Five wildfire tasks for language models and agents, each run bare and grounded and scored beside a non-LLM comparator. 77,644 stored responses; every paper table reproduces offline. - View it on GitHub