Each team may contribute up to 15 Q&A pairs by Oct 27. These will be added to the public test set on Nov. 7 and the leaderboard will refresh. Your team will know its own answers, but that’s fine — they are only a small part of the 250+ test questions. The goal is to add a layer of game-theory while helping WattBot grow into a stronger, community-driven benchmark. Please review submissions carefully and use good ole fashioned organic brains instead of ChatGPT; *any inaccuracies discovered during review will cause the entire set to be discarded and no further questions accepted from that team*.

Questions and answers should:
- follow the same conventions as train_QA.csv
- stick to sustainability topics surrounding AI  
- be grounded in explicit evidence (quote, table, or figure) with ref ID(s)  
- be phrased to yield short, checkable answers (numbers, ranges, or categorical names)  
- to generate good Q&A pairs (that challenge the best RAG systems), we recommend crafting Q&A pairs that have one or more of the following qualities:
    - combine evidence across multiple papers
    - involve visual reasoning over figures (bonus: without supporting text to back it up so it's visual reasoning only)
    - involve simple math/unit conversions/deductions
    
**Send your team's Q&A pairs to endemann@wisc.edu by Oct 27**.