Abstract page for arXiv paper 2610.23866: GardenBench: A lightweight, daily evaluation of LLM capabilities