YESciEval: Robust LLM-as-a-Judge for Scientific Question Answering

Large Language Models (LLMs) drive scientific question-answering on modern search engines, yet their evaluation robustness remains underexplored. We introduce YESciEval, an open-source framework that combines finegrained rubric-based assessment with reinforcement learning to mitigate optimism bias in LLM evaluators. We release multidisciplinary scienceQ&A datasets, including adversarial variants, with evaluation scores from multiple LLMs. Independent of proprietary models and human feedback, our approach enables scalable, cost-free evaluation. By advancing reliable LLM-as-a-judge models, this work supports AI alignment and fosters robust, transparent evaluation essential for scientific inquiry. Conciseness: Subtle Adversarial Domain: Physics Research Question: What are the recent advancements in X-ray laser technology, and how are they being applied across various fields? Paper titles: 1. Sub-38 nm resolution tabletop microscopy with 13 nm wavelength laser light 2. Coherent imaging of biological samples with femtosecond pulses at the free-electron laser FLASH 3. Femtosecond X-ray measurement of coherent lattice vibrations near the Lindemann stability limit 4. Picosecond Snapshot of the Speckles from Ferroelectric BaTiO3 by Means of X-Ray Lasers 5. Defect-tolerant extreme ultraviolet nanoscale printing ORKG Science Q&A Instance 23 Llama-3.1-8B Response Correctness: Extreme Adversarial