Research map / Large language models / Agents, code and math
WebDevJudge: Evaluating (M)LLMs as Critiques for Web Development Quality
Chunyang Li, Yilun Zheng, Xinting Huang, Tianqing Fang, Jiahao Xu, Lihui Chen, Yangqiu Song, Han Hu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use pairwise comparison instead of single-answer grading (ablation) | LLM judges · WebDevJudge | agreement rate: avg gain over 8.0% | Improved |
| use agentic workflow instead of vanilla LLM judge | GPT-4.1 / UI-TARS-1.5 · WebDevJudge | agreement rate: 60.55 vs 60.86 (single) | Hurt |
| add rubric or Likert guidance over Direct judgment (ablation) | pairwise LLM judges · WebDevJudge | agreement rate | No clear effect |
| use binary rubric instead of Likert scale (ablation) | single-answer LLM judges · WebDevJudge | agreement rate | Improved |
| remove code input (image only) (ablation) | GPT-4.1 pairwise · WebDevJudge | agreement rate: 70.34 -> 60.55 | Hurt |
| remove screenshot (code only) (ablation) | GPT-4.1 pairwise · WebDevJudge | agreement rate: 70.34 -> 69.57 | Hurt |
| use screenshots only for single-answer grading of Gemini-2.5-pro (ablation) | Gemini-2.5-pro · WebDevJudge | agreement rate: 50.31 -> 53.36 | Improved |
| add explicit instruction to ignore position bias (ablation) | LLM pairwise judges · WebDevJudge | consistency: consistency 83.3-89.6% | No clear effect |
| add rubric-tree guidance to annotation (ablation) | human annotators · annotation agreement (w/ tie) | inter-annotator agreement (%): 65.0 -> 92.0 | Improved |
| gate ensemble: agent OR LLM for dynamic rubric items (ablation) | GPT-4.1 + DeepSeek-V3 + UI-TARS-1.5 · WebDevJudge | agreement rate | Improved |
Closest papers by meaning
- WebGen-Bench: Evaluating LLMs on Generating Interactive and Functional Websites from Scratch (NeurIPS 2025 · similarity 0.90)
- REAL: Benchmarking Autonomous Agents on Deterministic Simulations of Real Websites (NeurIPS 2025 · similarity 0.86)
- How can we assess human-agent interactions? Case studies in software agent design (ICML 2026 · similarity 0.85)
- JudgeBench: A Benchmark for Evaluating LLM-Based Judges (ICLR 2025 · similarity 0.85)
- TimeWarp: Evaluating Web Agents by Revisiting the Past (NeurIPS 2026 · similarity 0.85)
- WebDS: An End-to-End Benchmark for Web-based Data Science (ICLR 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
