The Morning Preprint

Friday 16 October

Showing the most recent batch — Friday 16 October. No papers have been ingested today yet.

1443 papers ingested by the estate's stations that day — about 10 minutes to read. The counts and the chart are the estate's own ingest, not arXiv's submission figures.

This day's harvest looks short: 1443 papers against a busiest day of 8455 in the week shown below. Weekends and a broken fetch look the same in the chart, and the chart counts the estate's ingest, not arXiv's.

The 60-second version

Five things from this day's batch, best first — everything below the fold expands on them.

  1. New method boosts target detection in hyperspectral imaging

    Reconstruction-to-Vector Diffusion uses manifold purification to set a new state of the art for hyperspectral anomaly detection.

  2. Agent reasoning steps reduced by up to eighteen percent

    ContextWeaver introduces a selective memory framework for LLM agents, cutting agent steps by 13-18% on software engineering benchmarks.

  3. Multi-hop question answering runtime slashed by seventy percent

    ConRAG uses a consensus-driven evidence flow to select passages and transfer information, reducing total runtime by up to 75.9%.

  4. Sleep recalibrates neural computation for precision

    Neural Computation Runs on Dissipative Structures Under Cyclic Calibration suggests sleep resets function-structure coupling to preserve accuracy.

  5. Verification-gated framework ensures industrial system safety

    Verification-Gated Agentic Mission-State Governance uses synchronized state objects to ensure actions only commit after deterministic verification. arXiv 2608.29005

Across the stations

AI Radio

121 papers today

The day's focus on AI Radio was heavily weighted toward advanced reasoning and multimodal understanding. The work on Beyond Reconstruction proposed Reconstruction-to-Vector Diffusion to move beyond scalar residuals in hyperspectral anomaly detection by using a manifold purification origin, which established a new state-of-the-art for target detectability. We also saw work on ContextWeaver, which introduced a selective memory framework for LLM agents to organize reasoning steps into dependency graphs, improving resolve rates on software engineering benchmarks by using 13-18% fewer agent steps.

Other papers contributed varied areas of research. ConRAG enhanced multi-hop question answering by using a consensus-driven evidence flow to select relevant passages and transfer intermediate information, reducing total runtime by up to 75.9%. On the modeling side, NEMORA introduced Neural Equivariant Multipole Operators for long-range atomistic learning, which reduces force errors relative to short-range backbones by over an order of magnitude while operating in linear time and memory complexity.

The overall picture suggests a push toward more structured and efficient reasoning systems. While DynFS-MoE focused on dynamic functional-structural mixture-of-experts for epilepsy diagnosis, other works explored different facets of LLM capability. For instance, TurboVGGT achieved fast multi-view 3D reconstruction by using adaptive alternating attention to capture global relationships across frames.

The day on AI Radio

Astrophysics Radio

91 papers today

The day's focus on Astrophysics Radio centered on constraining emission physics from extreme environments, particularly around supermassive black holes and in galactic structures. The work by Neutrino production at the extreme vicinity of supermassive black hole in NGC 1068 used updated IceCube data to tightly constrain the emission region size from 3R s to 5σ, suggesting stochastic acceleration is the most viable particle injection mechanism.

Other papers contributed by focusing on galactic dynamics and observational techniques; A 13 CO Detection in a Massive Molecular Gas Flow in Abell 1795's Central Galaxy provided new estimates for molecular gas mass, suggesting direct entrainment by radio bubbles. Meanwhile, The Radius-Luminosity Relationship for Active Galactic Nuclei with Hierarchical Bayesian Reverberation Mapping presented a robust Bayesian framework to improve constraints on the R-L relation using large AGN time series data.

The remaining research explored diverse areas such as stellar and planetary systems; Secular Evolution of Quasiperiodic Eruptions: eRO-QPE2 analyzed recurrent X-ray flares, finding recurrence times decrease at a rate of about 5 times 10^-6. Furthermore, studies like Hydrostatic Equilibrium Constraints on the Fuzzy Dark Matter Mass from Dwarf Spheroidals: Method, Systematics, and Robustness used hydrostatic equilibrium to estimate fuzzy dark matter boson masses in Milky Way dwarf spheroidals.

The day on Astrophysics Radio

Genomics Radio

8 papers today

The day's work focuses heavily on how neural computation manages its internal states, with a key paper suggesting sleep acts as a necessary recalibration phase to maintain computational accuracy by counterbalancing entropic drift Neural Computation Runs on Dissipative Structures Under Cyclic Calibration: A Systems Perspective on Sleep. This work posits that neural activity disrupts its own substrate, causing parameter drift, and sleep resets function-structure coupling to preserve precision.

Other research explored different facets of biological information processing; for instance, one study introduced CellNet to allow panoramic modeling of single-cell transcriptomes by integrating DNA identities and RNA expression Panoramic modelling of single-cell transcriptomes at scale with CellNet. Another piece addressed how mindfulness training reorganizes brain dynamics, revealing dissociable axes in response to different clinical statuses Two dissociable axes of brain hierarchical dynamics track universal and population-specific effects of mindfulness.

Finally, some work focused on applying computational models to specific problems; one study developed a dual LLM pipeline to objectively document treatment phenotypes for Rett syndrome by analyzing clinical notes Documenting Trofinetide Treatment Phenotypes in Rett Syndrome with a Dual-LLM Consensus Pipeline. This paper used LLMs to extract 18 treatment responses from patient notes, showing high agreement among experts on observed features like motor impairment and communication issues.

The day on Genomics Radio

Quantum Radio

176 papers today

The day's work focused heavily on exploring fundamental physics, with a major contribution coming from the study of singular wave functions and quantum collapse in inverse-power potentials. This work produced analytical forms and exact solutions for stationary wave functions under an inverse-power potential, showing how regularized versions restore a discrete spectrum of bound states.

Other papers contributed diverse ideas; for instance, agent-assisted experimental strategy transfer across quantum processors demonstrated how an agent can adapt experimental procedures across different quantum hardware to realize common tasks with high fidelity. The remaining submissions covered topics ranging from simulating topological superconductors using photon-mediated interactions to developing methods for accurate fidelity estimation in noisy quantum devices.

Collectively, these papers touch upon the theoretical limits of quantum simulation and experimental practice, leaving an open question about the practical implementation costs of complex quantum algorithms. While one work details how agent assistance scales experimental tasks across processors, another explores upper bounds on zero-error communication capacity. The day suggests a shift toward developing robust tools for scaling and characterizing increasingly noisy quantum hardware.

The day on Quantum Radio

Robotics Radio

31 papers today

The work on Verification-Gated Agentic Mission-State Governance focuses on ensuring that multi-robot industrial systems maintain task dependencies and safety during execution by using a verification-gated framework. This approach uses synchronized state objects, including an evolving task forest and a governed blackboard, to ensure that any proposed action or repair only commits after passing deterministic verification.

Other papers contribute to different areas of robotics research. For instance, SPARC introduces a risk-aware framework that automatically labels robot demonstrations with structured spatial annotations and assigns reliability scores, which helps in grounding policies for applications like motion planning.

Finally, several works address control and perception challenges; Task-space model-based control of pneumatic soft actuators provides a dynamic model for controlling soft actuators with high precision by fusing a quasi-static feedforward inverse model with a task-space PI controller.

The day on Robotics Radio

Security Radio

52 papers today

Today’s Security Radio focus is on how software provenance can be tracked through module placement using PlaceMark. This work shows that encoding a payload in where modules are placed in a repository allows for recovery even when the modules are moved, proving this placement acts as a useful carrier for watermarks embedded in source text.

The remaining papers touch on several areas of concern; NOMAI introduces a scalable federated learning system that simultaneously guarantees model, data, and integrity privacy using MPC. Furthermore, studies like ExecWeave reconstruct execution provenance by mapping logical operations to machine activity, while work on Agent Step-Up authorization tests binding mechanisms for agent approvals against various attack vectors.

The combination of these findings suggests that security needs to account for both data privacy in distributed training and the integrity of agent actions. One open question remains whether the observed carrier effect in PlaceMark can be reliably exploited across different programming environments like Django and Airflow.

The day on Security Radio

The day, in brief

AI Radio's framework allows for sample-efficient locally differentially private learning for structured univariate distributions like Gaussian mixture models AKMPC, which enables building models from sensitive data without compromising privacy through local differential privacy guarantees. Astrophysics Radio's analysis of dust activity evolution in interstellar comets shows that before perihelion the effective scattering cross section brightened across a region of two times ten four kilometers because it was proportional to r H-gamma. Genomics Radio's Panoramic modelling of single-cell transcriptomes at scale with CellNet introduces a scalable framework for interpreting large amounts of single-cell data across many species. Quantum Radio's three-axis active magnetic compensation method incorporates spatial feedforward matrix decoupling and a hysteresis PI regulation scheme to suppress Zeeman decoherence in cold atom quantum memories. Robotics Radio's vision-guided closed-loop autonomous navigation of light-driven Janus microbots showed that over ninety percent of measured trajectory deviations remained below six micrometers across various path geometries. Security Radio's MAppPrint introduces a behavior-level traffic fingerprint called the Traffic Behavior Correlation Graph or TBCG to identify applications by modeling unit traffic behaviors and their correlations.

The week in the estate's ingest

Seven days of papers the estate ingested. Weekends are striped — the weekend is visible in the data.

1118Sat
893Mon
8455Wed
1425Thu
1443Fri

Show the answer

1443 papers were ingested on Friday 16 October.

1443 papers, three ways

Where today's batch landed

By the paper's primary arXiv category.
cs.LG228
cs.CV181
cs.AI161
cs.CL147
cs.RO121
quant-ph108
cs.CR52
astro-ph.IM35

The index

The day's strongest 60, by score. Every title links its paper here and on arXiv.

  1. Exploring the Design Space of Permutation Diffusion LLMs [arXiv] [episode]
  2. When Does Corruption Break the Plan? A Degradation-Tolerance Benchmark for Camera-Only End-to-End Driving [arXiv] [episode]
  3. Agent Gym: A Framework for Continuous Evaluation and Evolution of LLM Agents Through Human-in-the-Loop Feedback [arXiv] [episode]
  4. Optimal Watermark Localization in Mixed-Source Large Language Model Texts [arXiv] [episode]
  5. Optimal Unambiguous DNFs and Alon-Saks-Seymour [arXiv] [episode]
  6. Phase Retrievability of Super Operators and Measurements [arXiv] [episode]
  7. Parity-Based Time-Bin Encoding Enabling SWAP Between Polarization and Time-Bin Qubits [arXiv] [episode]
  8. Beyond Relevance-Centric Retrieval: Rubric-Oriented Document Set Selection and Ranking [arXiv] [episode]
  9. Circuit Claims Depend on What Is Extracted and How It Is Compared [arXiv] [episode]
  10. MemSyco-Bench: Benchmarking Sycophancy in Agent Memory [arXiv] [episode]
  11. Replica Symmetry Breaking and Algorithmic Thresholds in Empirical Risk Minimization under Multi-Index Model [arXiv] [episode]
  12. CSWinUNETR: Segmentation of Thin Anatomical Structures in Medical Images [arXiv] [episode]
  13. Let's Ask Gauss: Improved One-Run Privacy Auditing [arXiv] [episode]
  14. Linear Strategic Classification with Endogenous Improvements [arXiv] [episode]
  15. High-fidelity molecular quantum logic gates resilient to interaction fluctuation [arXiv] [episode]
  16. Language-Based Agent Control [arXiv] [episode]
  17. Mind the Delay: Latency-aware Query Routing for Heterogeneous LLM pools [arXiv] [episode]
  18. PREPING: Zero-Task Agent Memory Construction via Self-Practice [arXiv] [episode]
  19. Mind the Gap: Structure-Aware Consistency in Preference Learning [arXiv] [episode]
  20. ORACLE-SWE: Quantifying the Contribution of Critical Information Signals on SWE Agents [arXiv] [episode]
  21. Where to Steer: Input-Dependent Layer Selection for Steering Improves LLM Alignment [arXiv] [episode]
  22. Towards Better Aligned LLM Judges for Radiology Report Evaluation across Modalities and Anatomies [arXiv] [episode]
  23. Double-weak-link interferometer of hard-core bosons in one dimension [arXiv] [episode]
  24. SEVerA: Verified Synthesis of Self-Evolving Agents [arXiv] [episode]
  25. Benchmarking Efficient & Effective Camera Pose Estimation Strategies for Novel View Synthesis [arXiv] [episode]
  26. High-quality single photons from cavity-enhanced biexciton-to-exciton transition [arXiv] [episode]
  27. Unified Perceptive Fall Safety for Humanoids [arXiv] [episode]
  28. Composition-RL: Compose Your Verifiable Prompts for Reinforcement Learning of Large Language Models [arXiv] [episode]
  29. Geometric Optimization over Quantum State Spaces: Tight Uncertainty Relations and Resource Certification [arXiv] [episode]
  30. CoCopula: Conformal Calibration and Copula Aggregation for Reliable Renewable Energy Forecasting [arXiv] [episode]
  31. DeFlow: Decoupling Behavior-Prior Modeling and Value Maximization for Offline Policy Extraction [arXiv] [episode]
  32. Why finite-size non-Hermitian eigenvalues can miss asymptotic eigenstates in topological physics [arXiv] [episode]
  33. Learning to Pose Problems: Reasoning-Driven and Solver-Adaptive Data Synthesis [arXiv] [episode]
  34. How Do LLM-Generated Texts Impact Term-Based Retrieval Models? [arXiv] [episode]
  35. UniFGVC: Universal Training-Free Few-Shot Fine-Grained Vision Classification via Attribute-Aware Multimodal Retrieval [arXiv] [episode]
  36. BOLT: Universal Classifier Training via a Bayes-Error Upper Bound [arXiv] [episode]
  37. CoTS: Competitive Token Selection for Latent Reasoning in Large Language Models [arXiv] [episode]
  38. PRISM: PRIvileged Sibling-guided Multi-Agent Topology Design [arXiv] [episode]
  39. Filling Before Advancing: Capability-Gap-Driven Post-Training for Scenario-Specialized Remote Sensing MLLMs [arXiv] [episode]
  40. Discriminative Barrier Functions for Safe Adversarial Imitation Learning from Observation [arXiv] [episode]
  41. KACE: Knowledge-Adaptive Context Engineering for Mathematical Reasoning [arXiv] [episode]
  42. EPIC-Bench: A Perception-Centric Benchmark for Fine-Grained Embodied Visual Grounding in Vision-Language Models [arXiv] [episode]
  43. Naka-GS: Naka-Guided Chroma Correction and Point-Prior Compaction for Low-Light 3D Gaussian Splatting [arXiv] [episode]
  44. PhyMix: Towards Physically Consistent Single-Image 3D Indoor Scene Generation with Implicit--Explicit Optimization [arXiv] [episode]
  45. Large Language Models Naively Recover Ethnicity from Individual Records [arXiv] [episode]
  46. Uncovering Grounding IDs: How External Cues Shape Multi-Modal Binding [arXiv] [episode]
  47. Towards Reliable Long-Context Reasoning: Detecting Unanswerable Questions via FactGuard [arXiv] [episode]
  48. Scaling Multimodal Search and Recommendation with Small Language Models via Upside-Down Reinforcement Learning [arXiv] [episode]
  49. MBBQ: A Dataset for Cross-Lingual Comparison of Stereotypes in Generative LLMs [arXiv] [episode]
  50. Unaware, Unfunded and Uneducated: A Systematic Review of SME Cybersecurity [arXiv] [episode]
  51. KNOU-Flood: Advancing global streamflow forecasting with lead-aware probabilistic machine learning [arXiv] [episode]
  52. Reasoning Error from Known Fact: Step-Level Self-Consistency Group Relative Policy Optimization for LLMs [arXiv] [episode]
  53. Words Speak Louder Than Code: Investigating Cognitive Heuristics in LLM-Based Code Vulnerability Detection [arXiv] [episode]
  54. VOiLA: Vectorized Online Planning with Learned Diffusion Models for POMDP Agents [arXiv] [episode]
  55. Multi-Agent Transactive Memory [arXiv] [episode]
  56. Fine-grained Human Motion Understanding with Language Models [arXiv] [episode]
  57. Spatially Conditioned Diffusion Policy: Learning Precise and Robust Manipulation with a Single RGB Camera [arXiv] [episode]
  58. EvoFlux: Inference-Time Evolution of Executable Tool Workflows for Compact Agents [arXiv] [episode]
  59. Rethinking Calibration for High-Sparsity LLM Pruning: From Single Sources to Capability Coverage [arXiv] [episode]
  60. From 3D Perception to Safety Reasoning: A Graph-Based Framework for Real-Time Underground Mine Monitoring [arXiv] [episode]

Listen

Everything the estate published on 2026-10-16, in the player. Playing from this page is how the edition learns whether anyone did.

Previous editions