There's growing evidence that LLMs can p-hack.
But p-hacking also points to something bigger: a data science multiverse of defensible analytical choices.
We wrote a paper (arxiv.org/abs/2602.18710) on using LLM agents to map this multiverse systematically. 🧵
