-Query via Prediction](https://arxiv.org/abs/2610.04137) `2610.04137` | 0.30 | 0.45 | 0.30 | 0.00 | 0.48 | — | — |
| 32.6 | [Deep Learning for Sleep Heart Rate Estimation from Accelerometers: Toward Population-Scale Cardiac Insight Without Optical Sensors](https://arxiv.org/abs/2610.06823) `2610.06823` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [On Learning Optimal Corners in Orthogonal Partially Observable Cooperative Guard Art Galleries](https://arxiv.org/abs/2610.06777) `2610.06777` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Hyperbolic Graph Representation Learning: Embed in One Metric, Optimize with Another](https://arxiv.org/abs/2610.06745) `2610.06745` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Out-of-control Hamiltonian Learning](https://arxiv.org/abs/2610.06709) `2610.06709` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Aligning Multimodal Patient Evidence with Biomedical Knowledge Graphs for Clinical LLMs](https://arxiv.org/abs/2610.06685) `2610.06685` | 0.30 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 32.6 | [How Sparse Probability Maps Shape Mixture-of-Experts Routing](https://arxiv.org/abs/2610.06677) `2610.06677` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [TrustmeWatcher: An Application for Workplace Micro-Sensing and Explainable Well-Being Feedback](https://arxiv.org/abs/2610.06657) `2610.06657` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Considering Context: When World Models Need Context Encoders](https://arxiv.org/abs/2610.06651) `2610.06651` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Language models can notice an impossible engineering problem yet still report it as solved](https://arxiv.org/abs/2610.06668) `2610.06668` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Long-Horizon Textual World Modeling through Structured Reasoning](https://arxiv.org/abs/2610.06637) `2610.06637` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [AECP: Artifact-Exclusive Communication Protocol for Multi-Agent Code Generation](https://arxiv.org/abs/2610.06481) `2610.06481` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Better Call Reward: Reward Hacking as Strategic Abstention in Legal Reasoning Models](https://arxiv.org/abs/2610.06439) `2610.06439` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Commercial Intent in Human-AI Conversations: A Corpus Audit and Architecture for Website Sales Agents](https://arxiv.org/abs/2610.06205) `2610.06205` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Quality-Aware Cross-Model Computation Reuse](https://arxiv.org/abs/2610.05285) `2610.05285` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Reconstruction-Aware Distribution Matching for Inverse Problems with Unpaired Data](https://arxiv.org/abs/2610.06297) `2610.06297` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Global Communication or Graph-Specific Memory?](https://arxiv.org/abs/2610.05874) `2610.05874` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Quantitative Universality of Approximate Message Passing for Rank-One Quadratic Sensing](https://arxiv.org/abs/2610.05327) `2610.05327` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.1 | [Base Models Can Reason By Taking a Cue From Training Data](https://arxiv.org/abs/2610.06851) `2610.06851` | 0.00 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 32.1 | [Generate What You Can Trust: Content Credibility in Generative Recommenders](https://arxiv.org/abs/2610.05670) `2610.05670` | 0.35 | 0.40 | 0.60 | 0.00 | 0.00 | — | — |
| 32.1 | [Kandinsky 6.0 Video: Foundation Models for Synchronized Video and Audio Generation](https://arxiv.org/abs/2610.05608) `2610.05608` | 0.00 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 32.1 | [ASCENT: Online Test-Time Training of Long-Horizon Agents via Self-Distillation of Verified Experience](https://arxiv.org/abs/2610.05303) `2610.05303` | 0.30 | 0.40 | 0.30 | 0.00 | 0.50 | — | — |
| 31.6 | [MatrixFormer: A Foundation Model for Matrix Completion](https://arxiv.org/abs/2610.06751) `2610.06751` | 0.20 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 31.6 | [Programmatic Search Agents: Extending Agentic Search Beyond Query Reformulation](https://arxiv.org/abs/2610.06689) `2610.06689` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [Cut Binary Cross Entropy: Efficient Large-Vocabulary Loss and Gradient Kernels for Sequential Recommendation](https://arxiv.org/abs/2610.05559) `2610.05559` | 0.00 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 31.1 | [SOL: Measuring Gaps between Text Distributions by Double Sliced Wasserstein Metrics](https://arxiv.org/abs/2610.06513) `2610.06513` | 0.35 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 31.1 | [OSWorld-Pro: Process-based Evaluation for Computer Use Agents](https://arxiv.org/abs/2609.24890) `2609.24890` | 0.30 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 30.5 | [Block Disentanglement in CRL: Bridging Identifiability and Visual State Estimation](https://arxiv.org/abs/2610.06809) `2610.06809` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [SAFE-MR: Evidence Sufficiency Learning for Selective Multimodal Rumor Detection](https://arxiv.org/abs/2610.06708) `2610.06708` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [LoGRA: Scaling LLM Reinforcement Learning with Low-Rank Gradient Sketches](https://arxiv.org/abs/2610.06647) `2610.06647` | 0.00 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 30.5 | [Sampling Allocation of LinUCB: Optimal Design Limits in the Small-Gap Regime](https://arxiv.org/abs/2610.06213) `2610.06213` | 0.50 | 0.40 | 0.15 | 0.00 | 0.20 | — | — |
| 30.5 | [Poisson Autoregression on a Large Network with a Stochastic Block Model Structure](https://arxiv.org/abs/2610.05365) `2610.05365` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Anatomy of LLM Sycophancy: What a Flip Rate Hides](https://arxiv.org/abs/2610.06522) `2610.06522` | 0.30 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 30.5 | [Test-Time Adaptation of Reasoning Strategies with Bayesian Nonparametric Memory](https://arxiv.org/abs/2610.06516) `2610.06516` | 0.30 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 30.0 | [Data Unlearning via Inverse Distillation](https://arxiv.org/abs/2609.36099) `2609.36099` | 0.10 | 0.75 | 0.30 | 0.00 | 0.30 | — | — |
| 29.5 | [Mind the Execution Gap: Action-Semantic Mismatch in World-Model Control](https://arxiv.org/abs/2610.06582) `2610.06582` | 0.40 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 29.5 | [Errors of LLM-Assisted Literature Retrieval in Environmental Science: A Comparison Study of Abstract versus Full-text Based Prompts](https://arxiv.org/abs/2610.05690) `2610.05690` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [Training Numerical Intelligence via Auto-Diagnosis and Skill Discovery](https://arxiv.org/abs/2610.03872) `2610.03872` | 0.35 | 0.40 | 0.30 | 0.00 | 0.23 | — | — |
| 28.9 | [DAG-CLIP: A DAG Learning Framework in the Presence of Latent Variables](https://arxiv.org/abs/2610.06086) `2610.06086` | 0.65 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 28.9 | [Periscope: Extending Frozen Language Models Beyond Their Context Window](https://arxiv.org/abs/2610.04047) `2610.04047` | 0.00 | 0.40 | 0.40 | 1.00 | 0.10 | — | — |
| 28.9 | [What Matters for Latent Reasoning with Flow Matching](https://arxiv.org/abs/2610.06666) `2610.06666` | 0.30 | 0.45 | 0.15 | 0.00 | 0.43 | — | — |
| 28.4 | [Synthetic Cultural Agents from Aggregate Anchors](https://arxiv.org/abs/2610.06562) `2610.06562` | 0.40 | 0.75 | 0.00 | 0.00 | 0.00 | — | — |
| 28.4 | [Breaking Bureaucracy: Evaluating open-source LLMs for legal document review](https://arxiv.org/abs/2610.06345) `2610.06345` | 0.00 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 28.4 | [PACMI: Provenance-Aware Cascading Memory Invalidation for Long-Term LLM Agents](https://arxiv.org/abs/2610.05732) `2610.05732` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 27.9 | [T-Search: An Open Agentic Retriever and Playground for Hard Multi-Step Search](https://arxiv.org/abs/2610.06782) `2610.06782` | 0.15 | 0.80 | 0.30 | 0.00 | 0.00 | — | — |
| 27.4 | [Reading the Mood: Emotion-Guided Book-to-Music Recommendation via CGANs and LLMs](https://arxiv.org/abs/2610.06703) `2610.06703` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [Mind the Accent Gap: British Accent Robustness in Speech-Driven Financial Voice Assistants](https://arxiv.org/abs/2610.06587) `2610.06587` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [SCOUT: Supply-Aware Cold-Start Proactive Query Suggestion for Travel Search](https://arxiv.org/abs/2610.05619) `2610.05619` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.2 | [JEPA-TTT: Persistent Test-Time Training of Latent World Models for Planning under Dynamics Shifts](https://arxiv.org/abs/2610.00722) `2610.00722` | 0.25 | 0.40 | 0.40 | 0.00 | 0.16 | — | — |
| 26.3 | [Before Agent Tells The Lie: Has Deception Already Been Represented?](https://arxiv.org/abs/2610.06576) `2610.06576` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Ontology Concept Overlap as a Training Signal: Knowledge-Grounded Reinforcement Learning for Clinical Question Answering](https://arxiv.org/abs/2610.06360) `2610.06360` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.3 | [RAISED: Self-Distillation for Robustness to Prompt Injection in LLM Agents](https://arxiv.org/abs/2610.06401) `2610.06401` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 24.7 | [Differentially Private Mixing of Public Datasets Improves Private Learning](https://arxiv.org/abs/2610.06636) `2610.06636` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [Frozen Factor or Spectral Band? Disentangling Two Choices in Low-Rank LoRA](https://arxiv.org/abs/2610.06621) `2610.06621` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [Domain adaptation of Russian ModernBERT for long legal documents](https://arxiv.org/abs/2610.06715) `2610.06715` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [Learning to Read the Contextual Tokens in Diffusion Transformers](https://arxiv.org/abs/2610.06844) `2610.06844` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [How to scale your HEP ML models: A recipe for robust architecture comparisons at scale](https://arxiv.org/abs/2610.06784) `2610.06784` | 0.00 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [Beyond the Model: The Critical Role of Data Filtering in Clinical Machine Learning](https://arxiv.org/abs/2610.06640) `2610.06640` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 24.2 | [Recursive Video In-Context Learning for Agentic Robot](https://arxiv.org/abs/2610.06843) `2610.06843` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [MedPrune: Topology-Efficient Multimodal Multi-Agent Communication Evolution for Medical VQA Tasks](https://arxiv.org/abs/2610.06695) `2610.06695` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [The Assistance Dilemma: Learning to Teach via Multi-Turn Reinforcement Learning](https://arxiv.org/abs/2610.06446) `2610.06446` | 0.00 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [MATE: Adaptive Long- and Short-Term User Memory for LLM-Based Recommendation](https://arxiv.org/abs/2610.06050) `2610.06050` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 24.2 | [When Are Concept Bottleneck Model Explanations Faithful and Compact?](https://arxiv.org/abs/2610.06285) `2610.06285` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 22.6 | [Behavior-Preserving KV Cache Compression](https://arxiv.org/abs/2610.06479) `2610.06479` | 0.45 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 22.6 | [LiFT: Loop Flow Transformers](https://arxiv.org/abs/2610.05538) `2610.05538` | 0.35 | 0.00 | 0.40 | 0.00 | 0.20 | — | — |
| 21.1 | [MemStrata: 95% and 90.91% Source-Aware Accuracy on LongMemEval-500 and LoCoMo-1540 with a Local Qwen 3.8 27B Q4_K_M Reader](https://arxiv.org/abs/2610.05343) `2610.05343` | 0.00 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 20.2 | [QuantCode Model: Specializing Language Models for Executable Algorithmic Trading Code](https://arxiv.org/abs/2609.39420) `2609.39420` | 0.00 | 0.45 | 0.30 | 0.00 | 0.28 | — | — |
| 20.0 | [Diffusion Transformers are Provably Optimal In-context Generators](https://arxiv.org/abs/2610.05333) `2610.05333` | 0.00 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 17.9 | [Search Engines Never Say No: How Frozen Agents React When the Retrieval Tool Refuses](https://arxiv.org/abs/2610.05348) `2610.05348` | 0.00 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 16.8 | [JEV versus LLMs: Accuracy, Cost and Calibration on Seven Political Science Replications](https://arxiv.org/abs/2610.06625) `2610.06625` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.3 | [Improving Diversity in LLM Short Story Generation](https://arxiv.org/abs/2610.06729) `2610.06729` | 0.05 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 16.3 | [From Pixels, Without Pre-training: Joint Generative and Self-Supervised Representation Learning in One Model](https://arxiv.org/abs/2610.05711) `2610.05711` | 0.25 | 0.00 | 0.40 | 0.00 | 0.00 | — | — |
| 14.7 | [Wikidata Search Traces: A Dataset for Training Knowledge Graph Search Agents](https://arxiv.org/abs/2610.06650) `2610.06650` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 14.7 | [G-CARB: Graph-Localized Conformal Agent Risk Budget for Compositional Harm](https://arxiv.org/abs/2610.05563) `2610.05563` | 0.00 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 10.0 | [To Learn is to Wander: Learning Across Graphs and Tasks with Random Walks](https://arxiv.org/abs/2610.06694) `2610.06694` | 0.05 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |

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