2026-06-14·Daily digest

🔬 Research Pulse

Daily Digest

June 14, 2026

🤖 AI

🧠 LLMs

1. SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning

Authors: Seokju Cho, Ryo Hachiuma, Abhishek Badki... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How does the design of the action interface (how tools are invoked) constrain or enable open-ended spatial reasoning in tool-augmented VLM agents, particularly for complex 3D/4D tasks where existing single-pass code execution and structured tool-call interfaces fall short?

Summary: SpatialClaw is a training-free agentic framework for 3D/4D spatial reasoning that uses code as the action interface, with a stateful Python kernel and perception/geometry primitives that let a VLM iteratively compose analyses based on intermediate observations. It outperforms the prior best spatial agent by +11.2 points (59.9% avg) across 20 benchmarks and generalizes across six VLM backbones.

Key Results: SpatialClaw achieves 59.9% average accuracy across 20 spatial reasoning benchmarks spanning static and dynamic 3D/4D tasks, outperforming the prior recent spatial agent by +11.2 points. Gains are consistent across six VLM backbones from two model families, achieved without any benchmark-specific or model-specific adaptation, and the framework is training-free.

Key Findings:

Action interface design is a primary bottleneck for tool-augmented spatial reasoning agents, not just the underlying VLM capability
A stateful code-cell interface conditioning on prior text + visual outputs outperforms both single-pass code and structured tool-call interfaces
The approach generalizes: consistent gains across six VLM backbones from two model families with no per-model tuning

Technical Novelty: The key novelty is treating code as a stateful, iterative action interface rather than a single-pass program: a persistent Python kernel pre-loaded with frames and perception/geometry primitives lets the agent write one cell per step conditioned on all prior text AND visual outputs, enabling adaptive composition that neither single-pass code agents nor structured tool-call agents support.

What's New: Prior spatial agents either commit to a full analysis up-front (single-pass code) or are constrained by rigid tool-call schemas; SpatialClaw is the first to combine an interactive stateful Python kernel, multimodal intermediate feedback, and a pre-loaded primitive library specifically for open-ended 3D/4D spatial reasoning in a training-free manner.

Extension Opportunities:

Extend the stateful Python kernel with additional perception primitives (e.g., physics simulation, scene graph reasoning, or audio-spatial cues) to broaden the agent's reasoning toolkit beyond geometry
Apply the code-as-action interface paradigm to embodied/robotic agents that must combine spatial reasoning with real-world action planning and execution
Investigate fine-tuning or RL on the agent's code traces to distill the multi-step reasoning into smaller VLMs, reducing inference cost while preserving accuracy

Replicability: Abstract does not mention code/data release. Reproducing would require access to the 20 spatial reasoning benchmarks, the six VLM backbones (commercial API access likely needed), and compute for multi-step inference across all benchmarks — moderate to high inference budget but no training required since framework is training-free.

Research Gaps:

Lack of systematic study on how action interface design (not just tool quality) affects agentic reasoning performance
Existing spatial agents cannot flexibly adapt analysis strategy to intermediate visual observations during reasoning

2. Automated reproducibility assessments in the social and behavioral sciences using large language models

Authors: Tobias Holtdirk, Pietro Marcolongo, Anna Steinberg Schulten... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: Can large language models automate the reproducibility assessment of empirical studies in the social and behavioral sciences, which is currently a resource-intensive manual process that doesn't scale?

Summary: The authors build an LLM pipeline that automatically reanalyzes published behavioral/social science studies to assess reproducibility, and benchmark it against human reanalysts on 76 studies with predefined claims. The LLM matches or outperforms humans on both effect-size recovery (41% vs 34%) and qualitative conclusions (96% vs 74%), suggesting LLMs can serve as a scalable substitute for manual reproducibility audits.

Key Results: Tested on N=76 published behavioral/social science studies with predefined claims. LLM pipeline recovered original effect sizes in 41% of studies (within ±0.05 Cohen's d tolerance) vs. 34% for human reanalysts. LLM matched qualitative conclusions in 96% of cases vs. 74% for humans. LLM failed to produce viable effect size estimates in only 7 of 76 studies.

Key Findings:

LLM-driven reanalysis outperforms human reanalysts on qualitative agreement with original conclusions (96% vs 74%)
LLM effect-size recovery rate is modestly better than humans (41% vs 34% within ±0.05 Cohen's d)
The pipeline fails outright on roughly 9% of studies (7/76), indicating remaining limits in data parsing or method interpretation

Technical Novelty: Prior LLM-for-science work focused on summarization, peer review assistance, or code generation. This is the first end-to-end pipeline where an LLM independently performs statistical reanalysis from raw data and published methods descriptions, then is benchmarked against trained human reanalysts on a held-out reproducibility corpus.

What's New: Frames reproducibility assessment — historically a labor-bound bottleneck in meta-science — as a tractable LLM agent task, and provides a head-to-head human-vs-LLM benchmark rather than just demonstrating capability in isolation.

Extension Opportunities:

Build a continuous auditing service that ingests preprints/published papers with open data and outputs automated reproducibility reports, integrated with journals or repositories like OSF
Extend the pipeline beyond behavioral/social science to fields with messier data (clinical trials, economics, neuroimaging) by adding domain-specific analysis priors and statistical method libraries
Develop a feedback loop where the LLM flags ambiguous methodological descriptions in papers — turning reproducibility checks into a pre-publication tool that improves manuscript clarity

Replicability: Abstract does not mention code/data availability explicitly. Reproducing would require access to the 76 study dataset (likely from prior reproducibility projects like SCORE or ManyLabs), API access to the underlying LLM, and modest compute (inference-only, no training). Estimated cost: low-to-moderate API spend (likely <$1K).

Research Gaps:

Doesn't characterize why the LLM (or humans) fail on the studies where effect sizes diverge — error taxonomy would guide future improvements
Limited to behavioral/social sciences with predefined claims; generalization to fields with ambiguous hypotheses or proprietary data is unclear

3. Agents-K1: Towards Agent-native Knowledge Orchestration

Authors: Zongsheng Cao, Bihao Zhan, Jinxin Shi... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can LLM-based research agents move beyond shallow paper representations (abstracts, surface mentions, flat citation edges) to orchestrate deep scientific knowledge — entities, claims, evidence, mechanisms, and method lineages — needed for genuine scientific reasoning?

Summary: Agents-K1 is an end-to-end pipeline that converts raw scientific papers into agent-native knowledge graphs by combining a multimodal full-paper parser, a 4B GRPO-trained extraction model, and a tri-source retrieval CLI. Applied to 2.46M papers across six disciplines, it produces Scholar-KG (1M-paper subset released) and reportedly outperforms prior work on scientific IE, KG construction, and multi-hop reasoning.

Key Results: Built Agents-K1, an end-to-end pipeline combining a 5-module multimodal parser, a 4B IE backbone trained via GRPO with rule-based rewards, and a tri-source graphanything CLI. Processed 2.46M papers across six subjects into Scholar-KG, releasing a 1M-paper subset. Reports superior performance on scientific IE, KG construction, and multi-hop scientific reasoning benchmarks (specific numbers not given in abstract).

Key Findings:

A purpose-built schema capturing entities, evidence, citations, and typed relations across full papers outperforms abstract-only or citation-flat representations for downstream scientific reasoning
A small (4B) IE backbone trained with GRPO + rule-based rewards is sufficient to drive large-scale, high-quality KG construction
Unifying web search, multimodal graph retrieval, and cross-document traversal in a single agent interface enables stronger multi-hop scientific reasoning than any single source

Technical Novelty: Unlike prior agent work that treats papers as flat citation graphs with abstract-level text, Agents-K1 defines an agent-native KG schema covering full-paper entities, multimodal evidence, and typed inter-entity relations. The 4B IE model trained with GRPO + rule-based reward (rather than SFT or larger general models) and the tri-source agent CLI unifying web, multimodal graph retrieval, and cross-document traversal are the key novel pieces.

What's New: Reframes the research-agent stack from 'agent orchestration' to 'knowledge orchestration', introducing an agent-native KG schema, a GRPO-trained compact extractor, and a tri-source retrieval interface — together producing one of the largest released scientific KGs.

Extension Opportunities:

Extend the schema and pipeline to non-scientific corpora (legal, medical records, patents) leveraging the authors' note about general-domain applicability
Use the schema-conformant KG as a grounding source for training or evaluating reasoning LLMs on multi-hop scientific QA, beyond retrieval
Add temporal/version tracking of method lineages to power 'idea evolution' agents that trace how a technique propagates across subfields

Replicability: A 1M-paper subset of Scholar-KG is publicly released; full KG accessible via an SCP link. No explicit code release mentioned in the abstract. Training a 4B model with GRPO and processing 2.46M papers (multimodal parsing included) implies substantial compute — likely multi-GPU cluster for weeks; inference/use of the released KG is far cheaper.

Research Gaps:

Existing research agents reduce papers to abstracts and flat citation edges, losing mechanisms, claims, evidence, and method lineages
Lack of agent-native, multimodal, full-paper knowledge graphs and standardized retrieval interfaces over them

🦾 ROBOTICS

1. LabVLA: Grounding Vision-Language-Action Models in Scientific Laboratories

Authors: Baochang Ren, Xinjie Liu, Xi Chen... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can Vision-Language-Action (VLA) models be adapted to execute scientific laboratory protocols, given that existing policies are trained on household/tabletop data and lack exposure to lab instruments, transparent liquids, and fixed protocol workflows across diverse robot embodiments?

Summary: LabVLA adapts vision-language-action models to scientific laboratory protocol execution by combining a simulation-based data engine (RoboGenesis) that produces multi-embodiment lab demonstrations with a two-stage training recipe: FAST action-token pretraining on a Qwen3-VL-4B backbone followed by flow-matching posttraining with a DiT action expert. It achieves state-of-the-art average success rate on the LabUtopia benchmark in both in- and out-of-distribution settings.

Key Results: LabVLA achieved the highest average success rate among all evaluated baselines on the LabUtopia benchmark under both in-distribution and out-of-distribution settings. Built on a Qwen3-VL-4B-Instruct backbone with FAST action token pretraining followed by flow-matching posttraining with a DiT action expert. Specific success-rate percentages were not provided in the abstract.

Key Findings:

Data scarcity and embodiment diversity — not just model architecture — are the central bottlenecks for deploying VLAs in scientific labs.
A two-stage recipe (discrete FAST tokens first, then continuous flow-matching control) outperforms direct continuous-control training by making the backbone action-aware before specialization.
Simulation-generated, filtered, multi-embodiment demonstrations are sufficient to achieve leading performance on lab benchmarks under both in- and out-of-distribution evaluation.

Technical Novelty: Two-stage training recipe combining FAST discrete action token pretraining (to make the VLM action-aware before continuous control) with flow-matching posttraining via a DiT action expert under 'knowledge insulation', plus RoboGenesis — a simulation data engine that composes lab workflows from atomic skills and exports structured demonstrations across multiple robot embodiments.

What's New: First VLA system explicitly targeted at scientific laboratory protocols (transparent liquids, instruments, fixed workflows) rather than household/tabletop settings, paired with a simulation data engine designed for multi-embodiment lab demonstrations and a novel pretraining-then-flow-matching training pipeline with knowledge insulation.

Extension Opportunities:

Extend RoboGenesis with additional atomic skills for specialized domains (e.g., cell culture, electrochemistry, cryogenics) to broaden protocol coverage beyond what LabUtopia evaluates.
Bridge sim-to-real by collecting paired real-lab demonstrations and fine-tuning LabVLA to quantify the simulation transfer gap on physical wet-lab hardware.
Integrate LabVLA with an LLM-based protocol planner (e.g., from literature-derived SOPs) to create an end-to-end pipeline from natural-language experimental design to robotic execution.

Replicability: Abstract does not mention public code/data release for LabVLA, RoboGenesis, or LabUtopia. Reproduction would require a Qwen3-VL-4B-Instruct backbone, multi-GPU training for flow-matching posttraining of a DiT expert, and a simulator capable of rendering laboratory scenes with transparent liquids — likely 8+ high-memory GPUs (A100/H100 class) for training, plus simulator compute for data generation.

Research Gaps:

No sim-to-real validation reported — performance on physical laboratory hardware remains unverified.
Coverage of atomic skills and protocol types is bounded by LabUtopia; long-horizon, multi-step experimental campaigns (e.g., full synthesis or assay runs) are not directly evaluated.

2. Low cost, easily manufactured, highly flexible strain and touch sensitive fiber for robotics applications

Authors: Christian Diaz Herrera, Srushti Raste, Simin Liu... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can stretch and touch sensors for robotics be made affordable, fast to manufacture, and flexible without requiring specialized equipment or expensive materials?

Summary: The paper introduces a conductive fiber sensor built from cheap commercial conductive thread inside silicone tubing, manufacturable by hand in minutes. It demonstrates the same fiber serving as both a resistive strain sensor and a knitted capacitive touch/proximity sensor across five robotics applications, plus a method to repair cut fibers.

Key Results: Demonstrated a conductive fiber built from off-the-shelf parts (conductive thread at $0.07/ft, silicone tubing at $0.94/ft, $2 loop-style needle threader) that can be made in ~2 minutes for a 20 cm length. Validated across five applications: resistive strain sensing (triggering grasp in pneumatic assistive finger, pose sensing of a pneumatic robotic strap, pose estimation of a flexible solid) and capacitive sensing (touch-triggered robot arm motion, near-field hand-following). Also demonstrated a repair method for cut fiber.

Key Findings:

A single fiber design works in both resistive (strain) and capacitive (touch/near-field) modes, covering grasp triggering, soft-robot pose estimation, and hand-following
Manufacturing is radically cheaper and faster than prior soft sensors: ~$1/ft materials, 2 minutes per 20 cm, no specialized equipment
The fiber is flexible enough to be knitted into fabric-like capacitive sensors and can be repaired after being cut, improving practical longevity

Technical Novelty: Combining commodity conductive thread inside silicone tubing — threaded with a $2 needle threader — to yield a single fiber that functions as both a resistive strain sensor and a capacitive touch/proximity sensor, with a documented repair procedure. Prior soft sensors typically require liquid metals, custom molding, lithography, or specialized 3D printing.

What's New: Most soft strain/touch sensors rely on liquid metals (eGaIn), custom elastomer molding, or printed conductive composites with specialized equipment. This work shows that a thread-in-tube construction using only hobby-grade parts can match key sensing functions while being knit-compatible and repairable.

Extension Opportunities:

Develop an automated knitting/threading rig to scale manufacturing beyond hand-fabrication while preserving the low BOM cost
Integrate the fiber into a full soft-robot proprioceptive skin with multi-channel readout and ML-based pose regression across many fibers
Characterize long-term durability, hysteresis, and washability for wearable/assistive deployments and add closed-loop control demos using the strain signal

Replicability: The abstract describes only commodity parts and hand tools, so physical replication is highly accessible (sub-$5 of materials per fiber, ~2 min build time). No mention of released code, datasets, or CAD; reproducing the robot-arm and pneumatic demos would require the corresponding hardware but no significant compute.

Research Gaps:

No quantitative comparison of sensitivity, linearity, drift, or hysteresis against established soft-sensor baselines is described
Manufacturing scalability is discussed conceptually but not demonstrated with an automated process or yield/variance data

3. WT-UMI: Tactile-based Whole-Body Manipulation via Force-Supervised Contact-Aware Planning

Authors: Jaehwi Jang, Zhaoyuan Gu, Alfred Cueva... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can humanoid robots perform whole-body manipulation of bulky, deformable, and shared-load objects with explicit contact-force regulation, given that imitation policies typically treat force only implicitly and demonstration sources (human vs. teleoperation) offer complementary but incomplete modalities?

Summary: WT-UMI introduces a wearable whole-body tactile interface usable by both humans and humanoids that records tactile images, contact forces, and end-effector poses, then trains a force-supervised planner with a pose-correction module to bridge human demonstrations and robot teleoperation. The system outputs end-effector pose and force trajectories that drive a tactile admittance controller, improving success and contact tracking on five contact-rich whole-body tasks over four baselines.

Key Results: Across five contact-rich tasks spanning deformable objects, bulky rigid objects, and human-humanoid collaboration, WT-UMI improves success rate and reduces contact-position tracking error compared to four policy baselines. Specific numeric values for success rates and error reductions are not disclosed in the abstract.

Key Findings:

Unifying human and teleoperation demonstrations through a shared wearable tactile interface yields complementary force and action data
Explicitly supervising on contact force (not just pose) improves contact-rich whole-body manipulation performance
A learned correction module can convert natural human poses into robot-executable contact-aware targets, outperforming four baseline policies on five tasks

Technical Novelty: A force-conditioned target-pose correction module that learns to translate human-demonstrated poses into robot-executable contact-aware targets using teleoperation data, combined with a force-supervised planner that jointly predicts pose chunks and contact-force trajectories used as references for a tactile admittance controller — bridging the human-vs-teleoperation demonstration gap explicitly via force.

What's New: Prior imitation policies treat contact force implicitly and rely on a single demonstration source. WT-UMI is novel in (1) a dual-mode wearable tactile interface usable by humans and humanoids alike, (2) explicit force supervision in the planner, and (3) a learned human-to-robot pose correction grounded in force.

Extension Opportunities:

Extend the wearable tactile interface to bimanual coordination with object-centric force budgeting for collaborative carrying with multiple humanoids
Replace the admittance controller with a learned residual force policy that adapts gains online based on tactile image distribution shifts
Apply the force-conditioned target-pose correction module to mobile manipulation where base motion couples with contact dynamics (e.g., pushing carts, opening heavy doors)

Replicability: Project page at https://wt-umi.github.io/WTUMI/ is referenced but the abstract does not explicitly confirm code/data release. Reproduction would require a humanoid platform, the custom wearable tactile interface (hardware fabrication), tactile sensors, and standard GPU compute for policy training — a high hardware barrier.

Research Gaps:

Quantitative generalization to unseen object geometries, masses, and deformability regimes is not characterized in the abstract
Scaling to long-horizon tasks and dynamic disturbances (e.g., a human partner suddenly releasing load) is not addressed

💻 COMPUTE

1. Measurement-Based Performance Evaluation of SmartRSUs with Heterogeneous Antenna Architectures for V2X Communications

Authors: Marco Savarese, Gaetano Orazio Cauchi, Salvatore Iandolo... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How do different V2X antenna architectures (integrated all-in-one rooftop vs. external dual ITS-G5) on SmartRSUs affect real-world communication performance, and what trade-offs do they present for C-ITS infrastructure deployment?

Summary: The paper empirically compares two custom SmartRSU designs — one with an integrated rooftop GNSS/comms module and one with external dual ITS-G5 antennas — using a co-located commercial RSU as a ground-truth benchmark. Measurements of coverage, RSSI, packet loss, and bidirectional latency reveal antenna-design and placement trade-offs that inform next-generation C-ITS infrastructure choices.

Key Results: The authors built two custom SmartRSUs on a proprietary OBU platform — one with an integrated GNSS+comms rooftop module, one with external dual ITS-G5 (IEEE 802.11p) antennas at 5.9 GHz plus a dedicated GNSS antenna — and measured coverage, RSSI, packet loss, and bidirectional end-to-end latency (OBU↔infrastructure). A co-located commercial COTS V2X RSU served as a ground-truth benchmark under identical environmental conditions. Specific numerical results are not disclosed in the abstract.

Key Findings:

Antenna design and physical placement materially affect V2X link reliability and latency, not just radio/MAC parameters
Integrated all-in-one rooftop modules and external dual-antenna setups exhibit distinct trade-offs across coverage, RSSI, and packet loss
A co-located commercial RSU benchmark is a viable methodology for validating custom SmartRSU performance under identical environmental conditions

Technical Novelty: A controlled, co-located A/B comparison of integrated vs. external V2X antenna architectures on the same proprietary OBU-derived SmartRSU platform, validated against a commercial RSU as ground truth — most prior V2X measurement work studies a single RSU or compares vendors rather than isolating the antenna-architecture variable.

What's New: Rather than benchmarking a single RSU or comparing vendors, the authors hold the OBU/software stack constant and vary only the antenna architecture, enabling a clean attribution of performance differences to antenna design — and they ground-truth the results with a co-located commercial RSU.

Extension Opportunities:

Repeat the measurement campaign with C-V2X (PC5/sidelink, 3GPP Rel-16/17) instead of ITS-G5 to compare antenna-architecture sensitivity across radio access technologies
Add NLOS, urban-canyon, and adverse weather scenarios plus multi-vehicle concurrent traffic to stress-test the antenna designs beyond clean line-of-sight conditions
Build an open-source digital twin / ray-tracing model calibrated against the measured RSSI and packet-loss data so designers can predict SmartRSU placement trade-offs before deployment

Replicability: No code, datasets, or hardware schematics are mentioned in the abstract. Reproduction would require custom OBU hardware adapted for infrastructure use, two antenna configurations, a commercial ITS-G5 RSU benchmark, and a field test site with instrumented vehicles — compute needs are negligible, but hardware and field-campaign costs are substantial.

Research Gaps:

Limited to ITS-G5 (IEEE 802.11p) at 5.9 GHz; C-V2X PC5 and hybrid stacks are not evaluated
Abstract does not indicate testing under NLOS, dense urban, high-mobility, or adverse-weather conditions that dominate real C-ITS deployments

2. Observation of Non-Gaussian Magnon Dynamics in a Two-Dimensional Long-Range XY Model

Authors: S. -A. Guo, J. -Y. Tan, J. Ye... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can one experimentally observe and verify non-Gaussian magnon dynamics in a 2D long-range XY model while overcoming decoherence, statistical fluctuations, and calibration errors that typically obscure such signatures?

Summary: The paper experimentally demonstrates the crossover from Gaussian to non-Gaussian magnon dynamics in a 2D long-range XY model on a trapped ion quantum simulator. By comparing measured high-order spin correlations against mean-field and Holstein-Primakoff predictions, the authors witness genuine non-Gaussian behavior in a way that is robust to experimental calibration errors, marking a verifiable boundary between classically tractable and potentially quantum-advantaged regimes.

Key Results: Demonstrated the Gaussian-to-non-Gaussian crossover on a 2D XY model with long-range, spatially structured interactions using a trapped ion quantum simulator. Verified single-spin observable dynamics across varying initial magnon excitation densities and compared high-order spin correlations against mean-field and Holstein-Primakoff approximations, showing non-Gaussian behavior in a calibration-error-independent manner.

Key Findings:

Engineered a 2D XY Hamiltonian with long-range, spatially structured interactions on a trapped ion platform and verified single-spin dynamics across varying magnon densities
High-order spin correlations diverge from both mean-field and Holstein-Primakoff (Gaussian) predictions, providing a clean signature of non-Gaussian dynamics
The witness protocol is independent of calibration errors, addressing a key obstacle in experimental verification of non-Gaussian many-body dynamics

Technical Novelty: First experimental demonstration of a calibration-error-independent witness of non-Gaussian magnon dynamics in a 2D long-range XY model, comparing high-order spin correlations directly against both mean-field and Holstein-Primakoff predictions on a trapped ion platform with engineered spatially structured interactions.

What's New: Combines a 2D long-range XY simulator with a calibration-robust comparison against two distinct Gaussian baselines (mean-field and Holstein-Primakoff), enabling unambiguous experimental witness of non-Gaussian behavior — prior work largely focused on 1D systems or lacked calibration-independent verification.

Extension Opportunities:

Scale the trapped ion simulator to larger 2D lattices to probe regimes where Holstein-Primakoff and mean-field approximations definitively break down, mapping the quantum advantage frontier
Extend the protocol to other long-range spin models (e.g., Ising, Heisenberg, or frustrated geometries) to characterize universality of non-Gaussian crossovers
Develop classical ML surrogates trained on the verified Gaussian regime data, then benchmark them against the simulator's non-Gaussian regime to quantify exact crossover boundaries

Replicability: Abstract does not mention code/data release. Reproduction requires access to a programmable trapped ion quantum simulator capable of engineering 2D long-range XY Hamiltonians with spatially structured couplings and high-fidelity single-spin readout — significant specialized hardware investment.

Research Gaps:

Scaling and benchmarking the non-Gaussian regime against state-of-the-art classical simulation methods (e.g., tensor networks, neural quantum states) to formally establish a quantum advantage threshold
Extending verification protocols to other long-range models and to dynamics in the presence of stronger decoherence channels

3. Quantum optical photoelectron interferometry

Authors: Jonathan Dubois, Viviane Cotte, Richard Taïeb... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How do quantum statistical properties of light (photon correlations, squeezing, non-classical statistics) imprint themselves on photoelectron observables in multiphoton ionization, and can attosecond interferometry (RABBIT) be generalized to quantum light sources?

Summary: The paper develops a general theory connecting photon statistics of quantum light fields to photoelectron spectra in multiphoton ionization, and applies it to RABBIT attosecond interferometry. It shows that sideband amplitude, contrast, and phase encode the quantum-statistical properties of the driving harmonics and IR fields, validated by numerical simulations using squeezed-coherent IR states.

Key Results: Derived an analytical framework mapping photon autocorrelation and cross-correlation functions directly onto photoelectron spectra. Demonstrated that RABBIT sideband oscillations (amplitude, contrast, phase vs pump-probe delay) encode the quantum nature of the driving field. Validated analytical theory against numerical simulations for classical harmonics + squeezed-coherent IR field with 'excellent agreement.' Covered multiple regimes: correlated IR/harmonic modes, uncorrelated modes with non-classical harmonics.

Key Findings:

Photon auto- and cross-correlation functions map directly onto photoelectron spectral features
RABBIT sideband contrast and phase carry signatures of the light field's quantum statistics, distinguishing classical, squeezed, and non-classical configurations
Interplay between classical and quantum correlations between IR and harmonic modes governs photoemission coherence
Analytical predictions agree quantitatively with numerical simulations for squeezed-coherent IR + classical harmonics

Technical Novelty: First general theoretical framework directly linking arbitrary photon statistics (g^(1), g^(2), cross-correlations) to photoelectron observables in multiphoton/attosecond regimes. Prior RABBIT theory assumed classical coherent fields; this work treats both pump and probe as fully quantum and reveals statistics-dependent sideband contrast/phase — a genuinely new observable channel.

What's New: Bridges quantum optics and attosecond science by treating the driving field as fully quantum within RABBIT — a regime historically described with semi-classical fields. Introduces statistics-resolved sideband observables as a probe of light's quantum nature.

Extension Opportunities:

Implement experimental quantum-light RABBIT using squeezed IR sources (e.g., OPA-generated squeezed vacuum) to measure predicted contrast/phase signatures and validate the theory in a lab
Extend the framework to entangled photon pairs or Fock-state harmonics to explore whether photoelectron spectra can serve as a quantum-state tomography tool for XUV light
Couple this theory to molecular targets to investigate how vibrational/electronic dynamics interact with non-classical light statistics, opening quantum-enhanced attosecond chemistry

Replicability: No code or data availability mentioned in the abstract. Reproduction would require a TDSE/perturbative photoionization solver coupled to quantum-optical field representations (squeezed/coherent states). Compute is modest — likely single-workstation or small-cluster scale for the single-atom RABBIT simulations described.

Research Gaps:

No experimental demonstration yet — feasibility with current squeezed-XUV or squeezed-IR sources remains open
Extension to many-electron or molecular targets, and to entangled/Fock-state harmonics, is not addressed

⚡ ENERGY

1. Mathematical Modeling of HDV RNA, HBV DNA, and HBsAg Dynamics during Lonafarnib-Based Therapy: Insights from the LOWR HDV-1 Study

Authors: Adquate Mhlanga, Louis Shekhtman, Rami Zakh... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How do HDV RNA, HBV DNA, and HBsAg dynamics evolve during Lonafarnib (LNF)-based therapy in HBV/HDV coinfected patients, and what mechanisms explain observed viral kinetics including biphasic decline, viral breakthrough, and paradoxical HBV increases?

Summary: The paper presents a mathematical kinetic model fitted to 15 HBV/HDV coinfected patients receiving Lonafarnib-based regimens, quantifying HDV half-life (~1.26 days) and treatment efficacy (94%, rising to 98.9% in phase 2). It mechanistically explains paradoxical HBV DNA increases via release of HDV-mediated inhibition and shows HBsAg remains stable due to a constant pool of producing cells.

Key Results: Developed a mathematical model fitted to 15 HBV/HDV coinfected patients from the LOWR HDV-1 study. Estimated HDV RNA serum half-life of 1.26 days (95% CI: 1.05–1.47), treatment efficacy of 94% in inhibiting HDV RNA production across all regimens (95% CI: 89%–97%), with 2nd-phase efficacy reaching 98.9%. Quantified a median 4-fold (IQR: 1–28) increase in HBV DNA production rate when HDV declined below an inhibitory threshold. Showed HBsAg stability was explained by constant HBsAg-producing cell counts.

Key Findings:

HDV RNA serum half-life is approximately 1.26 days with LNF efficacy of ~94%, with a second phase reaching 98.9% efficacy via time-dependent increase
HBV DNA rises ~4-fold (median) when HDV falls below an inhibitory threshold, revealing HDV's suppressive effect on HBV replication
LNF combination with ritonavir or PEG-IFNα produces biphasic HDV decline without viral breakthrough, while LNF monotherapy yields flat-partial response often followed by breakthrough
HBsAg stability during treatment is mechanistically explained by a constant number of HBsAg-producing cells, independent of HDV decline

Technical Novelty: First mathematical model that jointly captures HDV, HBV, and HBsAg kinetics under LNF-based therapy, introducing a time-dependent efficacy term to explain biphasic HDV decline and a novel HDV-inhibitory-threshold mechanism that triggers HBV rebound — moving beyond standard single-virus viral kinetic models.

What's New: Unlike prior single-virus kinetic models, this work jointly models all three viral markers (HDV, HBV, HBsAg) under a drug that selectively targets HDV, isolating HDV's modulatory effect on HBV — and introduces a time-varying efficacy term plus an HDV-threshold mechanism for HBV rebound.

Extension Opportunities:

Extend the model to incorporate intracellular HDV/HBV replication dynamics and link in vitro prenylation inhibition data to in vivo efficacy parameters
Apply the model framework to newer HDV therapies (bulevirtide, nucleic acid polymers) to compare mechanism-of-action signatures and stratify responders
Build a clinical decision-support tool that uses early HDV kinetic markers (1st-phase slope, time to plateau) to predict viral breakthrough risk and guide regimen switching

Replicability: No code or data repository mentioned in the abstract. Reproduction would require access to the LOWR HDV-1 clinical trial dataset (serum HBV DNA, HDV RNA, HBsAg measurements for 15 patients). Compute requirements are modest — nonlinear mixed-effects ODE fitting on a standard workstation (e.g., Monolix, NONMEM, or R/Python).

Research Gaps:

Sample size of 15 patients limits generalizability; mechanistic basis for the time-dependent efficacy increase remains speculative
Model does not resolve intracellular dynamics or immune-mediated clearance, leaving the source of HDV inhibition on HBV mechanistically unclear

2. Micron-sized magnonic 3-port rectilinear circulator

Authors: Kevin S. Weber, Loic Temdie, Vincent Castel... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can non-reciprocal microwave devices (circulators) be miniaturized to micron-scale while remaining compatible with integrated circuit architectures, overcoming the bulky size of conventional ferrite-based circulators?

Summary: The authors present the first experimental micron-scale 3-port magnon circulator that uses chiral spin-wave excitation through nanowire gratings to route signals unidirectionally between three rectilinear channels. The device operates in a narrow band tunable across 2-8 GHz via external field (up to 100 mT) and grating dimensions, offering a path toward integrated non-reciprocal microwave components.

Key Results: First experimental characterization of a micron-sized 3-port magnon circulator using chiral spin-wave excitation via nanowire gratings. Demonstrated genuine spin-wave circulation between three rectilinear ports via full 3-port spin-wave spectroscopy. Tunable narrow-band operation across 2-8 GHz frequency range via external magnetic fields up to 100 mT and grating geometry that sets wavevectors.

Key Findings:

Chiral excitation via nanowire gratings can produce unidirectional rectilinear spin-wave beams suitable for circulator operation
Full 3-port spectroscopy confirms genuine circulation between ports at the micron scale
Operating frequency is broadly tunable (2-8 GHz) via field strength and grating geometry, enabling design flexibility

Technical Novelty: Novel rectilinear 3-channel circulator topology using chiral spin-wave excitation by nanowire gratings — exploits wavevector-selective unidirectional magnon emission rather than ferrite Faraday rotation. Replaces bulky ferrite junctions with a planar, lithographically defined micron-scale magnonic device.

What's New: Unlike conventional ferrite circulators that rely on Faraday rotation in bulky junctions, this work uses magnonic chirality with lithographically patterned nanowire gratings to achieve non-reciprocity in a planar, micron-scale geometry compatible with IC integration. The rectilinear 3-channel topology is itself an original design.

Extension Opportunities:

Integrate the magnon circulator with on-chip superconducting qubit architectures for quantum signal routing in cryogenic environments
Extend to broadband operation by cascading multiple gratings with different wavevectors or designing chirped/multi-resonant nanowire arrays
Develop reconfigurable/switchable variants using voltage-controlled magnetic anisotropy (VCMA) instead of external field tuning for low-power CMOS-compatible operation

Replicability: No code/data availability mentioned in the abstract. Reproduction requires a magnonics lab: YIG (or similar low-damping magnetic) thin films, e-beam lithography for nanowire gratings, vector network analyzer for spin-wave spectroscopy, and electromagnet capable of ~100 mT fields. Significant experimental infrastructure required; not computationally reproducible.

Research Gaps:

Insertion loss, isolation, and bandwidth performance metrics not specified in the abstract — quantitative benchmarking vs. conventional circulators is needed
Room-temperature CMOS-compatible operation without external bias fields remains unaddressed; the 100 mT field requirement is impractical for on-chip integration

3. Bandedge-state-limited single-photon emission from volumetric quantum design of 2D colloidal quantum wells

Authors: Xiao Liang, Bo Wang, Yue Yu... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can solution-processable single-photon sources overcome the particle-to-particle heterogeneity of 3D colloidal quantum dots while simultaneously solving the multiexciton emission and surface-exciton susceptibility problems that have historically prevented 2D colloidal quantum wells (CQWs) from serving as reliable single-photon emitters at room temperature?

Summary: This paper introduces volumetric quantum design (VQD) for 2D colloidal quantum wells, producing room-temperature single-photon emitters with exceptional uniformity, near-unity on-time (>99.5%), strong antibunching (g(2)(0)=0.041), and high linear polarization (73%) without cavity enhancement. By combining lateral excitonic confinement with a thick strain-relieved barrier shell, the design eliminates the multiexciton and surface-susceptibility issues that previously disqualified CQWs as single-photon sources, establishing them as a homogeneous, scalable alternative to colloidal quantum dots.

Key Results: The authors demonstrate that volumetric quantum design (VQD) of 2D CQWs achieves: (1) near-blinking-free emission with on-time >99.5%, (2) fluence-insensitive antibunching with g(2)(0) = 0.041 (well below the 0.5 single-photon threshold), (3) linear polarization up to 73% under cavity-free conditions, and (4) a bandedge-state-filling bottleneck that protects single-photon purity. These are verified via statistical single-particle spectroscopy at room temperature.

Key Findings:

VQD-CQWs achieve fluence-insensitive single-photon emission with g(2)(0) = 0.041, protected by a bandedge-state-filling bottleneck rather than fast Auger recombination
Near-blinking-free behavior (>99.5% on-time) at room temperature, addressing a chronic instability of colloidal emitters
High intrinsic linear polarization (up to 73%) arises from synergistic transition-dipole and electric-field anisotropies without external photonic structuring

Technical Novelty: The volumetric quantum design (VQD) concept itself — simultaneously laterally confining the bandedge excitonic domain within the exciton coherent area AND vertically decoupling it from surface states via a thick, strain-relieved quantum-barrier shell. This overcomes the long-standing thickness-confinement trade-off that has limited 2D CQWs, enabling a single, highly localized bandedge state with a state-filling bottleneck that suppresses multiexciton emission — something not achieved in prior CQW work.

What's New: Unlike 3D colloidal quantum dots which suffer particle-to-particle heterogeneity, and unlike prior 2D CQWs which suffered efficient multiexciton emission and surface losses, VQD is the first design to deliver a single highly localized bandedge state in a 2D platform by reconciling the thickness vs. confinement trade-off through a strain-relieved quantum-barrier shell — yielding uniform, reproducible, cavity-free single photons.

Extension Opportunities:

Integrate VQD-CQWs into on-chip photonic cavities (e.g., micro-pillar or photonic crystal cavities) to boost extraction efficiency and indistinguishability for scalable quantum networks
Extend VQD synthesis recipes to other II-VI/III-V material systems (e.g., InP/ZnS or perovskite nanoplatelets) to access telecom-wavelength single-photon emission for fiber-based quantum communication
Develop deterministic patterning/positioning of VQD-CQWs onto waveguides or electrically-driven LED structures to build solution-processable, electrically-pumped single-photon devices

Replicability: The abstract does not mention public code or data availability. Reproduction requires colloidal synthesis facilities (Schlenk lines, glovebox), advanced shell-growth chemistry for strain-relieved quantum barriers, and a single-particle spectroscopy setup with Hanbury Brown–Twiss interferometer, time-correlated single-photon counting, and polarization-resolved optics. No heavy compute needed; the barrier is wet-chemistry expertise and optical instrumentation.

Research Gaps:

Indistinguishability and Hong-Ou-Mandel two-photon interference visibility of VQD-CQW emission are not addressed, which is critical for quantum information applications
Performance at cryogenic temperatures, integration with photonic cavities, and electrical injection pathways remain to be demonstrated

🔬 MATERIALS

1. Band-Selective Tunneling and Anisotropic Multiband Superconductivity in V$_2$Ga$_5$

Authors: Jozef Haniš, Jozef Kačmarčík, Filip Košuth... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can directional pairing characteristics in multiband superconductors with structural anisotropy be disentangled, and why do directional tunneling experiments show apparent reversals of bulk gap hierarchies in V$_2$Ga$_5$?

Summary: The paper presents a combined thermodynamic, spectroscopic, and first-principles study of the tetragonal superconductor V$_2$Ga$5$ ($T_c \approx 3.5$ K), introducing an anisotropic multiband $s$-wave ($A{1g}$) pairing model that reconciles specific heat and upper critical field data. It demonstrates that apparent gap hierarchy reversals in directional STS arise from band-selective tunneling shaped by Fermi velocity hot spots and Fermi surface topology, not from intrinsic gap magnitudes.

Key Results: Combined AC calorimetry, directional low-temperature STS, and DFT on tetragonal V$2$Ga$5$ ($T_c \approx 3.5$ K). Built a self-consistent anisotropic multiband singlet $s$-wave ($A{1g}$) pairing model that reconciles specific heat and upper critical field $H{c2}$ anomalies, and showed that the apparent reversal of bulk gap hierarchies in directional tunneling arises from band-selective tunneling driven by Fermi velocity hot spots and Fermi surface topology rather than raw gap magnitudes.

Key Findings:

An anisotropic multiband singlet $s$-wave model in the $A_{1g}$ representation self-consistently reproduces both specific heat and upper critical field anomalies in V$_2$Ga$_5$
Directional tunneling spectra show an apparent reversal of bulk gap hierarchies that is explained by band-selective tunneling rather than by changes in the underlying gaps
Fermi velocity hot spots combined with Fermi surface topology govern which bands dominate tunneling along a given crystallographic direction

Technical Novelty: Integration of first-principles Fermi velocity 'hot spot' analysis with directional STS to explain band-selective tunneling, plus a self-consistent anisotropic multiband $A_{1g}$ singlet $s$-wave model that simultaneously fits specific heat and $H_{c2}$ anisotropy in V$_2$Ga$_5$.

What's New: Provides a microscopic, direction-resolved explanation of how Fermi velocity anisotropy filters which bands contribute to tunneling — reframing observed gap hierarchies as transport-constraint artifacts rather than thermodynamic features, in a previously underexplored intermetallic superconductor.

Extension Opportunities:

Apply the band-selective tunneling framework to other uniaxial multiband superconductors (e.g., MgB$_2$, layered pnictides) to test generality of the Fermi velocity hot spot mechanism
Develop an open-source simulation toolkit that combines DFT Fermi surface outputs with directional tunneling matrix elements to predict STS spectra from first principles
Extend the $A_{1g}$ multiband model to include impurity scattering or magnetic field rotation studies to probe gap nodes and pairing symmetry boundaries

Replicability: No code or dataset availability is mentioned in the abstract. Reproduction would require DFT compute (modest HPC cluster for electronic structure and Fermi surface), a dilution-refrigerator-equipped low-temperature STM with directional capability, AC calorimetry apparatus, and single-crystal V$_2$Ga$_5$ samples — i.e., specialized experimental infrastructure rather than commodity compute.

Research Gaps:

Lack of a general framework linking Fermi surface anisotropy to directional STS observables across diverse multiband superconductor families
Limited understanding of how band-selective tunneling effects modify interpretation of unconventional pairing signatures in other uniaxial systems

2. Andreev Reflection to Probe Momentum-Dependent Spin Polarization in Altermagnet CrSb

Authors: Yan Zhang, Yixuan Luo, Yue Yang... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can momentum-dependent spin polarization in altermagnets — predicted by their k-dependent spin-splitted band structure with zero net magnetization — be directly experimentally probed, and does CrSb exhibit such anisotropic spin textures distinct from conventional antiferromagnets?

Summary: The paper introduces Andreev reflection via Nb-tip point-contact spectroscopy as a new experimental paradigm for probing momentum-dependent spin polarization in altermagnets, applied to CrSb single crystals. The authors measure distinct spin polarizations along three crystallographic planes (73.4%, 67.9%, 61.9%) and resolve altermagnetic domains of 250–500 nm, providing strong evidence for g-wave altermagnetic order.

Key Results: Using mechanical point-contact spectroscopy (MPCS) with superconducting Nb tips on CrSb single crystals, the authors extracted spin polarizations of ~73.4% on the (0001) plane, ~67.9% on the (-1-120) plane, and ~61.9% on the (10-10) plane — demonstrating clear crystallographic anisotropy consistent with g-wave altermagnetic splitting. Spatial line-scan conductance spectra further revealed altermagnetic domains of characteristic size 250–500 nm separated by domain walls ~250 nm wide.

Key Findings:

CrSb exhibits strongly anisotropic, plane-dependent spin polarization (73.4%/67.9%/61.9% on (0001)/(-1-120)/(10-10)), consistent with g-wave altermagnetic band splitting rather than conventional antiferromagnetism.
Spatial conductance line-scans reveal altermagnetic domains of 250–500 nm separated by ~250 nm wide domain walls, providing the first MPCS-based real-space evidence of altermagnetic domain structure.
Andreev reflection at superconductor/altermagnet point contacts is established as a viable, momentum-selective probe of k-dependent spin textures despite the zero net magnetization.

Technical Novelty: First use of Andreev reflection spectroscopy as a momentum-selective probe of altermagnetic spin polarization, exploiting the crystallographic-orientation dependence of point contacts to access k-dependent spin textures — a regime previously addressable mainly by spin-ARPES. Also the first MPCS-based imaging of altermagnetic domain structure.

What's New: Unlike prior altermagnet studies that rely on spin-resolved ARPES or transport-symmetry signatures, this work leverages the orientation-dependent geometry of mechanical point contacts to extract momentum-projected spin polarization via Andreev reflection — a technique historically used for ferromagnets but never previously deployed to access altermagnetic k-dependent spin splitting or image its domain structure.

Extension Opportunities:

Apply the same Nb-tip Andreev reflection methodology to other predicted altermagnet candidates (RuO2, MnTe, KV2Se2O) to build a comparative database of momentum-resolved spin polarizations across d-, g-, and i-wave classes.
Combine MPCS with in-situ magnetic field rotation or strain tuning to map full angular dependence of spin polarization and test the predicted nodal structure of g-wave splitting in CrSb.
Pair the technique with cryogenic scanning probe microscopy (SP-STM or MFM) to directly correlate the 250–500 nm domain structure observed in conductance maps with real-space altermagnetic order parameter textures, enabling domain-wall engineering for spintronic devices.

Replicability: No code or data availability is mentioned in the abstract. Reproduction requires high-quality CrSb single crystals oriented along three principal facets, a dilution or He-flow cryostat (sub-Tc of Nb, ~9 K), mechanical point-contact rigs with Nb tips, and standard lock-in differential conductance electronics — accessible to any mid-sized condensed-matter lab but not trivially low-cost.

Research Gaps:

The extracted polarizations represent angular averages over the contact's Fermi-surface projection; a full theoretical model linking MPCS geometry to the g-wave nodal structure is still needed.
The microscopic origin and stability of the observed 250–500 nm domains, and their controllability via field, strain, or growth, remain unexplored.

3. Experimental Design Space Exploration of Ultra-Low Threshold Hybrid III-V/Si Quantum Dot Microring Lasers

Authors: Xucheng Yang, Preston Luong, Yatiraj Ramanujam... Published: 2026-06-11 | Citations: 0 arXiv | PDF

Research Question: How can hybrid III-V/Si quantum dot microring lasers be designed to simultaneously achieve ultra-low threshold current, high wall-plug efficiency, strong thermal stability, and high modulation bandwidth for compact 1.3 μm silicon photonic light sources?

Summary: The paper presents an experimental design-space exploration of hybrid III-V/Si InAs/GaAs quantum dot microring lasers at 1.3 μm, achieving record performance across threshold (<0.8 mA, 109 A/cm²), wall-plug efficiency (~10%), thermal stability (T₀ = 212 K), and bandwidth (5 GHz). It establishes QD microrings on silicon as a viable ultra-low-power light source for silicon photonic integration.

Key Results: Demonstrated InAs/GaAs QD microring lasers on silicon with sub-0.8 mA threshold current, >2 mW output power, ~10% wall-plug efficiency (record), 109 A/cm² threshold current density, characteristic temperature T₀ = 212 K (record, indicating near temperature-independent threshold), and 3-dB modulation bandwidth up to 5 GHz at 1.3 μm emission.

Key Findings:

Sub-0.8 mA threshold with >2 mW output and ~10% WPE — record for compact QD lasers on Si
Characteristic temperature T₀ = 212 K shows near-immunity of threshold to temperature, removing the need for active cooling in many deployments
High differential gain from the QD active region enables direct modulation up to 5 GHz despite the small device footprint

Technical Novelty: Systematic multi-dimensional experimental design-space exploration (rather than single-point optimization) of hybrid III-V/Si QD microrings, combining InAs/GaAs quantum dot gain (low threshold, temperature insensitivity) with silicon microring geometry to simultaneously set records for WPE, T₀, and threshold density on this platform.

What's New: Unlike prior demonstrations that optimize one figure of merit at a time, this work jointly explores a multi-dimensional design space and simultaneously sets records in threshold density, WPE, and T₀ for hybrid III-V/Si QD microring lasers.

Extension Opportunities:

Integrate these microring lasers into wavelength-division-multiplexed silicon photonic transceivers and benchmark BER performance at higher data rates (10-25 Gbps using equalization or PAM4)
Explore even smaller ring radii or coupled-ring/photonic-molecule geometries to push thresholds into the sub-100 µA regime while characterizing mode stability
Extend the multi-dimensional design exploration via Bayesian optimization or ML-driven inverse design across ring radius, coupling gap, QD layer count, and waveguide width to map the full Pareto frontier

Replicability: No code or dataset is mentioned in the abstract; reproduction would require a III-V/Si heterogeneous integration cleanroom (MBE/MOCVD for InAs/GaAs QD epitaxy, wafer bonding to SOI, lithography for microring patterning) plus cryogenic-to-elevated temperature L-I-V and small-signal RF characterization setups — substantial fab infrastructure rather than compute.

Research Gaps:

Long-term reliability, aging, and yield statistics across the design space are not addressed in the abstract
Modulation bandwidth (5 GHz) still trails commercial DFB/EML sources, limiting direct use in high-speed datacom without further design effort

🔥 GitHub Trending

1. 001TMF/harness-forge

⭐ 19 stars | Python

Turn Claude Code into its own Meta-Harness — a skill that evolves the scaffolding around a fixed model (memory, retrieval, context, prompts) via a native propose→score→Pareto loop. Native reimplementa

agent-skills agents claude-code claude-skill llm meta-harness

2. ingridtoulotte/clauditor

⭐ 4 stars | Python

Audit your Claude Code instruction stack — find the rules Claude silently ignores (dead, contradictory, vague, duplicate, bloated). Zero-dep CLI + GitHub Action.

agents ai anthropic claude claude-code claude-md

3. ingridtoulotte/claude-cost-intelligence

⭐ 4 stars | Python

See where LLM money goes, why, and how to cut waste. Local-first FinOps + spend intelligence for Claude Code & the Anthropic API. Pure stdlib, zero deps.

analytics anthropic claude cost-optimization dashboard developer-tools

4. ingridtoulotte/context-compressor

⭐ 4 stars | Python

Lossless-first semantic compression for LLM context windows. Shrink context 60-80% and prove nothing important was lost. Deterministic, extractive, stdlib-only, validated.

ai-agents context-engineering context-window cost-optimization deterministic llm

5. tabtabtabai/awesome-acp

⭐ 3 stars | Python

A curated list of everything in the Agent Client Protocol (ACP) ecosystem — connect any editor to any AI agent.

acp agent-client-protocol ai-agents awesome awesome-list coding-agents

6. silobust-hash/fable-to-opus

⭐ 3 stars | Shell

Adapt-first playbook: help any Claude model pick up Fable's effective work habits by auditing your own harness. miss you, fable 🤍

agentic anthropic claude claude-code claude-code-plugin llm

7. martin1847/evolab

⭐ 3 stars | Shell

A² · Agentic AI × Anything：多 agent 编排方法论与治理 skill 合集（旗舰 cto-orchestration）

agent-skills agentic-ai ai-agents ai-orchestration claude-code-skills developer-tools

8. PeytonNowlin/Nebula

⭐ 2 stars | Rust

A typed language for AI agents — author in Nebula, ship as Python. Rust compiler + JSON toolchain.

agentic ai-agents compiler domain-specific-language llm mcp

9. synthetic-recon/pi-fusion

⭐ 2 stars | TypeScript

Multi-model deliberation for pi, inspired by OpenRouter Fusion

coding-agent fusion llm pi-extension pi-package

10. rohanmistry231/ML-From-Scratch

⭐ 1 stars | Python

A compact collection of classic machine learning algorithms implemented from scratch in Python for learning and experimentation.

algorithms classification from-scratch-in-python machine-learning ml-algorithms python

11. rohanmistry231/AI-From-Scratch

⭐ 1 stars | Unknown

A unified hub for the Scratch Series — 200+ algorithms across ML, DL, NLP, LLM, GenAI, and TimeSeries, all implemented from scratch with NumPy.

ai-from-scratch deep-learning education from-scratch generative-ai interview-prep

12. Sanjana-crypto/phishing-email-detector

⭐ 1 stars | Python

A machine learning-based phishing email detection tool with a Streamlit web interface, built to assist SOC analysts in triaging suspicious emails by classifying them as Safe, Suspicious, or High-Risk

cybersecurity machine-learning nlp phishing-detection python soc-analyst

13. rohanmistry231/DL-From-Scratch

⭐ 1 stars | Python

A compact collection of foundational deep learning architectures implemented from scratch in Python for learning and experimentation.

attention-mechanism deep-learning dl-algorithm from-scratch neural-networks nlp

14. supzammy/EpiRNAC

⭐ 1 stars | Python

EpiRNAC-Scanner - C stands for CNN, a publicly deployed web framework that processes sequences from 41 bases to over 200,000 bases without manual preprocessing. The system introduces Epitranscriptomic

bioinformatics biotechnology cnn deep-learning

15. Saboor-Hamedi/lumina-notes

⭐ 1 stars | Unknown

A comprehensive engineering knowledge base: 400+ interconnected notes across AI/ML, system design, databases, DevOps, security, and software engineering. Built as an Obsidian vault with Mermaid diagra

ai-ml data-science databases deep-learning devops engineering

Generated by Research Pulse on 2026-06-14 06:07