Interesting papers
All published papers whose full text was read, ranked by interest score.
| # | Score | Paper | Year |
|---|---|---|---|
| 1 | 8.7 |
LLM Judges as Raters: A Pre-Registered Audit of Severity, Halo, Reliability, and Version Instability in LLM Essay Scoring on Public Corpora
arXiv · doi:10.48550/arxiv.2608.29517 Novelty 8.0 · Rigor 9.5 · Engineering 9.0 · Consensus 8.0 · Topic 10.0 |
2026 |
| 2 | 8.1 |
One Percent of the Bank: Adaptive Item Selection Reads Targeted Contamination Up While the Whole Bank Barely Moves
Zenodo (CERN European Organization for Nuclear Research) · doi:10.5281/zenodo.23156543 Novelty 8.0 · Rigor 8.5 · Engineering 7.5 · Consensus 8.0 · Topic 9.0 |
2026 |
| 3 | 8.0 |
Agreement Metrics for LLM-as-Judge Evaluation: What to Report and Why
arXiv · doi:10.48550/arxiv.2606.00093 Novelty 7.5 · Rigor 8.5 · Engineering 7.5 · Consensus 8.0 · Topic 8.5 |
2026 |
| 4 | 8.0 |
StyleShield: Exposing the Fragility of AIGC Detectors through Continuous Controllable Style Transfer
arXiv · doi:10.48550/arxiv.2605.00924 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 8.0 · Topic 9.0 |
2026 |
| 5 | 8.0 |
From 0-Order Selection to 2-Order Judgment: Combinatorial Hardening Exposes Compositional Failures in Frontier LLMs
arXiv · doi:10.48550/arxiv.2605.07268 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 8.0 · Topic 8.0 |
2026 |
| 6 | 8.0 |
Every Model Cheats: Prompt-Level Mitigation of Cheating on Offensive Cyber Tasks
arXiv · doi:10.48550/arxiv.2607.21763 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 8.0 · Topic 9.0 |
2026 |
| 7 | 8.0 |
Laplace-PSN-IRT: Uncertainty Quantification for Neural Item Response Theory Models of LLM Benchmarks
arXiv · doi:10.48550/arxiv.2607.25257 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 8.0 · Topic 9.0 |
2026 |
| 8 | 8.0 |
A Judge Should Know What Changed:Construct Validity for LLM-as-a-Judge Evaluation
arXiv · doi:10.48550/arxiv.2608.24419 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 8.0 · Topic 9.0 |
2026 |
| 9 | 7.8 |
Rubrics Carry the Standard: Shapley Attribution for Checklist-Based LLM Judges
arXiv · doi:10.48550/arxiv.2610.24971 Novelty 8.0 · Rigor 7.5 · Engineering 8.0 · Consensus 8.0 · Topic 7.0 |
2026 |
| 10 | 7.8 |
Detection of two-way outliers in multivariate data and application to cheating detection in educational tests
The Annals of Applied Statistics · doi:10.1214/21-aoas1564 Novelty 7.0 · Rigor 9.0 · Engineering 8.0 · Consensus 6.0 · Topic 9.0 |
2022 |
| 11 | 7.7 |
Diagnosing the Reliability of LLM-as-a-Judge via Item Response Theory
arXiv · doi:10.48550/arxiv.2602.00521 Novelty 7.5 · Rigor 8.5 · Engineering 7.0 · Consensus 7.5 · Topic 9.0 |
2026 |
| 12 | 7.7 |
Correct codes for the wrong reasons? validating LLMs as measurement instruments for theoretical constructs
arXiv · doi:10.48550/arxiv.2606.28574 Novelty 8.0 · Rigor 7.0 · Engineering 7.0 · Consensus 8.0 · Topic 9.0 |
2026 |
| 13 | 7.7 |
GAUGE: When Not to Trust LLM-as-a-Judge in User-Simulated Evaluation of Task-Oriented Agents
arXiv · doi:10.48550/arxiv.2609.12191 Novelty 8.0 · Rigor 8.0 · Engineering 7.0 · Consensus 7.0 · Topic 9.0 |
2026 |
| 14 | 7.6 |
Personality Without Persons? A Psychometric Critique of Big Five Testing in Large Language Models
arXiv · doi:10.48550/arxiv.2607.02325 Novelty 8.0 · Rigor 8.0 · Engineering 5.0 · Consensus 8.0 · Topic 8.0 |
2026 |
| 15 | 7.6 |
Faster AI, Uneven Frontier: Rapid Crossings, a Jagged Frontier, and the Repositioning of Human Judgment
arXiv · doi:10.48550/arxiv.2607.12125 Novelty 8.0 · Rigor 8.0 · Engineering 5.0 · Consensus 8.0 · Topic 9.0 |
2026 |
| 16 | 7.5 |
Latent Impact and Differential Item Functioning Analysis for Asymmetric IRT Models
arXiv · doi:10.48550/arxiv.2605.05684 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 7.0 |
2026 |
| 17 | 7.5 |
Do Coding Agents Deceive Us? Detecting and Preventing Cheating via Capped Evaluation with Randomized Tests
arXiv · doi:10.48550/arxiv.2606.07379 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 8.0 |
2026 |
| 18 | 7.5 |
CurveShift: Is Agent Progress Scalar? Separating Level from Shape
arXiv · doi:10.48550/arxiv.2608.00355 Novelty 7.0 · Rigor 8.0 · Engineering 8.0 · Consensus 7.0 · Topic 9.0 |
2026 |
| 19 | 7.5 |
Fighting Fire with Fire: On the Feasibility of Protecting Exercises Against AI Cheating
arXiv · doi:10.48550/arxiv.2608.01112 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 9.0 |
2026 |
| 20 | 7.5 |
Every Wrong Answer Counts: Option-Level Psychometrics for LLM Multiple-Choice Benchmarks
arXiv · doi:10.48550/arxiv.2608.02966 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 8.0 |
2026 |
| 21 | 7.5 |
EduClaw-Bench: A Long-Horizon Benchmark for Pedagogical LLM Agents with Simulated Learners
arXiv · doi:10.48550/arxiv.2608.03206 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 8.0 |
2026 |
| 22 | 7.5 |
Item Response Theory for AI Safety
arXiv · doi:10.48550/arxiv.2608.05086 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 8.0 |
2026 |
| 23 | 7.5 |
Every Token Counts: Exact Likert-Scale Distributions for Measuring LLM Attitudes and Biases
arXiv · doi:10.48550/arxiv.2608.10503 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 8.0 |
2026 |
| 24 | 7.5 |
A Dual-Dimensional LLM Framework for Automated Item Incidental Content Similarity Analysis in Large-Scale Assessments
arXiv · doi:10.48550/arxiv.2608.24825 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 9.0 |
2026 |
| 25 | 7.5 |
Check The Scoreboard: An Analysis of Scoring Schemes on Multiple-Choice Evaluation
arXiv · doi:10.48550/arxiv.2608.29887 Novelty 8.0 · Rigor 8.0 · Engineering 8.0 · Consensus 5.0 · Topic 8.0 |
2026 |