Benchmark

Weekly scoreboard of top papers by Signal Fusion. Data is updated weekly by our pipeline.

Week of 2026-03-02 · v1

Rank	Paper	Viability	Composite
1	RoboPocket: Improve Robot Policies Instantly with Your Phone	7.0	70.0
2	POET-X: Memory-efficient LLM Training by Scaling Orthogonal Transformation	3.0	30.0
3	The Spike, the Sparse and the Sink: Anatomy of Massive Activations and Attention Sinks	2.0	20.0
4	Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation	4.0	40.0
5	Reasoning Theater: Disentangling Model Beliefs from Chain-of-Thought	2.0	20.0
6	Towards Provably Unbiased LLM Judges via Bias-Bounded Evaluation	5.0	50.0
7	SurvHTE-Bench: A Benchmark for Heterogeneous Treatment Effect Estimation in Survival Analysis	4.0	40.0
8	Leveraging LLM Parametric Knowledge for Fact Checking without Retrieval	4.0	40.0
9	Distributed Partial Information Puzzles: Examining Common Ground Construction Under Epistemic Asymmetry	4.0	40.0
10	RealWonder: Real-Time Physical Action-Conditioned Video Generation	7.0	70.0
11	Residual RL--MPC for Robust Microrobotic Cell Pushing Under Time-Varying Flow	5.0	50.0
12	Planning in 8 Tokens: A Compact Discrete Tokenizer for Latent World Model	6.0	60.0
13	SAIL: Similarity-Aware Guidance and Inter-Caption Augmentation-based Learning for Weakly-Supervised Dense Video Captioning	6.0	60.0
14	Ensembling Language Models with Sequential Monte Carlo	3.0	30.0
15	RelaxFlow: Text-Driven Amodal 3D Generation	7.0	70.0
16	MobileFetalCLIP: Selective Repulsive Knowledge Distillation for Mobile Fetal Ultrasound Analysis	6.0	60.0
17	Dissociating Direct Access from Inference in AI Introspection	2.0	20.0
18	Judge Reliability Harness: Stress Testing the Reliability of LLM Judges	5.0	50.0
19	Legal interpretation and AI: from expert systems to argumentation and LLMs	2.0	20.0
20	Learning Causal Structure of Time Series using Best Order Score Search	3.0	30.0
21	PACE: A Personalized Adaptive Curriculum Engine for 9-1-1 Call-taker Training	7.0	70.0
22	Ailed: A Psyche-Driven Chess Engine with Dynamic Emotional Modulation	5.0	50.0
23	Building AI Coding Agents for the Terminal: Scaffolding, Harness, Context Engineering, and Lessons Learned	4.0	40.0
24	GALACTIC: Global and Local Agnostic Counterfactuals for Time-series Clustering	6.0	60.0
25	PersianPunc: A Large-Scale Dataset and BERT-Based Approach for Persian Punctuation Restoration	7.0	70.0
26	Latent-Mark: An Audio Watermark Robust to Neural Resynthesis	5.0	50.0
27	Med-V1: Small Language Models for Zero-shot and Scalable Biomedical Evidence Attribution	7.0	70.0
28	UniSTOK: Uniform Inductive Spatio-Temporal Kriging	6.0	60.0
29	WavSLM: Single-Stream Speech Language Modeling via WavLM Distillation	7.0	70.0
30	WebChain: A Large-Scale Human-Annotated Dataset of Real-World Web Interaction Traces	5.0	50.0
31	STRUCTUREDAGENT: Planning with AND/OR Trees for Long-Horizon Web Tasks	6.0	60.0
32	X-RAY: Mapping LLM Reasoning Capability via Formalized and Calibrated Probes	5.0	50.0
33	Whispering to a Blackbox: Bootstrapping Frozen OCR with Visual Prompts	4.0	40.0
34	GCAgent: Enhancing Group Chat Communication through Dialogue Agents System	4.0	40.0
35	Reclaiming Lost Text Layers for Source-Free Cross-Domain Few-Shot Learning	6.0	60.0
36	Recursive Inference Machines for Neural Reasoning	5.0	50.0
37	Boosting ASR Robustness via Test-Time Reinforcement Learning with Audio-Text Semantic Rewards	6.0	60.0
38	Not All Trust is the Same: Effects of Decision Workflow and Explanations in Human-AI Decision Making	2.0	20.0
39	The Geometric Inductive Bias of Grokking: Bypassing Phase Transitions via Architectural Topology	5.0	50.0
40	AI+HW 2035: Shaping the Next Decade	3.0	30.0
41	SPyCer: Semi-Supervised Physics-Guided Contextual Attention for Near-Surface Air Temperature Estimation from Satellite Imagery	7.0	70.0
42	KARL: Knowledge Agents via Reinforcement Learning	7.0	70.0
43	Early Warning of Intraoperative Adverse Events via Transformer-Driven Multi-Label Learning	6.0	60.0
44	Balancing Coverage and Draft Latency in Vocabulary Trimming for Faster Speculative Decoding	3.0	30.0
45	Stable-LoRA: Stabilizing Feature Learning of Low-Rank Adaptation	7.0	70.0
46	Logi-PAR: Logic-Infused Patient Activity Recognition via Differentiable Rule	7.0	70.0
47	Guidelines for the Annotation and Visualization of Legal Argumentation Structures in Chinese Judicial Decisions	3.0	30.0
48	C2-Faith: Benchmarking LLM Judges for Causal and Coverage Faithfulness in Chain-of-Thought Reasoning	5.0	50.0
49	Lifelong Language-Conditioned Robotic Manipulation Learning	4.0	40.0
50	SSR-GS: Separating Specular Reflection in Gaussian Splatting for Glossy Surface Reconstruction	7.0	70.0

Methodology

Rankings use Signal Fusion: (viability_score × 10) + (unicorn_probability × 30) + (votes × 0.1) + (star_velocity × 5). See our FAQ for full scoring definition.