# paper-forge daily report — 2026-09-25

146 new papers screened; 20 analyzed in depth; 0 reproduction runs.

## Reproductions

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

## Top candidates

### [Global Convergence of Third-Order Langevin Dynamics for Non-Convex Optimization via Simulated Annealing](https://arxiv.org/abs/2609.28611)
`2609.28611` · score 67.9 · analysis: heuristic · feasibility 8/10

We study global convergence guarantees of third-order Langevin dynamics for non-convex optimization via simulated annealing with fixed friction and decreasing noise. An explicit three-block distorted entropy transfers dissipation from the noisy auxiliary variable to the full state.

Rubric evidence: + cheap/closed-form ("Convex"); + toy/synthetic experiments ("Simulated"); + small/synthetic data ("Simulated"); + formal statement ("Convergence"); + claims simplicity ("simple"); + code: https://github.com/gagawjbytw/simulated-annealing-third-order-langevin

### [Orbital Error Dynamics: Self-Organized Criticality, Ephemeral Parameter Resonance, and Non-Linear Biological Ontologies in Zero-Storage Neural Synthesis](https://arxiv.org/abs/2609.30115)
`2609.30115` · score 56.8 · analysis: heuristic · feasibility 6/10

Modern deep neural networks treat parameters as static floating-point matrices stored in physical memory, incurring Von Neumann memory bottlenecks and representation collapse. We formulate Orbital Error Dynamics (OED), an analytical framework wherein synaptic weights are not stored masses (O(W)), but transient topological resonances (O(1)) derived procedurally from the complex quadratic polynomial map z_{n+1} = z_n^2 + c.

Claimed: accuracy 8.0 (abstract (regex extraction, unverified))
Rubric evidence: + cheap/closed-form ("convex"); + explicit algorithm/pseudocode ("Algorithm 1"); + formal statement ("Convergence"); - system/framework (many moving parts) ("framework"); + code: https://github.com/pCwOrM/mandelbrot-fractal-neural-synthesis

### [OBLIQ-IR: Training a Dense Retriever for Oblique Queries](https://arxiv.org/abs/2609.29649)
`2609.29649` · score 56.3 · analysis: heuristic · feasibility 6/10

Oblique retrieval, as exemplified by OBLIQ-Bench, asks a retriever to find documents whose relevance is determined by a latent attribute (an implicit stance, an analogous reasoning technique, an authorial fingerprint, or a vague tip-of-the-tongue recollection) that has little or no surface expression in the document. State-of-the-art dense encoders and agentic search pipelines built around frontier language models exhibit a large first-stage bottleneck on these tasks, while the same language models reliably verify relevance when shown candidates.

Claimed: NDCG@10 0.211 (abstract (regex extraction, unverified)); NDCG@10 0.171 (abstract (regex extraction, unverified)); NDCG@10 2.0 (abstract (regex extraction, unverified)); NDCG@10 0.222 (abstract (regex extraction, unverified)); NDCG@1 0.0 (abstract (regex extraction, unverified))
Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); - system/framework (many moving parts) ("agentic"); + code: https://github.com/DataScienceUIBK/obliq-ir; + keywords: retrieval

### [MF-SCBO : Multi-fidelity Scalable Constrained Bayesian Optimization](https://arxiv.org/abs/2609.29941)
`2609.29941` · score 54.2 · analysis: heuristic · feasibility 7/10

Many real-world optimization problems rely on expensive simulations or experiments, making the efficient use of available data essential. Multi-fidelity optimization of high-dimensional black-box functions subject to black-box constraints is increasingly relevant as the cost of objective evaluations continues to rise in applications such as machine learning, engineering, and control.

Rubric evidence: + toy/synthetic experiments ("simulations"); + public benchmarks ("benchmark"); + explicit algorithm/pseudocode ("Algorithm 1"); + formal statement ("convergence"); + claims simplicity ("simple"); no public code found; + keywords: sampling

### [Residual Correlation as a Diagnostic for Joint-Uncertainty Gains from GP Coregionalisation](https://arxiv.org/abs/2609.30085)
`2609.30085` · score 53.7 · analysis: heuristic · feasibility 7/10

In multi-target regression, correlated targets are often coupled through multi-output Gaussian processes with an intrinsic model of coregionalisation (GP-ICM), assuming that sharing statistical strength improves overall performance. In practice, the benefits are inconsistent.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + CPU-friendly ("lightweight"); + small/synthetic data ("synthetic"); + public benchmarks ("benchmarks"); + formal statement ("bound"); no public code found

### [Exact Bayes Regret and Asymptotic Optimality in High-Dimensional Gaussian Bandits](https://arxiv.org/abs/2609.28718)
`2609.28718` · score 52.6 · analysis: heuristic · feasibility 6/10

We study Bayesian linear bandits with an isotropic Gaussian parameter, independent Gaussian candidate arms, and Gaussian reward noise when the horizon is proportional to the dimension. The normalized posterior uncertainty has an explicit limit that is uniform over all causal policies.

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

### [Temporal Gradient Inversion for Private Trajectory Reconstruction in Embodied Reinforcement Learning](https://arxiv.org/abs/2609.30258)
`2609.30258` · score 52.6 · analysis: heuristic · feasibility 7/10

Distributed learning in embodied reinforcement-learning agents offers a degree of privacy by retaining raw sensor data on-device and transmitting only policy gradients to the server. Yet temporal structure can amplify this leakage beyond single-frame attacks.

Claimed: PSNR 3.0 (abstract (regex extraction, unverified))
Rubric evidence: + cheap/closed-form ("closed-form"); + concrete method ("closed-form"); + formal statement ("bound"); + claims simplicity ("simple"); no public code found; + keywords: regularization

### [Direct Message Approximation (DMA): A Consistency-Based Framework for Tractable Approximate Inference on Factor Graphs](https://arxiv.org/abs/2609.29466)
`2609.29466` · score 52.6 · analysis: heuristic · feasibility 7/10

Approximate message passing on factor graphs underlies two dominant families of probabilistic inference algorithms: expectation propagation (EP) and variational message passing (VMP). Both methods approximate the marginal at each factor edge, forcing an iterative round-robin schedule, risking negative-precision messages, and, for VMP, collapsing to point estimates at Dirac-delta factors.

Rubric evidence: + cheap/closed-form ("closed-form"); + explicit algorithm/pseudocode ("Algorithm 1"); + concrete method ("closed-form"); + formal statement ("We prove"); - system/framework (many moving parts) ("Framework"); no public code found; + keywords: variational

### [SmallReason-ColBERT: An Ultra-Small Late-Interaction Retriever for Reasoning Intensive Retrieval](https://arxiv.org/abs/2609.29652)
`2609.29652` · score 49.5 · analysis: heuristic · feasibility 5/10

Reasoning-intensive retrieval remains difficult for small models. Compact public ColBERTs are usually trained on general-purpose corpora and underperform reasoning-tuned 150M+ baselines on BRIGHT~\cite{bright} by several nDCG@10 points.

Claimed: nDCG@1 0.0 (abstract (regex extraction, unverified)); nDCG@10 1.21 (abstract (regex extraction, unverified))
Rubric evidence: + formal statement ("theorem"); + code: https://github.com/DataScienceUIBK/SmallReason-ColBERT; + keywords: retrieval

### [Decoupled Learning and Selection in Slate Recommendation for Privacy and Stability Under Noisy Scores](https://arxiv.org/abs/2609.29453)
`2609.29453` · score 46.8 · analysis: heuristic · feasibility 6/10

We formalize slate recommendation as a randomized score learner followed by deterministic selection. First, an appropriately scoped differential-privacy guarantee passes through selection and its audit trace by post-processing.

Rubric evidence: + toy/synthetic experiments ("simulations"); + formal statement ("bound"); + claims simplicity ("simple"); no public code found; + keywords: ranking

### [JevOut: Natural Context Can Flip Decision Models](https://arxiv.org/abs/2609.30243)
`2609.30243` · score 46.3 · analysis: heuristic · feasibility 4/10

Dedicated decision models such as Jev map unstructured language to probability distributions over finite choices, allowing their outputs to directly route requests, select tools, and trigger actions. Yet real-world inputs rarely arrive in isolation: they come with background details and surrounding context.

Rubric evidence: + code: https://github.com/xzx34/JevOut; + keywords: optimizer

### [The Fellowship of the Query: Learning Retrieval Actions](https://arxiv.org/abs/2609.28653)
`2609.28653` · score 46.3 · analysis: heuristic · feasibility 4/10

Retrieval-augmented question answering requires control decisions about when to decompose a question, search, reformulate, extract evidence, synthesize facts, verify progress, and stop. We study whether trajectory fine-tuning can improve small language models (SLMs) as next-action controllers.

Claimed: F1 0.6536 (abstract (regex extraction, unverified)); F1 0.7783 (abstract (regex extraction, unverified))
Rubric evidence: + code: https://github.com/padas-lab-de/agent-action-controller; + keywords: retrieval

### [Ingest-Time Fact Compilation for Cost-Efficient and Reliable Question Answering over Revised Corpora](https://arxiv.org/abs/2609.29661)
`2609.29661` · score 45.8 · analysis: heuristic · feasibility 6/10

Most agentic question answering (QA) systems do an important part of their semantic work at the worst possible time: every time someone asks a question. When a corpus contains revisions, drafts, revocations, deletions, and sources with different levels of authority, the model must reconstruct the governed current state on every read - then throw that work away and repeat it on the next query.

Rubric evidence: + toy/synthetic experiments ("synthetic"); + small/synthetic data ("synthetic"); + claims simplicity ("Simple"); - system/framework (many moving parts) ("agentic"); no public code found

### [A Nearly Quadratic Lower Bound for Linear Optimization over Convex Bodies in the Membership Oracle Model](https://arxiv.org/abs/2609.30215)
`2609.30215` · score 45.3 · analysis: heuristic · feasibility 6/10

We prove nearly quadratic lower bounds for randomized algorithms for linear optimization and uniform sampling over convex bodies in the membership oracle model. For linear optimization, this matches the known nearly quadratic upper bound up to a polylog factor in the dimension.

Rubric evidence: + cheap/closed-form ("Convex"); + formal statement ("Bound"); no public code found; + keywords: sampling

### [AERIAL: Adversarial Evaluation of Robustness in Accuracy-Preserving Low-Precision EEG Decoders](https://arxiv.org/abs/2609.30037)
`2609.30037` · score 44.7 · analysis: heuristic · feasibility 6/10

Deployment-oriented compression is attractive for resource-constrained brain--computer interfaces (BCIs), but whether it changes adversarial vulnerability remains unclear. On BCI Competition IV-2a, we compare 32-bit floating-point (FP32) EEGNet and ShallowConvNet models with global magnitude pruning and simulated INT8 post training quantization (PTQ) and quantization-aware training (QAT) across nine subjects and three seeds.

Claimed: accuracy 22.0 (abstract (regex extraction, unverified))
Rubric evidence: + toy/synthetic experiments ("simulated"); + small/synthetic data ("simulated"); no public code found

### [Claim-Gated Source-Risk Auditing for Generative Search](https://arxiv.org/abs/2609.29145)
`2609.29145` · score 44.7 · analysis: heuristic · feasibility 6/10

A generative search answer can cite a supported passage yet omit a source relationship that changes its interpretation. We specify a claim-gated audit of the query-source-answer tuple.

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

### [EvLink: Source-Grounded Evidence Linking for Graph RAG](https://arxiv.org/abs/2609.29695)
`2609.29695` · score 40.0 · analysis: heuristic · feasibility 5/10

Graph-based Retrieval-Augmented Generation (GraphRAG) supports multi-hop reasoning by organizing corpora into structured graphs. However, graph reachability often captures semantic association rather than evidence support, so a reachable passage may still fail to justify a required cross-passage transition.

Claimed: F1 2.4 (abstract (regex extraction, unverified))
Rubric evidence: + public benchmarks ("benchmarks"); + claims simplicity ("simple"); no public code found; + keywords: retrieval

### [From Interests to Semantic IDs: Retrieval-Grounded Credit Assignment for Generative Recommendation](https://arxiv.org/abs/2609.29983)
`2609.29983` · score 38.9 · analysis: heuristic · feasibility 4/10

Semantic IDs (SIDs) encode each catalog item as a short token sequence, enabling generative recommenders to predict the next item autoregressively. Reasoning-enhanced variants, an increasingly common extension, first generate a textual trace and then decode a next-item SID by beam search.

Rubric evidence: no public code found; + keywords: retrieval, ranking

### [Learning Better Reasoning for Generative Recommendation with Semantic IDs](https://arxiv.org/abs/2609.29973)
`2609.29973` · score 37.9 · analysis: heuristic · feasibility 4/10

Generative recommendation reformulates item retrieval as sequence generation, allowing a unified model to directly generate the next item from a user's interaction history. Semantic IDs further make this paradigm effective and scalable by representing each item as discrete codes, enabling knowledge sharing among semantically related items.

Rubric evidence: + public benchmarks ("benchmarks"); - system/framework (many moving parts) ("framework"); no public code found; + keywords: retrieval, ranking

### [A Systematic Multi-Domain Evaluation of Document Retrievers](https://arxiv.org/abs/2609.29455)
`2609.29455` · score 36.8 · analysis: heuristic · feasibility 4/10

Document retrieval is a crucial component of many modern AI systems, directly influencing their effectiveness, robustness, and fairness in downstream tasks. While recent years have seen a growing number of retrievers, comparative studies in the literature are typically limited in scope or focused on singular benchmarks, domains, or model families.

Rubric evidence: + public benchmarks ("benchmarks"); no public code found; + keywords: retrieval

## All screened papers

| score | paper | compute | data | clarity | code | interest | analysis | laptop-feasible |
|---:|---|---:|---:|---:|---:|---:|---|---|
| 67.9 | [Global Convergence of Third-Order Langevin Dynamics for Non-Convex Optimization via Simulated Annealing](https://arxiv.org/abs/2609.28611) `2609.28611` | 0.85 | 0.75 | 0.70 | 1.00 | 0.00 | heuristic | True |
| 56.8 | [Orbital Error Dynamics: Self-Organized Criticality, Ephemeral Parameter Resonance, and Non-Linear Biological Ontologies in Zero-Storage Neural Synthesis](https://arxiv.org/abs/2609.30115) `2609.30115` | 0.70 | 0.40 | 0.75 | 1.00 | 0.00 | heuristic | True |
| 56.3 | [OBLIQ-IR: Training a Dense Retriever for Oblique Queries](https://arxiv.org/abs/2609.29649) `2609.29649` | 0.65 | 0.75 | 0.30 | 1.00 | 0.20 | heuristic | True |
| 54.2 | [MF-SCBO : Multi-fidelity Scalable Constrained Bayesian Optimization](https://arxiv.org/abs/2609.29941) `2609.29941` | 0.65 | 0.45 | 1.00 | 0.00 | 0.20 | heuristic | True |
| 53.7 | [Residual Correlation as a Diagnostic for Joint-Uncertainty Gains from GP Coregionalisation](https://arxiv.org/abs/2609.30085) `2609.30085` | 0.80 | 0.80 | 0.55 | 0.00 | 0.00 | heuristic | True |
| 52.6 | [Exact Bayes Regret and Asymptotic Optimality in High-Dimensional Gaussian Bandits](https://arxiv.org/abs/2609.28718) `2609.28718` | 0.70 | 0.40 | 0.75 | 0.00 | 0.40 | heuristic | True |
| 52.6 | [Temporal Gradient Inversion for Private Trajectory Reconstruction in Embodied Reinforcement Learning](https://arxiv.org/abs/2609.30258) `2609.30258` | 0.70 | 0.40 | 0.90 | 0.00 | 0.20 | heuristic | True |
| 52.6 | [Direct Message Approximation (DMA): A Consistency-Based Framework for Tractable Approximate Inference on Factor Graphs](https://arxiv.org/abs/2609.29466) `2609.29466` | 0.70 | 0.40 | 0.90 | 0.00 | 0.20 | heuristic | True |
| 49.5 | [SmallReason-ColBERT: An Ultra-Small Late-Interaction Retriever for Reasoning Intensive Retrieval](https://arxiv.org/abs/2609.29652) `2609.29652` | 0.50 | 0.40 | 0.55 | 1.00 | 0.20 | heuristic | True |
| 46.8 | [Decoupled Learning and Selection in Slate Recommendation for Privacy and Stability Under Noisy Scores](https://arxiv.org/abs/2609.29453) `2609.29453` | 0.65 | 0.40 | 0.70 | 0.00 | 0.20 | heuristic | True |
| 46.3 | [JevOut: Natural Context Can Flip Decision Models](https://arxiv.org/abs/2609.30243) `2609.30243` | 0.50 | 0.40 | 0.40 | 1.00 | 0.20 | heuristic | True |
| 46.3 | [The Fellowship of the Query: Learning Retrieval Actions](https://arxiv.org/abs/2609.28653) `2609.28653` | 0.50 | 0.40 | 0.40 | 1.00 | 0.20 | heuristic | True |
| 45.8 | [Ingest-Time Fact Compilation for Cost-Efficient and Reliable Question Answering over Revised Corpora](https://arxiv.org/abs/2609.29661) `2609.29661` | 0.65 | 0.75 | 0.45 | 0.00 | 0.00 | heuristic | True |
| 45.3 | [A Nearly Quadratic Lower Bound for Linear Optimization over Convex Bodies in the Membership Oracle Model](https://arxiv.org/abs/2609.30215) `2609.30215` | 0.70 | 0.40 | 0.55 | 0.00 | 0.20 | heuristic | True |
| 44.7 | [AERIAL: Adversarial Evaluation of Robustness in Accuracy-Preserving Low-Precision EEG Decoders](https://arxiv.org/abs/2609.30037) `2609.30037` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | heuristic | True |
| 44.7 | [FB-GDM: Fully-Bayesian Guided Diffusion Models for High-Dimensional Linear Inverse Problems via Unsupervised Variational Inference](https://arxiv.org/abs/2609.29216) `2609.29216` | 0.55 | 0.40 | 0.60 | 0.00 | 0.40 | — | — |
| 44.7 | [NPBoost: Neural Processes with Gradient-Boosted Fixed Effects](https://arxiv.org/abs/2609.28122) `2609.28122` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 44.7 | [Claim-Gated Source-Risk Auditing for Generative Search](https://arxiv.org/abs/2609.29145) `2609.29145` | 0.65 | 0.75 | 0.40 | 0.00 | 0.00 | heuristic | True |
| 43.2 | [Multi-Dimensional Matching](https://arxiv.org/abs/2609.29958) `2609.29958` | 0.70 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 43.2 | [Low-Cost Assays for Measuring Model Behavior Across Vendors and Releases](https://arxiv.org/abs/2609.30012) `2609.30012` | 0.30 | 0.40 | 0.70 | 1.00 | 0.00 | — | — |
| 43.2 | [C3M: Cross-Session Multimodal Memory Maintenance for Long-Horizon Tasks](https://arxiv.org/abs/2609.29735) `2609.29735` | 0.50 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 42.1 | [Boolean threshold functions, neuron capacity, and memory retrieval](https://arxiv.org/abs/2609.29756) `2609.29756` | 0.70 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 41.6 | [A Contraction Framework for Stochastic Operators with Bootstrapping: Application to TD Learning](https://arxiv.org/abs/2609.29961) `2609.29961` | 0.65 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 41.6 | [Sufficiently Reduced Distributional Regression](https://arxiv.org/abs/2609.29291) `2609.29291` | 0.65 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 40.5 | [Let Training Guide Selection: Online Synthetic Data Filtering via Real-Anchored Utility](https://arxiv.org/abs/2609.29988) `2609.29988` | 0.45 | 0.80 | 0.30 | 0.00 | 0.20 | — | — |
| 40.0 | [Does a model's stated reason for rejecting a candidate do any work?](https://arxiv.org/abs/2609.30151) `2609.30151` | 0.30 | 0.40 | 0.55 | 1.00 | 0.00 | — | — |
| 40.0 | [Beyond Average Safety: Chance-Constrained LLM Fine-tuning](https://arxiv.org/abs/2609.29960) `2609.29960` | 0.50 | 0.40 | 0.60 | 0.00 | 0.20 | — | — |
| 40.0 | [Riemannian Gradient Descent for Gaussian Mixture Models with unknown diagonal covariances](https://arxiv.org/abs/2609.30220) `2609.30220` | 0.70 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 40.0 | [Neural Transport Nested Sampling](https://arxiv.org/abs/2609.29413) `2609.29413` | 0.50 | 0.45 | 0.40 | 0.00 | 0.40 | — | — |
| 40.0 | [EvLink: Source-Grounded Evidence Linking for Graph RAG](https://arxiv.org/abs/2609.29695) `2609.29695` | 0.50 | 0.45 | 0.55 | 0.00 | 0.20 | heuristic | True |
| 38.9 | [Aim Short to Reach Far: Your Frozen World Model Can Plan Better Than You Think](https://arxiv.org/abs/2609.30036) `2609.30036` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 38.9 | [Shrinking-Tube Concentration for Adaptive Markovian Stochastic Approximation](https://arxiv.org/abs/2609.29833) `2609.29833` | 0.50 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 38.9 | [Back to the Definition: Estimating Step-Level Advantages via Trajectory Graphs for Agentic Reinforcement Learning](https://arxiv.org/abs/2609.28963) `2609.28963` | 0.30 | 0.45 | 0.30 | 1.00 | 0.20 | — | — |
| 38.9 | [From Interests to Semantic IDs: Retrieval-Grounded Credit Assignment for Generative Recommendation](https://arxiv.org/abs/2609.29983) `2609.29983` | 0.50 | 0.40 | 0.40 | 0.00 | 0.40 | heuristic | True |
| 38.4 | [Beyond Compression: Training Latent Representations for Stable Long-Horizon Rollout in Neural Surrogate Solvers](https://arxiv.org/abs/2609.30198) `2609.30198` | 0.65 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 38.4 | [TTLab at AlexandriaX-2026: A Fine-Tuned Surface Tagger for Arabic Machine-Translation Error-Span Detection and Classification](https://arxiv.org/abs/2609.29633) `2609.29633` | 0.25 | 0.40 | 0.40 | 1.00 | 0.20 | — | — |
| 37.9 | [Diverse Geometries, Frozen Weights: Robust Heterogeneous Treatment-Effect Estimation via Causal Expert Ensembles](https://arxiv.org/abs/2609.29974) `2609.29974` | 0.70 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 37.9 | [Style, Not Self: Surface Cues Explain Zero-Shot Code Attribution by Large Language Models](https://arxiv.org/abs/2609.30048) `2609.30048` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures](https://arxiv.org/abs/2609.29429) `2609.29429` | 0.30 | 0.45 | 0.40 | 1.00 | 0.00 | — | — |
| 37.9 | [Learning Better Reasoning for Generative Recommendation with Semantic IDs](https://arxiv.org/abs/2609.29973) `2609.29973` | 0.50 | 0.45 | 0.30 | 0.00 | 0.40 | heuristic | True |
| 37.4 | [Artificial Societies Benchmark: A Validation Framework for Synthetic Research](https://arxiv.org/abs/2609.30030) `2609.30030` | 0.65 | 0.80 | 0.00 | 0.00 | 0.00 | — | — |
| 36.8 | [To Trust or Not to Trust: Retrieval-Augmented Fact Checking in Speech](https://arxiv.org/abs/2609.30227) `2609.30227` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [Self-Play Pretraining with Zero Data](https://arxiv.org/abs/2609.30063) `2609.30063` | 0.40 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 36.8 | [ChunkRank: Model-Aware Text Chunking and Abstention-Aware Answer Selection for LLM Pipelines](https://arxiv.org/abs/2609.29828) `2609.29828` | 0.30 | 0.40 | 0.40 | 1.00 | 0.00 | — | — |
| 36.8 | [Machine Unlearning for Gibbs Supervised Learning Algorithms](https://arxiv.org/abs/2609.29409) `2609.29409` | 0.50 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 36.8 | [When Does Action Credit Need Updating?](https://arxiv.org/abs/2609.29007) `2609.29007` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.8 | [Memory-Conditioned Diffusion Model for Generalized Langevin Dynamics](https://arxiv.org/abs/2609.28371) `2609.28371` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 36.8 | [A Systematic Multi-Domain Evaluation of Document Retrievers](https://arxiv.org/abs/2609.29455) `2609.29455` | 0.50 | 0.45 | 0.40 | 0.00 | 0.20 | heuristic | True |
| 36.8 | [Stochastic Semantic Evidence Graphs: Uncertainty Propagation and Governance for Agentic AI](https://arxiv.org/abs/2609.29703) `2609.29703` | 0.50 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 36.3 | [PoEM: Predicting RL Outcomes from Existing Policies](https://arxiv.org/abs/2609.30226) `2609.30226` | 0.45 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 36.3 | [Screen Before You Serve: Simulation for Production Customer Experience AI Agents at 140M Scale](https://arxiv.org/abs/2609.30137) `2609.30137` | 0.45 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 35.8 | [Intrinsic-Extrinsic Coupling in Learning Dynamics](https://arxiv.org/abs/2609.30185) `2609.30185` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [On the SoS Certifiability of Log-Concave Distributions](https://arxiv.org/abs/2609.30105) `2609.30105` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [Return or Revise? Learning When Revision Helps Retrieval-Augmented QA](https://arxiv.org/abs/2609.30087) `2609.30087` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Nuclear Norm-Regularized Bayesian Matrix Completion](https://arxiv.org/abs/2609.30078) `2609.30078` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [Tracking States or Tracking Cosets? An Algebraic Account of Learned State Tracking](https://arxiv.org/abs/2609.29951) `2609.29951` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [A New Gap Sequence for Shellsort: RL-Driven Algorithm Discovery Beyond $N^{4/3}$](https://arxiv.org/abs/2609.29881) `2609.29881` | 0.50 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 35.8 | [How Sensitive Are LLM Leaderboard Claims to Hidden Model Selection?](https://arxiv.org/abs/2609.28177) `2609.28177` | 0.30 | 0.40 | 0.55 | 0.00 | 0.40 | — | — |
| 35.8 | [OneTrans-V2: Unifying Retrieval, Pre-rank, and Fine-rank with One Transformer in Industrial Recommender](https://arxiv.org/abs/2609.28589) `2609.28589` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 35.8 | [SALI: Shot-Aware Late Interaction for Cross-Shot Relation Matching in Text-to-Video Retrieval using Film-Grammar Knowledge](https://arxiv.org/abs/2609.29721) `2609.29721` | 0.50 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 34.7 | [Selective Inference for Deep Clustering in Latent Spaces](https://arxiv.org/abs/2609.28756) `2609.28756` | 0.40 | 0.75 | 0.30 | 0.00 | 0.00 | — | — |
| 34.2 | [Local Geometric Mixing via Dobrushin Contraction with Applications to Diffusion Path Monte Carlo and the Proximal Sampler](https://arxiv.org/abs/2609.28338) `2609.28338` | 0.35 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 33.7 | [Anchored Extra-Proximal Methods: Optimal Higher-Order Methods for Monotone Inclusion Problems](https://arxiv.org/abs/2609.30212) `2609.30212` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [Graph-Based Inference and Topology-Aware Multi-Agent Reinforcement Learning for Large-Scale Railway Network Management](https://arxiv.org/abs/2609.30150) `2609.30150` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [AT-SKM-Net: An Accelerated Trainable Sampling Kaczmarz-Motzkin Framework for Linear Hard-Constraint Feasibility on Dynamic Graphs](https://arxiv.org/abs/2609.30088) `2609.30088` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.7 | [Spatio-temporally complementary feature propagation on graphs for longitudinal AADT estimation](https://arxiv.org/abs/2609.29906) `2609.29906` | 0.50 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 33.7 | [PUBG Ally: A Conversational Embodied Agent as an AI Teammate](https://arxiv.org/abs/2609.29837) `2609.29837` | 0.50 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 33.6 | [IterSynth: Rethinking Deep Search Agents via Role-Decoupled Iterative Synthesis](https://arxiv.org/abs/2609.29444) `2609.29444` | 0.30 | 0.05 | 0.40 | 1.00 | 0.26 | — | — |
| 32.6 | [Do Audio Language Models Hear and Read Distinctive Features Alike?](https://arxiv.org/abs/2609.30167) `2609.30167` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Canopy: Exploiting Piecewise Smooth Tree Priors for Multi-Fidelity Bandits](https://arxiv.org/abs/2609.30017) `2609.30017` | 0.30 | 0.40 | 0.55 | 0.00 | 0.20 | — | — |
| 32.6 | [When Temporal Perturbations Act Like Sensor Biases: Label-Free Auditing of Wearable Activity Recognizers](https://arxiv.org/abs/2609.29937) `2609.29937` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Does per-frame early exit pay? A compute-matched study of dynamic depth for on-device speech enhancement](https://arxiv.org/abs/2609.29867) `2609.29867` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Beyond Model Size: Redesigning LiSenNet for embedded speech enhancement](https://arxiv.org/abs/2609.29866) `2609.29866` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [A Training Criterion with Token-Level Tolerance to Transcription Ambiguity for Automatic Speech Recognition](https://arxiv.org/abs/2609.30160) `2609.30160` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [A Native-Reference Phone-Class Geometry for Second-Language Pronunciation Analysis](https://arxiv.org/abs/2609.30075) `2609.30075` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [TTLab at StanceEval-2026: A Cloze-Style Prompting Approach for Arabic-Language Stance Detection (CLASP-Ar)](https://arxiv.org/abs/2609.29733) `2609.29733` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [YODAS v3: Over 1 Million Hours of High-Bandwidth, Stereophonic, Multilingual Speech](https://arxiv.org/abs/2609.29448) `2609.29448` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Functional dynamic mode decomposition: Learning infinite-dimensional systems from data](https://arxiv.org/abs/2609.29159) `2609.29159` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Feature Space Selection and Heterogeneous Effect Estimation for Blood-Brain Barrier Permeability: A Random Forest to the Generalized Random Forest Pipeline](https://arxiv.org/abs/2609.29076) `2609.29076` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Personalised federated learning for Riemannian and Euclidean EEG decoding](https://arxiv.org/abs/2609.29037) `2609.29037` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [RLVR landscapes for iterated multiplications can be benign: Insights from spin-glass theory](https://arxiv.org/abs/2609.28625) `2609.28625` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [Rank-One Signal Recovery in Sparse Wishart Noise](https://arxiv.org/abs/2609.28163) `2609.28163` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [SGA: Uncertainty Quantification for Multi-Step Forecasting in Time Series Foundation Models](https://arxiv.org/abs/2609.28582) `2609.28582` | 0.30 | 0.40 | 0.40 | 0.00 | 0.40 | — | — |
| 32.6 | [Improving Ensemble Filters with Flow Matching](https://arxiv.org/abs/2609.28015) `2609.28015` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [ScalarLens: Numerical Embeddings with Stable Coordinates and Contextual Responses for CTR Prediction](https://arxiv.org/abs/2609.29182) `2609.29182` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.6 | [An Exploratory Ablation of a Small MLA--SSM Hybrid Language Model](https://arxiv.org/abs/2609.29618) `2609.29618` | 0.50 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 32.1 | [X-Rec Technical Report](https://arxiv.org/abs/2609.29180) `2609.29180` | 0.35 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 32.1 | [Fair Feed Ranking for Participatory Budgeting](https://arxiv.org/abs/2609.29819) `2609.29819` | 0.65 | 0.40 | 0.00 | 0.00 | 0.20 | — | — |
| 31.6 | [Reachability-Based Formal Verification of Graph Neural Networks with Node and Edge Features](https://arxiv.org/abs/2609.30079) `2609.30079` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [Multimodal Thinking with Renderable Programs](https://arxiv.org/abs/2609.30130) `2609.30130` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 31.6 | [Seek: Self-Evaluative Exploration for Knowledge Retrieval](https://arxiv.org/abs/2609.28980) `2609.28980` | 0.30 | 0.45 | 0.30 | 0.00 | 0.40 | — | — |
| 31.6 | [Advancing Model Research in AgentX: Long-Horizon Autonomy for Industrial Recommender Systems](https://arxiv.org/abs/2609.30001) `2609.30001` | 0.50 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [MQSS-Selector: RL-Guided Pass Selection for an MLIR Compilation Pipeline](https://arxiv.org/abs/2609.30104) `2609.30104` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [KernelOPT: Dispatch-Aware Agentic Search for GPU Kernel Optimization](https://arxiv.org/abs/2609.30059) `2609.30059` | 0.30 | 0.40 | 0.30 | 0.00 | 0.40 | — | — |
| 30.5 | [Error- and Prediction-Driven Motor Learning in the Cortico-Cerebellar Loop](https://arxiv.org/abs/2609.29945) `2609.29945` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Path-specific harm decomposition: A partial identification framework](https://arxiv.org/abs/2609.29938) `2609.29938` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Cost-Sensitive Online Window Size Selection for Portfolio Management](https://arxiv.org/abs/2609.29887) `2609.29887` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Multi-Task Learning by using Contextualized Word Representations for Syntactic Parsing of a Morphologically Rich Language](https://arxiv.org/abs/2609.29855) `2609.29855` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Robust Detection of LLM-Generated Text under Contamination](https://arxiv.org/abs/2609.29935) `2609.29935` | 0.30 | 0.45 | 0.55 | 0.00 | 0.00 | — | — |
| 30.5 | [Two Emojis of Difference: What Multilingual Affective Generation Benchmarks Actually Measure](https://arxiv.org/abs/2609.29445) `2609.29445` | 0.30 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 30.5 | [Cross-Country Code-Mixing for Generative Recommendation](https://arxiv.org/abs/2609.28972) `2609.28972` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [Asymmetric Dynamic Routing: Balancing Reasoning Depth and Computational Efficiency in Hypergraph RAG](https://arxiv.org/abs/2609.29282) `2609.29282` | 0.30 | 0.40 | 0.45 | 0.00 | 0.20 | — | — |
| 30.5 | [Anatomy of a Decision: Uncertainty-aware Hierarchical Intent Learning via Flow Matching for Multimodal Recommendation](https://arxiv.org/abs/2609.29609) `2609.29609` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.5 | [ASIRF: An Agentic Framework for Context-Dependent Sensitive Information Redaction](https://arxiv.org/abs/2609.29191) `2609.29191` | 0.50 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 30.0 | [Cultural Divergence Preservation: Diagnosing Flattening and Caricature in LLM-Simulated Survey Populations](https://arxiv.org/abs/2609.29928) `2609.29928` | 0.45 | 0.75 | 0.00 | 0.00 | 0.00 | — | — |
| 29.5 | [ARGUS: Role-Aware Event Knowledge Graphs for U.S. Employment-Discrimination Complaints](https://arxiv.org/abs/2609.30184) `2609.30184` | 0.30 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 29.5 | [MILO: Efficient Many-shot In-Context Learning with Block-wise Low-rank Compression](https://arxiv.org/abs/2609.29913) `2609.29913` | 0.30 | 0.45 | 0.50 | 0.00 | 0.00 | — | — |
| 29.5 | [CORDIAL: Calibrating Ordinal LLM Outputs from Few Labels](https://arxiv.org/abs/2609.29807) `2609.29807` | 0.30 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 28.9 | [From Processing to Functionality: Engineering Accessible Material States in Cu-Embedded SiO$_x$ Memristive Devices](https://arxiv.org/abs/2609.30047) `2609.30047` | 0.65 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 28.9 | [Augur: A Synthetic Decision Lab for Rehearsing Reactions to Product and Policy Changes](https://arxiv.org/abs/2609.29952) `2609.29952` | 0.15 | 0.75 | 0.40 | 0.00 | 0.00 | — | — |
| 28.9 | [GCUL: Ambiguity Identification in Text Emotion Classification via Cluster-Guided Learning](https://arxiv.org/abs/2609.29327) `2609.29327` | 0.65 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 28.4 | [SemMSA: Latent Semantic-Aided Robust Multimodal Sentiment Analysis with Incomplete Data](https://arxiv.org/abs/2609.30238) `2609.30238` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 28.4 | [LSF-SR: Latent Semantic Fusion for Sequential Recommendation via Flow-based Conditional Variational Autoencoders](https://arxiv.org/abs/2609.29815) `2609.29815` | 0.30 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 27.9 | [Transformers as Cross-Task Learners: Shared Structure Drives Sample Efficiency in In-Context Learning](https://arxiv.org/abs/2609.29060) `2609.29060` | 0.25 | 0.40 | 0.55 | 0.00 | 0.00 | — | — |
| 27.9 | [From Graphs to Feeders: Constraint-Guided Diffusion for Rule-Compliant Feeder Generation](https://arxiv.org/abs/2609.29879) `2609.29879` | 0.35 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [GRASP: Generating, Revising, and Assessing for Strategic Planning with Agentic AI](https://arxiv.org/abs/2609.30147) `2609.30147` | 0.30 | 0.40 | 0.30 | 0.00 | 0.20 | — | — |
| 27.4 | [What, When, and How: Audio Description as Constrained Global Optimization](https://arxiv.org/abs/2609.30121) `2609.30121` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [PrivDrift: Auditing User-Secret Leakage Under Topic Drift in Active LLM Conversations](https://arxiv.org/abs/2609.30094) `2609.30094` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Scoring Both Directions: LLMs realize the MRS they cannot reliably parse](https://arxiv.org/abs/2609.30071) `2609.30071` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [Benchmarking and Domain Adaptation of Automatic Speech Recognition (ASR) for Adolescent Health Communication in Ghanaian Languages](https://arxiv.org/abs/2609.29798) `2609.29798` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 27.4 | [JEV vs. LLMs as Rubric Judges: Cheaper, Faster, and Wrong in the Same Places](https://arxiv.org/abs/2609.29769) `2609.29769` | 0.30 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 26.8 | [ExplorationBench: Measuring AI Systems' Exploration in Verifiable Alien Worlds](https://arxiv.org/abs/2609.30199) `2609.30199` | 0.25 | 0.45 | 0.30 | 0.00 | 0.20 | — | — |
| 26.3 | [Learning and interpreting policies for simultaneous entanglement requests in quantum networks](https://arxiv.org/abs/2609.30157) `2609.30157` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [Learning to Ideate for Scientific Impact](https://arxiv.org/abs/2609.29802) `2609.29802` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [NumericJev: Jev-like LLM Numerical Decoding with Multiway Decision Trees](https://arxiv.org/abs/2609.28587) `2609.28587` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 26.3 | [SEEK: Skill-Routed Evaluation with Evolvable Knowledge for Industrial Search](https://arxiv.org/abs/2609.29803) `2609.29803` | 0.30 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.7 | [VietPrism: A large-scale Vietnamese speech and deepfake corpus with diverse dialects and code-switching](https://arxiv.org/abs/2609.30005) `2609.30005` | 0.25 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 24.2 | [Agentic Detection of Online Conspiracies](https://arxiv.org/abs/2609.30250) `2609.30250` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 24.2 | [Not All Confusion Is Equal: A Source-Aware Uncertainty Diagnosis for Fine-Grained Aircraft Detection](https://arxiv.org/abs/2609.29959) `2609.29959` | 0.50 | 0.40 | 0.00 | 0.00 | 0.00 | — | — |
| 24.2 | [agentic-ger: terminology recovery in long-form speech using global context](https://arxiv.org/abs/2609.29428) `2609.29428` | 0.30 | 0.40 | 0.30 | 0.00 | 0.00 | — | — |
| 23.2 | [Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs](https://arxiv.org/abs/2609.29845) `2609.29845` | 0.20 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 22.1 | [Improving Calibration of Black-Box Radiology AI Using Test-Time Augmentation](https://arxiv.org/abs/2609.29931) `2609.29931` | 0.50 | 0.00 | 0.30 | 0.00 | 0.00 | — | — |
| 21.1 | [Automated Regulatory Compliance Question Answering in Financial Services with Domain-Adapted Retrieval-Augmented Generation](https://arxiv.org/abs/2609.30009) `2609.30009` | 0.00 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 21.1 | [An Empirical Study of VLM Pipelines for Long-Document QA](https://arxiv.org/abs/2609.29933) `2609.29933` | 0.00 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 21.1 | [Reinforcement Learning with Verifiable Rewards for Small Search Agents](https://arxiv.org/abs/2609.28765) `2609.28765` | 0.00 | 0.45 | 0.40 | 0.00 | 0.20 | — | — |
| 20.0 | [How Reproducible Are Evaluation Conclusions? A Self-Audit of LLM-Inferred Prompt Structure](https://arxiv.org/abs/2609.30074) `2609.30074` | 0.00 | 0.40 | 0.40 | 0.00 | 0.20 | — | — |
| 19.5 | [GridSFM: A Foundation Model for Solving AC Optimal Power Flow](https://arxiv.org/abs/2609.30173) `2609.30173` | 0.05 | 0.40 | 0.45 | 0.00 | 0.00 | — | — |
| 19.5 | [TimeBraid: Unifying Time Series and Language for Understanding and Forecasting](https://arxiv.org/abs/2609.29792) `2609.29792` | 0.05 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 18.4 | [Adaptive Fisher-Whitened Cross-Covariance for Low-Resource Speech Recognition](https://arxiv.org/abs/2609.29800) `2609.29800` | 0.05 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 18.4 | [Learning a Flow to Self-Supervised Representations](https://arxiv.org/abs/2609.29350) `2609.29350` | 0.25 | 0.05 | 0.45 | 0.00 | 0.00 | — | — |
| 17.9 | [GHOST-Q: Towards Studying Grounding Hallucinations Overlooked Under Same-score TradeOffs in Quantized VLMS](https://arxiv.org/abs/2609.29999) `2609.29999` | 0.00 | 0.45 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [Minimally Invasive Steering of Language Models](https://arxiv.org/abs/2609.30218) `2609.30218` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [The Alignment Illusion in Multimodal Large Language Models](https://arxiv.org/abs/2609.30210) `2609.30210` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [R-DEIM Net: An Efficient Rationale-Augmented Dual-Expert Interaction Model for Paraphrase Detection](https://arxiv.org/abs/2609.30100) `2609.30100` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 16.8 | [CodeGraph: Open-Taxonomy Knowledge Graph for Source Code with Wikidata Grounding](https://arxiv.org/abs/2609.29474) `2609.29474` | 0.00 | 0.40 | 0.40 | 0.00 | 0.00 | — | — |
| 15.8 | [Encoded but Not Decoded: Layer-Localized Evidence for a Three-Level Gap in LLM Syntax](https://arxiv.org/abs/2609.29848) `2609.29848` | 0.00 | 0.45 | 0.30 | 0.00 | 0.00 | — | — |

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