Skip to content
← Back to explorer

HFEPX Hub

CS.AI + General Papers

Updated from current HFEPX corpus (Apr 9, 2026). 314 papers are grouped in this hub page.

Read Full Context

Updated from current HFEPX corpus (Apr 9, 2026). 314 papers are grouped in this hub page. Common evaluation modes: Automatic Metrics, Simulation Env. Most common rater population: Domain Experts. Common annotation unit: Trajectory. Frequent quality control: Calibration. Frequently cited benchmark: HotpotQA. Common metric signal: accuracy. Use this page to compare protocol setup, judge behavior, and labeling design decisions before running new eval experiments. Newest paper in this set is from Mar 22, 2026.

Papers: 314 Last published: Mar 22, 2026 Global RSS Tag RSS
Cs.AIGeneral

Researcher Quick Triage

This hub is best used for protocol triage and replication planning from abstract-level evidence. Quality band: High .

Analysis blocks below are computed from the currently loaded sample (60 of 314 total papers in this hub).

High-Signal Coverage

100.0%

60 / 60 sampled papers are not low-signal flagged.

Replication-Ready Set

14

Benchmark + metric + eval mode explicitly present.

Judge/Human Comparability

3

Papers containing both `human_eval` and `llm_as_judge`.

  • 14 papers are replication-ready (benchmark + metric + explicit evaluation mode).
  • 3 papers support judge-vs-human agreement analysis.
  • 9 papers report explicit quality controls (calibration/adjudication/IAA).

Primary action: Start with the top 2 papers in “Start Here”, then validate assumptions in the protocol matrix.

Currently showing only replication-ready papers in ranking and matrix sections (14 papers).

Need evaluators for this research workflow?

Post a Job →

Why This Matters For Eval Research

  • 62.1% of papers report explicit human-feedback signals, led by pairwise preferences.
  • automatic metrics appears in 47.8% of papers in this hub.
  • HotpotQA is a recurring benchmark anchor for cross-paper comparisons in this page.

Protocol Takeaways

  • 2 sampled papers report both human evaluation and LLM-as-judge, supporting direct agreement checks.
  • Most common quality-control signal is rater calibration (2.9% of papers).
  • Rater context is mostly domain experts, and annotation is commonly trajectory-level annotation; use this to scope replication staffing.

Benchmark Interpretation

  • HotpotQA appears in 1.6% of hub papers (5/314); use this cohort for benchmark-matched comparisons.
  • DROP appears in 1% of hub papers (3/314); use this cohort for benchmark-matched comparisons.

Metric Interpretation

  • accuracy is reported in 23.2% of hub papers (73/314); compare with a secondary metric before ranking methods.
  • cost is reported in 10.5% of hub papers (33/314); compare with a secondary metric before ranking methods.
Researcher Checklist (Expanded)

Researcher Checklist

  • Strong: Papers with explicit human feedback

    Coverage is strong (62.1% vs 45% target).

  • Gap: Papers reporting quality controls

    Coverage is a replication risk (5.4% vs 30% target).

  • Gap: Papers naming benchmarks/datasets

    Coverage is a replication risk (17.5% vs 35% target).

  • Strong: Papers naming evaluation metrics

    Coverage is strong (49.7% vs 35% target).

  • Gap: Papers with known rater population

    Coverage is a replication risk (12.4% vs 35% target).

  • Moderate: Papers with known annotation unit

    Coverage is usable but incomplete (29.6% vs 35% target).

Strengths

  • Strong human-feedback signal (62.1% of papers).
  • Contains both human-eval and LLM-as-judge protocols for head-to-head methodology comparison.
  • Agentic evaluation appears in 44.9% of papers.

Known Gaps

  • Only 5.4% of papers report quality controls; prioritize calibration/adjudication evidence.
  • Rater population is under-specified (12.4% coverage).
  • Benchmark coverage is thin (17.5% of papers mention benchmarks/datasets).

Suggested Next Analyses

  • Compare papers that report both human_eval and llm_as_judge to quantify judge-human agreement drift.
  • Stratify by benchmark (HotpotQA vs DROP) before comparing methods.
  • Track metric sensitivity by reporting both accuracy and cost.
Recommended Queries (Expanded)

Recommended Queries

Start with These 3

Use these when you need one protocol anchor, one benchmark anchor, and one recent comparison point before reading the wider hub.

Start Here (Best First 6)

Ranked for protocol completeness (human signal, benchmark + metric anchors, quality controls, and judge/human overlap).

Protocol Matrix (Top 12)

Use this to quickly compare protocol ingredients instead of scanning long prose.

Paper HF Signal Eval Modes Benchmarks Metrics QC
AgentHER: Hindsight Experience Replay for LLM Agent Trajectory Relabeling

Mar 22, 2026

Yes Human Eval , Llm As Judge WebArena , ToolBench Precision , Pass@1 Not Reported
PoSh: Using Scene Graphs To Guide LLMs-as-a-Judge For Detailed Image Descriptions

Oct 21, 2025

Yes Human Eval , Llm As Judge CAPArena Spearman Not Reported
SCOPE: Selective Conformal Optimized Pairwise LLM Judging

Feb 13, 2026

Yes Automatic Metrics MT Bench , LMSYS Chatbot Arena Error rate Calibration
TraceSafe: A Systematic Assessment of LLM Guardrails on Multi-Step Tool-Calling Trajectories

Apr 8, 2026

Yes Automatic Metrics Tracesafe Bench Accuracy Not Reported
SODIUM: From Open Web Data to Queryable Databases

Mar 19, 2026

Yes Automatic Metrics Sodium Bench Accuracy Not Reported
When AI Meets Early Childhood Education: Large Language Models as Assessment Teammates in Chinese Preschools

Mar 25, 2026

Yes Automatic Metrics Interaction2eval Agreement , Cost Not Reported
Stabilizing Rubric Integration Training via Decoupled Advantage Normalization

Mar 27, 2026

Yes Automatic Metrics Olympiadbench Accuracy Not Reported
DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment

Mar 23, 2026

Yes Automatic Metrics MT Bench , AlpacaEval Accuracy Not Reported
Elo-Evolve: A Co-evolutionary Framework for Language Model Alignment

Feb 14, 2026

Yes Automatic Metrics MT Bench , AlpacaEval Elo Not Reported
LMUnit: Fine-grained Evaluation with Natural Language Unit Tests

Dec 17, 2024

Yes Human Eval Biggenbench , Rewardbench Agreement Inter Annotator Agreement Reported
Critique-GRPO: Advancing LLM Reasoning with Natural Language and Numerical Feedback

Jun 3, 2025

Yes Automatic Metrics AIME Pass@1 Not Reported
VerifyBench: Benchmarking Reference-based Reward Systems for Large Language Models

May 21, 2025

Yes Automatic Metrics Verifybench Accuracy Not Reported

Protocol Diff (Top Papers)

Fast side-by-side comparison for the highest-ranked papers in this hub.

Signal AgentHER: Hindsight Experience Replay for LLM Agent… PoSh: Using Scene Graphs To Guide LLMs-as-a-Judge F… SCOPE: Selective Conformal Optimized Pairwise LLM J…
Human Feedback DemonstrationsRubric RatingPairwise Preference
Evaluation Modes Human Eval, Llm As JudgeHuman Eval, Llm As JudgeAutomatic Metrics
Benchmarks WebArena, ToolBenchCAPArenaMT Bench, LMSYS Chatbot Arena
Metrics Precision, Pass@1SpearmanError rate
Quality Controls Not reportedNot reportedCalibration
Rater Population UnknownDomain ExpertsUnknown
Annotation Unit TrajectoryMulti Dim RubricPairwise
Suggested Reading Order (Extended)

This section is intentionally expanded only when needed; use “Start Here” above for a faster pass.

Suggested Reading Order

  1. From Intuition to Calibrated Judgment: A Rubric-Based Expert-Panel Study of Human Detection of LLM-Generated Korean Text

    Start here for detailed protocol reporting and quality-control evidence. Signals: automatic metrics + rubric ratings. Focus: accuracy. Abstract: We present LREAD, a Korean-specific instantiation of a rubric-based expert-calibration.

  2. 37e66347-dcaf-4178-8b3b-169baef9860d

    High citation traction makes this a strong baseline for protocol comparison.

  3. TraceSafe: A Systematic Assessment of LLM Guardrails on Multi-Step Tool-Calling Trajectories

    High citation traction makes this a strong baseline for protocol comparison. Signals: automatic metrics + red-team protocols. Focus: Tracesafe-Bench / accuracy. Abstract: As large language models (LLMs) evolve.

  4. 1561e4a4-8626-43d6-ad64-5e297062a260

    High citation traction makes this a strong baseline for protocol comparison.

  5. AgentHER: Hindsight Experience Replay for LLM Agent Trajectory Relabeling

    Include a human-eval paper to calibrate against judge-based evaluation settings. Signals: human evaluation + demonstration data. Focus: WebArena / precision. Abstract: AgentHER realises this idea through a four-stage.

  6. LMUnit: Fine-grained Evaluation with Natural Language Unit Tests

    Include a human-eval paper to calibrate against judge-based evaluation settings. Signals: human evaluation + pairwise preferences. Focus: Biggenbench / agreement. Abstract: As language models become integral to critical.

  7. PoSh: Using Scene Graphs To Guide LLMs-as-a-Judge For Detailed Image Descriptions

    Include an LLM-as-judge paper to test judge design and agreement assumptions. Signals: human evaluation + rubric ratings. Focus: CAPArena / spearman. Abstract: In this work, we introduce PoSh,.

  8. Validating Political Position Predictions of Arguments

    Adds human evaluation with pairwise preferences for broader protocol coverage within this hub. Signals: human evaluation + pairwise preferences. Focus: agreement. Abstract: Real-world knowledge representation often requires capturing.

Known Limitations

Known Limitations

  • Only 5.4% of papers report quality controls; prioritize calibration/adjudication evidence.
  • Rater population is under-specified (12.4% coverage).
  • Narrative synthesis is grounded in metadata and abstracts only; full-paper implementation details are not parsed.
Research Utility Snapshot (Detailed)

Research Utility Snapshot

Human Feedback Mix

  • Pairwise Preference (105)
  • Demonstrations (32)
  • Red Team (30)
  • Rubric Rating (21)

Evaluation Modes

  • Automatic Metrics (150)
  • Simulation Env (58)
  • Llm As Judge (26)
  • Human Eval (16)

Top Benchmarks

  • HotpotQA (5)
  • DROP (3)
  • LMSYS Chatbot Arena (3)
  • MT Bench (3)

Top Metrics

  • Accuracy (73)
  • Cost (33)
  • Latency (13)
  • Agreement (12)

Rater Population Mix

  • Domain Experts (38)
  • Mixed (1)

Quality Controls

  • Calibration (9)
  • Inter Annotator Agreement Reported (7)
  • Adjudication (4)
  • Gold Questions (1)
Coverage diagnostics (sample-based): human-feedback 95.0% · benchmarks 38.3% · metrics 68.3% · quality controls 15.0%.

Top Papers

  • AgentHER: Hindsight Experience Replay for LLM Agent Trajectory Relabeling

    Liang Ding · Mar 22, 2026 · Citations: 0

    Demonstrations Human EvalLlm As Judge Long Horizon

    LLM agents fail on the majority of real-world tasks -- GPT-4o succeeds on fewer than 15% of WebArena navigation tasks and below 55% pass@1 on ToolBench (Zhou et al., 2024; Qin et al., 2024) -- yet every failed trajectory is routinely…

  • LMUnit: Fine-grained Evaluation with Natural Language Unit Tests

    Jon Saad-Falcon, Rajan Vivek, William Berrios, Nandita Shankar Naik, Matija Franklin · Dec 17, 2024 · Citations: 0

    Pairwise Preference Human Eval

    We introduce natural language unit tests, a paradigm that decomposes response quality into explicit, testable criteria, along with a unified scoring model, LMUnit, which combines multi-objective training across preferences, direct ratings,…

  • PoSh: Using Scene Graphs To Guide LLMs-as-a-Judge For Detailed Image Descriptions

    Amith Ananthram, Elias Stengel-Eskin, Lorena A. Bradford, Julia Demarest, Adam Purvis · Oct 21, 2025 · Citations: 0

    Rubric Rating Human EvalLlm As Judge

    In this work, we introduce PoSh, a metric for detailed image description that uses scene graphs as structured rubrics to guide LLMs-as-a-Judge, producing aggregate scores grounded in fine-grained errors (e.g.

  • SCOPE: Selective Conformal Optimized Pairwise LLM Judging

    Sher Badshah, Ali Emami, Hassan Sajjad · Feb 13, 2026 · Citations: 0

    Pairwise Preference Automatic Metrics

    Large language models (LLMs) are increasingly used as judges to replace costly human preference labels in pairwise evaluation.

  • TraceSafe: A Systematic Assessment of LLM Guardrails on Multi-Step Tool-Calling Trajectories

    Yen-Shan Chen, Sian-Yao Huang, Cheng-Lin Yang, Yun-Nung Chen · Apr 8, 2026 · Citations: 0

    Red Team Automatic Metrics Long Horizon

    As large language models (LLMs) evolve from static chatbots into autonomous agents, the primary vulnerability surface shifts from final outputs to intermediate execution traces.

  • SODIUM: From Open Web Data to Queryable Databases

    Chuxuan Hu, Philip Li, Maxwell Yang, Daniel Kang · Mar 19, 2026 · Citations: 0

    Expert Verification Automatic Metrics Multi Agent

    Existing systems struggle with SODIUM tasks: we evaluate 6 advanced AI agents on SODIUM-Bench, with the strongest baseline achieving only 46.5% accuracy.

  • Elo-Evolve: A Co-evolutionary Framework for Language Model Alignment

    Jing Zhao, Ting Zhen, Junwei Bao, Hongfei Jiang, Yang Song · Feb 14, 2026 · Citations: 0

    Pairwise Preference Automatic Metrics Multi Agent

    Current alignment methods for Large Language Models (LLMs) rely on compressing vast amounts of human preference data into static, absolute reward functions, leading to data scarcity, noise sensitivity, and training instability.

  • Measuring AI Ability to Complete Long Software Tasks

    Thomas Kwa, Ben West, Joel Becker, Amy Deng, Katharyn Garcia · Mar 18, 2025 · Citations: 0

    Expert Verification Automatic Metrics Tool Use

    Despite rapid progress on AI benchmarks, the real-world meaning of benchmark performance remains unclear.

  • When AI Meets Early Childhood Education: Large Language Models as Assessment Teammates in Chinese Preschools

    Xingming Li, Runke Huang, Yanan Bao, Yuye Jin, Yuru Jiao · Mar 25, 2026 · Citations: 0

    Rubric Rating Automatic Metrics

    In this paper, we investigate whether AI can serve as a scalable assessment teammate by extracting structured quality indicators and validating their alignment with human expert judgments.

  • Stabilizing Rubric Integration Training via Decoupled Advantage Normalization

    Zelin Tan, Zhouliang Yu, Bohan Lin, Zijie Geng, Hejia Geng · Mar 27, 2026 · Citations: 0

    Rubric Rating Automatic Metrics

    We propose Process-Aware Policy Optimization (PAPO), a method that integrates process-level evaluation into Group Relative Policy Optimization (GRPO) through decoupled advantage normalization, to address two limitations of existing reward…

  • DSPA: Dynamic SAE Steering for Data-Efficient Preference Alignment

    James Wedgwood, Aashiq Muhamed, Mona T. Diab, Virginia Smith · Mar 23, 2026 · Citations: 0

    Pairwise Preference Automatic Metrics

    Preference alignment is usually achieved by weight-updating training on preference data, which adds substantial alignment-stage compute and provides limited mechanistic visibility.

  • BEACON: Language-Conditioned Navigation Affordance Prediction under Occlusion

    Xinyu Gao, Gang Chen, Javier Alonso-Mora · Mar 10, 2026 · Citations: 0

    Automatic MetricsSimulation Env Web Browsing

    As a result, they struggle to infer target locations in occluded regions, typically caused by furniture or moving humans.

  • Critique-GRPO: Advancing LLM Reasoning with Natural Language and Numerical Feedback

    Xiaoying Zhang, Yipeng Zhang, Hao Sun, Kaituo Feng, Chaochao Lu · Jun 3, 2025 · Citations: 0

    Critique Edit Automatic Metrics

    We show that plateaued RL models can successfully refine failed solutions when given natural language critiques.

  • VerifyBench: Benchmarking Reference-based Reward Systems for Large Language Models

    Yuchen Yan, Jin Jiang, Zhenbang Ren, Yijun Li, Xudong Cai · May 21, 2025 · Citations: 0

    Pairwise Preference Automatic Metrics

    In this paper, we introduce VerifyBench and its challenging variant VerifyBench-Hard, two benchmarks specifically designed to assess reference-based reward systems.

Related Hubs

Need human evaluators for your AI research? Scale annotation with expert AI Trainers.