# paper-forge daily report — 2026-10-15

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

## Reproductions

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

## Top candidates

### [Neural Wasserstein Splines: Representation-Invariant Second-Order Priors for Trajectory Inference](https://arxiv.org/abs/2610.19878)
`2610.19878` · score 75.3 · analysis: heuristic · feasibility 8/10

Population dynamics are often observed only as unpaired snapshots, which leave the evolution between measurements underdetermined. Second-order regularization can constrain it, but a penalty on the accelerations of sampled particles is a prior on joint particle paths, and different particle representations of the same marginal curve can have different acceleration costs.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + explicit algorithm/pseudocode ("Algorithm 1"); + concrete method ("closed form"); + formal statement ("we prove"); + claims simplicity ("Simple"); + code: https://github.com/amartya21/Neural-Wasserstein-Splines.git; + keywords: regularization, ranking

### [Fractional is Better: Learnable Derivative Orders in Neural Operator Learning](https://arxiv.org/abs/2610.20363)
`2610.20363` · score 64.2 · analysis: heuristic · feasibility 7/10

Neural operators learn mappings between function spaces, enabling fast surrogate solutions to partial differential equations. Despite remarkable architectural diversity, these methods often share a common input representation: raw coordinate-value pairs $(x, u(x))$.

Rubric evidence: + cheap/closed-form ("closed form"); + public benchmarks ("benchmark"); + concrete method ("closed form"); + formal statement ("We prove"); + claims simplicity ("simple"); + code: https://github.com/FaresBMehouachi/delNO; + keywords: regularization

### [Testing Conditional Distribution Equality Using Generative Models](https://arxiv.org/abs/2610.20447)
`2610.20447` · score 58.4 · analysis: heuristic · feasibility 8/10

Conditional distribution equality testing is a core problem with applications in causal inference, model diagnostics, and transfer learning. Existing methods often lose power when the conditioning variable is high-dimensional.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + concrete method ("We propose a new"); + formal statement ("Theorem"); + claims simplicity ("straightforward"); no public code found; + keywords: kernel

### [Continuum Attention](https://arxiv.org/abs/2610.20586)
`2610.20586` · score 56.2 · analysis: heuristic · feasibility 6/10

We introduce Continuum Attention (CA), a causal Transformer attention operator that replaces discrete key and value lookup with integration over a continuous, piecewise linear memory curve. Each query induces an exponential relevance density along the curve.

Rubric evidence: + cheap/closed-form ("closed form"); + concrete method ("closed form"); + code: https://github.com/akhadangi/continuum-attention; + 2 HF upvotes

### [Recovering Reliable Supervision via Critique-Guided On-Policy Distillation](https://arxiv.org/abs/2610.20572)
`2610.20572` · score 55.3 · analysis: heuristic · feasibility 6/10

On-policy distillation (OPD) trains a student model on its own rollouts with token-level supervision from a teacher, assuming that the teacher can reach a correct answer from any prefix the student writes. We call this property the recoverability of the teacher and show it fails on incorrect rollouts: the teacher follows student errors, although it recognizes them.

Rubric evidence: + public benchmarks ("benchmarks"); + explicit algorithm/pseudocode ("Algorithm 1"); + code: https://github.com/kzhao5/CriticOPD; + 61 HF upvotes

### [The Accuracy Well: Nonmonotone Accuracy Dynamics at Vanishing Initialization](https://arxiv.org/abs/2610.20439)
`2610.20439` · score 54.2 · analysis: heuristic · feasibility 7/10

We study classification dynamics under small initialization and identify a pronounced non-monotone behavior of accuracy that we call the accuracy well: accuracy rises rapidly early in training, subsequently drops, and only much later recovers. We observe this phenomenon in deep linear networks, fully connected ReLU networks on MNIST, deep convolutional networks on CIFAR-10, and controlled synthetic tasks, across different losses and optimizers and in both training and test accuracy.

Claimed: accuracy 100.0% (abstract (regex extraction, unverified))
Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("MNIST"); + formal statement ("we prove"); + claims simplicity ("simple"); no public code found; + keywords: optimizer

### [A Large-Dimensional Analysis of Least-Squares Federated Learning via Generalized Total Variation Minimization](https://arxiv.org/abs/2610.20075)
`2610.20075` · score 54.2 · analysis: heuristic · feasibility 7/10

Federated learning over decentralized networks inherently trades off the suppression of local finite-sample noise through collaboration against the bias that collaboration introduces between heterogeneous nodes. We study this tension by performing a large-dimensional analysis of Generalized Total Variation Minimization with least-squares models.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + formal statement ("convergence"); + claims simplicity ("straightforward"); no public code found; + keywords: regularization

### [MCMC Can See Can't See](https://arxiv.org/abs/2610.20020)
`2610.20020` · score 52.6 · analysis: heuristic · feasibility 5/10

It is folklore that passing "convergence" diagnostics does not guarantee that a Markov chain has converged. We prove this as an impossibility for any finite checklist of diagnostics.

Rubric evidence: + formal statement ("convergence"); + claims simplicity ("simple"); + code: https://github.com/BaroqueObama/mcmc-impossibility-code; + keywords: sampling

### [What Does Disagreement Mean? A Semantic Identifiability Frontier for Pairwise Preference Learning](https://arxiv.org/abs/2610.20450)
`2610.20450` · score 51.1 · analysis: heuristic · feasibility 7/10

Pairwise comparison data are a standard supervision signal in preference learning, but disagreement in such data can reflect stable latent heterogeneity, ephemeral stochasticity, or non-binary responses hidden by a forced-choice interface. We study which of these semantic distinctions are identifiable from the feedback channel itself.

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

### [MA-Bench: Measuring Operational Planning in Language Models](https://arxiv.org/abs/2610.19138)
`2610.19138` · score 47.9 · analysis: heuristic · feasibility 6/10

How well can a language model turn a large, incomplete and partly contradictory body of documents into a plan when no single correct answer exists? We introduce Mission Analysis (MA)-Bench, built on mission analysis, the first step of military operational planning, in which a staff reads the full situation and states the problem before anyone proposes a solution.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); no public code found; + keywords: ranking

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 75.3 | [Neural Wasserstein Splines: Representation-Invariant Second-Order Priors for Trajectory Inference](https://arxiv.org/abs/2610.19878) `2610.19878` | 0.65 | 0.80 | 1.00 | 1.00 | 0.40 | heuristic | True |
| 64.2 | [Fractional is Better: Learnable Derivative Orders in Neural Operator Learning](https://arxiv.org/abs/2610.20363) `2610.20363` | 0.70 | 0.45 | 0.90 | 1.00 | 0.20 | heuristic | True |
| 58.4 | [Testing Conditional Distribution Equality Using Generative Models](https://arxiv.org/abs/2610.20447) `2610.20447` | 0.65 | 0.75 | 0.90 | 0.00 | 0.20 | heuristic | True |
| 56.2 | [Continuum Attention](https://arxiv.org/abs/2610.20586) `2610.20586` | 0.70 | 0.40 | 0.60 | 1.00 | 0.16 | heuristic | True |
| 55.3 | [Recovering Reliable Supervision via Critique-Guided On-Policy Distillation](https://arxiv.org/abs/2610.20572) `2610.20572` | 0.50 | 0.45 | 0.70 | 1.00 | 0.30 | heuristic | True |
| 54.2 | [The Accuracy Well: Nonmonotone Accuracy Dynamics at Vanishing Initialization](https://arxiv.org/abs/2610.20439) `2610.20439` | 0.65 | 0.75 | 0.70 | 0.00 | 0.20 | heuristic | True |
| 54.2 | [A Large-Dimensional Analysis of Least-Squares Federated Learning via Generalized Total Variation Minimization](https://arxiv.org/abs/2610.20075) `2610.20075` | 0.65 | 0.75 | 0.70 | 0.00 | 0.20 | heuristic | True |
| 52.6 | [MCMC Can See Can't See](https://arxiv.org/abs/2610.20020) `2610.20020` | 0.50 | 0.40 | 0.70 | 1.00 | 0.20 | heuristic | True |
| 51.1 | [What Does Disagreement Mean? A Semantic Identifiability Frontier for Pairwise Preference Learning](https://arxiv.org/abs/2610.20450) `2610.20450` | 0.65 | 0.75 | 0.70 | 0.00 | 0.00 | heuristic | True |
| 47.9 | [MA-Bench: Measuring Operational Planning in Language Models](https://arxiv.org/abs/2610.19138) `2610.19138` | 0.65 | 0.75 | 0.40 | 0.00 | 0.20 | heuristic | True |
| 47.9 | [Latent Compaction with Test-Time Training](https://arxiv.org/abs/2610.20582) `2610.20582` | 0.50 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 47.4 | [Conformalized Physics-Informed Weak Regression for Parameter Estimation in Epidemic Models](https://arxiv.org/abs/2610.19058) `2610.19058` | 0.70 | 0.45 | 0.60 | 0.00 | 0.20 | — | — |
| 46.8 | [Optimizing Sparse Boolean Rule Networks using Mixed Integer Linear Programming](https://arxiv.org/abs/2610.20170) `2610.20170` | 0.65 | 0.80 | 0.45 | 0.00 | 0.00 | — | — |
| 46.3 | [Co-Evolving Acting and Forecasting for Language Agents](https://arxiv.org/abs/2610.20597) `2610.20597` | 0.40 | 0.80 | 0.30 | 1.00 | 0.00 | — | — |
| 45.8 | [Noisy-Label Calibration with Multiple Unreliable Labels](https://arxiv.org/abs/2610.19543) `2610.19543` | 0.65 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 45.8 | [Seeing, Maintaining, and Learning: The Evolution of Multimodal Agent Memory](https://arxiv.org/abs/2610.19366) `2610.19366` | 0.50 | 0.40 | 0.30 | 1.00 | 0.30 | — | — |
| 45.3 | [The Physics Is the Network: Fast, Training-Free Neural Emulation](https://arxiv.org/abs/2610.20525) `2610.20525` | 0.80 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 45.3 | [Convergence Rates of Gaussian Process Thompson Sampling for Efficient Global Optimization](https://arxiv.org/abs/2610.19925) `2610.19925` | 0.50 | 0.40 | 0.70 | 0.00 | 0.40 | — | — |
| 45.3 | [Large-Step Last-Iterate Convergence of Random-Reshuffling Stochastic Gradient Descent](https://arxiv.org/abs/2610.19444) `2610.19444` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 44.7 | [Continual Learning for Reliable In-Sequence Molecular Communication](https://arxiv.org/abs/2610.20426) `2610.20426` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [From Human Corrections to Learned Segmentations: A Critical Mathematical Review of Information, Learning, and Human Effort](https://arxiv.org/abs/2610.20396) `2610.20396` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Repairing Speech Access After Text Updates with Frozen Affine Fields](https://arxiv.org/abs/2610.20253) `2610.20253` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [From Global to Tail Convergence: Dynamics of $α$-Stable Score-Based Generative Models](https://arxiv.org/abs/2610.19773) `2610.19773` | 0.35 | 0.40 | 0.55 | 1.00 | 0.20 | — | — |
| 44.7 | [Share the Symbol, Not the Vector: Training-Free Fact Communication Across Language Models](https://arxiv.org/abs/2610.19696) `2610.19696` | 0.50 | 0.40 | 0.40 | 1.00 | 0.10 | — | — |
| 44.2 | [VASPO: Self-Distillation with Privileged Value Estimation](https://arxiv.org/abs/2610.20516) `2610.20516` | 0.50 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 44.2 | [Joint Subspace Sparse Attention for Efficient Long-Context Decoding](https://arxiv.org/abs/2610.19026) `2610.19026` | 0.00 | 0.40 | 0.75 | 1.00 | 0.60 | — | — |
| 43.6 | [NavGPT-3: Harnessing Context in a Hierarchical Navigation Runtime](https://arxiv.org/abs/2610.10787) `2610.10787` | 0.50 | 0.40 | 0.30 | 1.00 | 0.16 | — | — |
| 43.2 | [Lost in Translation: Why Lower Bias Scores Don't Mean Fairer Models](https://arxiv.org/abs/2610.20479) `2610.20479` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 43.2 | [Outperforming 1000 Euclidean Layers With Hyperbolic Goal-Conditioned RL](https://arxiv.org/abs/2610.20009) `2610.20009` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 42.9 | [Diagnosing and Correcting Reward Hacking in Verifiable Instruction Following](https://arxiv.org/abs/2610.19525) `2610.19525` | 0.65 | 0.45 | 0.40 | 0.00 | 0.28 | — | — |
| 42.6 | [Cascaded Retrieval with Cross-Modal Query Refinement for Efficient Visual Document Search](https://arxiv.org/abs/2610.20367) `2610.20367` | 0.65 | 0.45 | 0.30 | 0.00 | 0.40 | — | — |
| 42.6 | [Gap Detection: Mitigating Homogenization through Demand and Defensibility](https://arxiv.org/abs/2610.20162) `2610.20162` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.6 | [Multi-Sensitive Distributional Fairness: Joint Wasserstein Audits and Shapley Allocation](https://arxiv.org/abs/2610.19252) `2610.19252` | 0.65 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 42.1 | [High-Probability Regret Bounds for Feel-Good Thompson Sampling](https://arxiv.org/abs/2610.19513) `2610.19513` | 0.50 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 41.6 | [KV4-RL: Unlocking Faster Asynchronous Rollouts for RLVR with INT4 KV Cache Quantization](https://arxiv.org/abs/2610.14448) `2610.14448` | 0.50 | 0.45 | 0.40 | 0.00 | 0.50 | — | — |
| 41.6 | [The Thinking Leakage in Reinforcement Learning for LLMs](https://arxiv.org/abs/2610.14270) `2610.14270` | 0.30 | 0.40 | 0.40 | 1.00 | 0.30 | — | — |
| 41.6 | [BigCell: Generating Gigapixel Whole-Slide Images](https://arxiv.org/abs/2610.14261) `2610.14261` | 0.45 | 0.75 | 0.40 | 0.00 | 0.20 | — | — |
| 41.1 | [FACES: Fine-grained Gated Semantic Alignment for Multimodal Emotional Support Conversation](https://arxiv.org/abs/2610.20051) `2610.20051` | 0.50 | 0.40 | 0.30 | 1.00 | 0.00 | — | — |
| 41.1 | [One Size Does Not Fit All: Curvature-Aware Adaptive Perturbation for SAM](https://arxiv.org/abs/2610.20415) `2610.20415` | 0.70 | 0.35 | 0.55 | 0.00 | 0.00 | — | — |
| 40.9 | [Long-Horizon RL Needs Muon](https://arxiv.org/abs/2610.20256) `2610.20256` | 0.50 | 0.40 | 0.30 | 0.00 | 0.66 | — | — |
| 40.5 | [$τ$-Emotion-Bench: Exposing the Interaction Fragility of LLM Agents through Real-World User Emotions](https://arxiv.org/abs/2610.19560) `2610.19560` | 0.30 | 0.45 | 0.30 | 1.00 | 0.30 | — | — |
| 39.5 | [PerSo: Investigating Facets of LLMs' Social Intelligence in Real-World Persuasion Scenarios](https://arxiv.org/abs/2610.20548) `2610.20548` | 0.45 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 39.5 | [Efficient contextual partial monitoring with general function approximation](https://arxiv.org/abs/2610.19804) `2610.19804` | 0.45 | 0.75 | 0.30 | 0.00 | 0.20 | — | — |
| 38.9 | [Tracking a Small Set of Experts in Online Convex Optimization](https://arxiv.org/abs/2610.20416) `2610.20416` | 0.70 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 38.9 | [Posterior Frequencies and the Cost of Not Fitting: A Correction to a Memorization Lower Bound](https://arxiv.org/abs/2610.20402) `2610.20402` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Reading the Neighborhood: Text-Attentive Graph Reranking for Automated Subject Indexing](https://arxiv.org/abs/2610.20048) `2610.20048` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [Minimax estimation of the pseudo-spectral gap of continuous Markov chains, with applications to empirical concentration bounds](https://arxiv.org/abs/2610.20007) `2610.20007` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.4 | [Detection of Potato Plant Leaf Diseases using Convolutional Neural Networks](https://arxiv.org/abs/2610.20544) `2610.20544` | 0.65 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 38.4 | [State-Matched Sequence Conformal Prediction](https://arxiv.org/abs/2610.19484) `2610.19484` | 0.65 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 38.4 | [Subdata Selection for Measurement-Constrained Linear Regression under Model Misspecification](https://arxiv.org/abs/2610.19351) `2610.19351` | 0.65 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 38.4 | [Text-Informed Training Data Filtering for Enhanced Virtual Flow Meter Performance](https://arxiv.org/abs/2610.19860) `2610.19860` | 0.45 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 37.4 | [Price of Anarchy of First-Price Auctions in Agentic Markets](https://arxiv.org/abs/2610.20587) `2610.20587` | 0.65 | 0.75 | 0.05 | 0.00 | 0.00 | — | — |
| 37.4 | [XViewRank: Overlap-Conditioned Score Fusion for LLM-Augmented Sparse Retrieval](https://arxiv.org/abs/2610.20122) `2610.20122` | 0.45 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 37.4 | [Annealed Lévy Diffusion via Gaussian Surrogate Reverse Dynamics](https://arxiv.org/abs/2610.19692) `2610.19692` | 0.35 | 0.80 | 0.30 | 0.00 | 0.20 | — | — |
| 37.4 | [Student Copula Markov Networks](https://arxiv.org/abs/2610.19135) `2610.19135` | 0.65 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 36.8 | [The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free Adaptation](https://arxiv.org/abs/2610.20456) `2610.20456` | 0.25 | 0.40 | 0.40 | 1.00 | 0.10 | — | — |
| 36.8 | [Equal CER, Unequal Damage: Severity-Weighted Character Error Rate for OCR](https://arxiv.org/abs/2610.20461) `2610.20461` | 0.30 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 36.8 | [RCER: Reconstructing Cross-Evidence Relations for Iterative Retrieval-Augmented Generation](https://arxiv.org/abs/2610.19398) `2610.19398` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [SpectrumQuant: Spectrum Alignment for LLM Quantization](https://arxiv.org/abs/2610.20303) `2610.20303` | 0.50 | 0.40 | 0.60 | 0.00 | 0.00 | — | — |
| 36.8 | [Projected Wasserstein-Fisher-Rao Flows for Hawkes Background-Intensity Estimation](https://arxiv.org/abs/2610.19588) `2610.19588` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.7 | [Are We Testing Too Much? A Survey of Benchmark Compression for Efficient Evaluation of Large Language Models](https://arxiv.org/abs/2610.18248) `2610.18248` | 0.30 | 0.45 | 0.00 | 1.00 | 0.46 | — | — |
| 35.8 | [Revisiting Simple Baselines for Relational Deep Learning](https://arxiv.org/abs/2610.20534) `2610.20534` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Hindsight-Privileged RLVR: Single-Rollout Token Credit Without Group Sampling](https://arxiv.org/abs/2610.20497) `2610.20497` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Average-Reward Policy Identification from a Single Continuing Trajectory](https://arxiv.org/abs/2610.20486) `2610.20486` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [SafeAdapt-Poker: Counterfactual Risk Calibration for Budgeted Adaptation under Opponent Shift](https://arxiv.org/abs/2610.20473) `2610.20473` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [HRTF Upsampling and Personalization via Neural Operators](https://arxiv.org/abs/2610.20400) `2610.20400` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Hypothesis Selection via Weighted Medians](https://arxiv.org/abs/2610.20375) `2610.20375` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [MMLA-RSI: Recursive Self-Improvement through RTT and RTMA](https://arxiv.org/abs/2610.20294) `2610.20294` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [The minimal width of neural networks and immersions of real projective space](https://arxiv.org/abs/2610.20229) `2610.20229` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [On the Optimisation Landscape of Overparameterised Weighted MMD](https://arxiv.org/abs/2610.19861) `2610.19861` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [On Global Joint Lower Bounds in Zeroth-Order Nonconvex Optimization](https://arxiv.org/abs/2610.19652) `2610.19652` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [CSWBench: Evaluating Context-Sensitive Code-Switching in Creative Writing](https://arxiv.org/abs/2610.20272) `2610.20272` | 0.30 | 0.45 | 0.30 | 1.00 | 0.00 | — | — |
| 35.3 | [Hyperspherical Contrastive Representations Unlock Planning and Exploration](https://arxiv.org/abs/2610.20409) `2610.20409` | 0.45 | 0.40 | 0.60 | 0.00 | 0.00 | — | — |
| 35.3 | [Neural Supervised Clustering for Learning Individualized Treatment Rules with Many Treatments](https://arxiv.org/abs/2610.19104) `2610.19104` | 0.65 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 35.3 | [SwarmRSI: Recursive Self-Improvement through Evolving Scientific Agent Swarm](https://arxiv.org/abs/2610.20431) `2610.20431` | 0.50 | 0.40 | 0.30 | 0.00 | 0.30 | — | — |
| 35.3 | [Taking the PULSE of Medical Reasoning: Step-Level Retrieval Timing for Knowledge-Grounded Inference](https://arxiv.org/abs/2610.20292) `2610.20292` | 0.45 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 35.3 | [Structure-aware Bayesian full waveform inversion via diffusion-regularized Stein variational gradient descent](https://arxiv.org/abs/2610.19187) `2610.19187` | 0.35 | 0.40 | 0.45 | 0.00 | 0.40 | — | — |
| 35.3 | [LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators](https://arxiv.org/abs/2610.17246) `2610.17246` | 0.15 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 34.7 | [Bridging the Topological Divide: Dual-Branch Graph Distillation with Semantic Priors for Robust Cold-Start Recommendation](https://arxiv.org/abs/2610.19578) `2610.19578` | 0.50 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 34.7 | [Function-Space Temporal Self-Distillation for Policy Optimization](https://arxiv.org/abs/2610.19687) `2610.19687` | 0.50 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 34.2 | [Embedding Geometry Shapes LLM Routing: Tracing Failure Modes and Improving Generalization](https://arxiv.org/abs/2610.19198) `2610.19198` | 0.45 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 33.7 | [Learnable Hodge-Laplacian Driven Mixture of Experts for Steady-PDE Surrogates on Coarse Unstructured Meshes](https://arxiv.org/abs/2610.20585) `2610.20585` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [A Two-Level Clinical Novelty Prioritization Framework for Disease Trajectories Discovered from Longitudinal Electronic Health Records](https://arxiv.org/abs/2610.20512) `2610.20512` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Beyond Expert Rankings: Contrast Calibration for Mixed-Precision MoE Quantization](https://arxiv.org/abs/2610.20353) `2610.20353` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Non-asymptotic Convergence of Average-reward Q-learning with Options](https://arxiv.org/abs/2610.20347) `2610.20347` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [When Occupation-derived Skills Don't Help: On the Limits of Skill-Augmented Career Path Prediction](https://arxiv.org/abs/2610.20186) `2610.20186` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [MultiPartyTurnBench: Benchmarking Floor Control in Multi-Party Spoken Dialogue](https://arxiv.org/abs/2610.20172) `2610.20172` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Creating SquashFS Images to Support HPC and AI Datasets](https://arxiv.org/abs/2610.20578) `2610.20578` | 0.50 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 33.7 | [Weighted Adjoint Matching: Finetuning Flow Matching Models with Time-Dependent Control Cost](https://arxiv.org/abs/2610.20035) `2610.20035` | 0.10 | 0.75 | 0.55 | 0.00 | 0.20 | — | — |
| 33.7 | [On Inherent Privacy of Posterior Sampling: A Unified Rényi-Divergence Framework](https://arxiv.org/abs/2610.19573) `2610.19573` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.2 | [Hierarchical Gaussian Processes for Survey Weighting: Local Equivalent Weights in Structured Populations](https://arxiv.org/abs/2610.20321) `2610.20321` | 0.65 | 0.40 | 0.05 | 0.00 | 0.20 | — | — |
| 32.6 | [ActiveEvidence: Decision-Focused Evidence Graphs with Verifiable Acquisition Gaps](https://arxiv.org/abs/2610.20532) `2610.20532` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Bayesian Learning to Select Fair Classifiers](https://arxiv.org/abs/2610.20514) `2610.20514` | 0.30 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 32.6 | [Decoupled Marginal Sharpening for Training-Free Inference-Time Scaling](https://arxiv.org/abs/2610.20483) `2610.20483` | 0.30 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 32.6 | [Selective Growing Q-Networks (SGQN): An Analysis of Adaptive Action Grids in Decoupled Q-Learning](https://arxiv.org/abs/2610.20463) `2610.20463` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Training Language Models to Fail Gracefully](https://arxiv.org/abs/2610.20401) `2610.20401` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Do AI Agents Gossip? Evaluative Talk in Human and AI-Agent Online Communities](https://arxiv.org/abs/2610.20360) `2610.20360` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Why Generate a Click? Plug-in Decision Heads for Web Agents](https://arxiv.org/abs/2610.20107) `2610.20107` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [AdCoEvo: A Co-Evolving Multi-Agent Advertising Delivery System](https://arxiv.org/abs/2610.19454) `2610.19454` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [What Information Buys in Active Inference](https://arxiv.org/abs/2610.19046) `2610.19046` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.1 | [FOLD: Position-Free Far Memory with Local Order for Long-Context Speculative Decoding](https://arxiv.org/abs/2610.20392) `2610.20392` | 0.45 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 31.6 | [OrionMoE-v1.0: A Mixture-of-Experts Foundation Model for Tabular In-Context Learning](https://arxiv.org/abs/2610.20418) `2610.20418` | 0.20 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 31.6 | [ReaDy: Recurrent States as Dynamic Context for Speculative Drafting in Hybrid Linear Attention Large Language Models](https://arxiv.org/abs/2610.20503) `2610.20503` | 0.00 | 0.45 | 0.40 | 1.00 | 0.20 | — | — |
| 31.6 | [TreeFlex: Agents Navigating Their Own Search Trees](https://arxiv.org/abs/2610.20334) `2610.20334` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [TreeSpark: Learning to Draft the Trees You Verify for Speculative Decoding](https://arxiv.org/abs/2610.20175) `2610.20175` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [RM-OPD: On-Policy Distillation for Generative Reward Models](https://arxiv.org/abs/2610.20174) `2610.20174` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [MAEM: Mitigating Recency Bias via a Replay-Free, Capacity-Controlled Dynamic Mixture of Experts for Class-Incremental Graph Learning](https://arxiv.org/abs/2610.20238) `2610.20238` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [SNACK: A Sequential Notation Framework for Probabilistic Graph Generation](https://arxiv.org/abs/2610.19781) `2610.19781` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.1 | [When Does Momentum Help in Reinforcement Learning with Verifiable Rewards?](https://arxiv.org/abs/2610.20413) `2610.20413` | 0.15 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 31.1 | [Asymptotically Exact Negative Guidance of Diffusion Models via Positive-Unlabeled Learning](https://arxiv.org/abs/2610.19802) `2610.19802` | 0.25 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 31.1 | [Discrete Diffusion Models are Minimax Optimal Under Sparsity](https://arxiv.org/abs/2610.19020) `2610.19020` | 0.35 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 31.1 | [More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement Learning](https://arxiv.org/abs/2610.19789) `2610.19789` | 0.00 | 0.45 | 0.30 | 1.00 | 0.30 | — | — |
| 30.9 | [StartLux-Decision: AutoThink Decision Models with Reinforcement Learning for Calibrated Deliberative Decisions](https://arxiv.org/abs/2610.17309) `2610.17309` | 0.00 | 0.45 | 0.40 | 1.00 | 0.16 | — | — |
| 30.5 | [Hierarchical Residual Policy Learning for Real-World Mobile Manipulation with Sparse Human Guidance](https://arxiv.org/abs/2610.20607) `2610.20607` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [One Task, Many Interfaces: Simple Cross-Interface Consistency Training Improves LLM Agent Robustness](https://arxiv.org/abs/2610.20576) `2610.20576` | 0.30 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 30.5 | [VascuSR: Iterative Anatomy-Aware Diffusion for CT Super-Resolution](https://arxiv.org/abs/2610.20509) `2610.20509` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Graph-Enhanced Disease Trajectory Discovery and Temporal Endpoint Forecasting from Longitudinal Electronic Health Records](https://arxiv.org/abs/2610.20502) `2610.20502` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [NeuroDETR: Multiscale Discrete Spatiotemporal Tokenization for Asynchronous EEG Event Detection](https://arxiv.org/abs/2610.20349) `2610.20349` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Generalizing Identifiability Results in Causal Discovery from Complete Data to Missing Data Settings](https://arxiv.org/abs/2610.19211) `2610.19211` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [What Makes a Social Agent's Interactions Effective? Learning from Verifiable Outcomes in Paired Worlds](https://arxiv.org/abs/2610.20195) `2610.20195` | 0.30 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [Scaling Discovery Requires Forgetting](https://arxiv.org/abs/2610.20563) `2610.20563` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [ChronoCulture: Evaluating Cultural Awareness Across Time in Large Language Models](https://arxiv.org/abs/2610.20472) `2610.20472` | 0.30 | 0.45 | 0.00 | 1.00 | 0.00 | — | — |
| 29.5 | [Analyzing performance of Retrieval Augmented Generation models for Question Answering Tasks](https://arxiv.org/abs/2610.20419) `2610.20419` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [From Triage to Impact: A Multi-Hop Reasoning Benchmark for Explainable Persian Hate Speech](https://arxiv.org/abs/2610.20370) `2610.20370` | 0.30 | 0.05 | 0.40 | 1.00 | 0.00 | — | — |
| 29.5 | [Which Knowledge Should LLMs Trust? Uncovering a Generalizable Steering Direction Across Knowledge Conflicts](https://arxiv.org/abs/2610.20067) `2610.20067` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 29.5 | [Calibration and Risk-Budgeted Decisions with System One Models](https://arxiv.org/abs/2610.19811) `2610.19811` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 28.9 | [From Plato to Aristotle: Toward a General Theory of Representation Hypotheses](https://arxiv.org/abs/2610.20286) `2610.20286` | 0.25 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 28.9 | [Mixed Poisson diffusion models and application to posterior sampling](https://arxiv.org/abs/2610.20218) `2610.20218` | 0.35 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 28.4 | [Lifted State Policy Optimization: Internalizing Self-Correction through Transition Geometry](https://arxiv.org/abs/2610.20350) `2610.20350` | 0.00 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 27.9 | [Maximum likelihood, neural networks and posterior mean for inferring fractional Brownian motion parameters](https://arxiv.org/abs/2610.19888) `2610.19888` | 0.35 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Risk-Controlled DFT Selection for Machine-Learned Interatomic Potentials](https://arxiv.org/abs/2610.19630) `2610.19630` | 0.40 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 27.4 | [Anuvani-LID: Robust and Calibrated Spoken Language Identification for Indian Languages](https://arxiv.org/abs/2610.20263) `2610.20263` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Cross-Model In-Context Neurofeedback Reveals (Limited) Activation Control in LLMs](https://arxiv.org/abs/2610.20179) `2610.20179` | 0.00 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 27.4 | [Sparse Autoencoders for Mechanistic Interpretability in NLP: A Critical Survey of Recoverability, Interpretability, and Utility](https://arxiv.org/abs/2610.20139) `2610.20139` | 0.50 | 0.40 | 0.00 | 0.00 | 0.20 | — | — |
| 27.4 | [Designing Textual Interfaces To Enhance Llm Reasoning On Audio-Visual Question Answering](https://arxiv.org/abs/2610.20085) `2610.20085` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [CASI: Credit Assignment via Set Intervention for Agentic Retrieval](https://arxiv.org/abs/2610.20293) `2610.20293` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [HEM-Med: A Hierarchical Agentic Memory Framework for Safe Multi-Center Medication Recommendation](https://arxiv.org/abs/2610.20191) `2610.20191` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 26.3 | [From Unreliable to Usable: Decomposing Failure Modes and Confidence Signals in LLM Entity Extraction - Evidence from Cyber Threat Intelligence](https://arxiv.org/abs/2610.20417) `2610.20417` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [What Is a Clarification? Estimating Model-Dependent Input Ambiguity](https://arxiv.org/abs/2610.20052) `2610.20052` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Twenty Days Later: A Pre-Registered Test of History-Based Forecasts of Brand Mentions in Gemini Answers](https://arxiv.org/abs/2610.19608) `2610.19608` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [Modeling and Understanding Neural Weights of Humanoid Policies with Diffusion](https://arxiv.org/abs/2610.20606) `2610.20606` | 0.35 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 25.8 | [Neural POS Tagging for the Low-Resource Saraiki](https://arxiv.org/abs/2610.20279) `2610.20279` | 0.25 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 25.8 | [Loss Curvature as a Sampling-Free Proxy for Memorization in Diffusion Models](https://arxiv.org/abs/2610.19042) `2610.19042` | 0.35 | 0.40 | 0.15 | 0.00 | 0.20 | — | — |
| 25.3 | [AFCS: Atomic-Fact-Conditioned Synthesis for Factual Knowledge Pretraining](https://arxiv.org/abs/2610.20086) `2610.20086` | 0.00 | 0.80 | 0.40 | 0.00 | 0.00 | — | — |
| 25.3 | [Same Claim, Different Interpretations: How LLM Agents Resolve Ambiguity in Scientific Claims](https://arxiv.org/abs/2610.20457) `2610.20457` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 25.3 | [Scaling Multilingual Safety and Reasoning Evaluation through Reasoning-Guided Translation](https://arxiv.org/abs/2610.20221) `2610.20221` | 0.30 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 24.7 | [ReAl: Refine neural operators via aligning dynamical flow](https://arxiv.org/abs/2610.20369) `2610.20369` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [Scaling Multilingual Mixture-of-Experts: How Language Coverage Changes Capacity and Compute Allocation](https://arxiv.org/abs/2610.20247) `2610.20247` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [Language Exposure Matters: Distribution and Order in Multilingual BabyLM Pre-training](https://arxiv.org/abs/2610.20102) `2610.20102` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [LLM-Based Policy Compliance Verification of Loan Applications](https://arxiv.org/abs/2610.20155) `2610.20155` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 23.7 | [Many Right Answers, One Training Example: Removing Responder Noise from GRPO for LLM Recommenders](https://arxiv.org/abs/2610.17771) `2610.17771` | 0.00 | 0.40 | 0.15 | 1.00 | 0.10 | — | — |
| 23.2 | [Revisiting Generative vs. Dense Retrieval: A Static Value Prior Reverses the Gap](https://arxiv.org/abs/2610.19065) `2610.19065` | 0.00 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 22.6 | [AdaRSI: On-Demand Recursive Self-Improvement for Embodied Agents](https://arxiv.org/abs/2610.20474) `2610.20474` | 0.25 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 22.6 | [Apertus 1.5: From Open Text Models to Open Multimodal Reasoners](https://arxiv.org/abs/2610.15542) `2610.15542` | 0.00 | 0.45 | 0.40 | 0.00 | 0.30 | — | — |
| 22.1 | [Your Transformer Is Secretly Elastic](https://arxiv.org/abs/2610.20588) `2610.20588` | 0.00 | 0.00 | 0.40 | 1.00 | 0.20 | — | — |
| 22.1 | [Retrieve What Helps, Only When It Helps: Agentic Reinforcement Learning for Selective Retrieval-Augmented Machine Translation](https://arxiv.org/abs/2610.19528) `2610.19528` | 0.30 | 0.40 | 0.05 | 0.00 | 0.20 | — | — |
| 21.6 | [Cortex: Consolidating Context into Parametric Memory](https://arxiv.org/abs/2610.20399) `2610.20399` | 0.00 | 0.40 | 0.40 | 0.00 | 0.30 | — | — |
| 21.1 | [Plan or Retrieve First? A Scale Study of Tool Discovery Architectures](https://arxiv.org/abs/2610.19068) `2610.19068` | 0.00 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 20.0 | [Manacá-1B-Instruct-2: Reproducible Safety Alignment for a Native Brazilian-Portuguese LLM. Negative DPO Results, a Controlled Composition Scan, and a Silent Failure Mode of the LLM Judge](https://arxiv.org/abs/2610.20570) `2610.20570` | 0.00 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 20.0 | [MolLean: A Process-Verifiable Language for Chemical Reasoning](https://arxiv.org/abs/2610.20459) `2610.20459` | 0.00 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 20.0 | [Lexical Persistence Without Semantic Entrenchment in AI Agent Communities](https://arxiv.org/abs/2610.20365) `2610.20365` | 0.10 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 19.5 | [Generative Representation Learning for Scalable Latent Tokenization](https://arxiv.org/abs/2610.20533) `2610.20533` | 0.35 | 0.00 | 0.40 | 0.00 | 0.00 | — | — |
| 18.4 | [Learning by Examples: Power Grid Steady State Analysis with a Tabular Foundation Model](https://arxiv.org/abs/2610.20462) `2610.20462` | 0.05 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 17.9 | [Anuvani-Diarization: Zero Shot Speaker Diarization and Annotation Conventions Across 22 Languages](https://arxiv.org/abs/2610.20327) `2610.20327` | 0.00 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 16.7 | [Diff Diffusion: Predicting Edit Paths with Diffusion Models for Efficient Code Editing](https://arxiv.org/abs/2610.17831) `2610.17831` | 0.15 | 0.45 | 0.00 | 0.00 | 0.16 | — | — |
| 8.4 | [PrismaCoder: Pushing the Boundaries of Open Multimodal Code Intelligence through Decomposed Rubric Synthesis](https://arxiv.org/abs/2610.20538) `2610.20538` | 0.00 | 0.00 | 0.40 | 0.00 | 0.00 | — | — |

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