datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
2026.RA.Fairness-Counterfactual-Pairs
2026.RA.Fairness-Counterfactual-Pairs
Action-level contrastive pairs from five-party private-information negotiations: at every turn an LLM took,
what a computable ideal agent would have done at that same decision point.
Each row is one (episode, turn, counterfactual_type). rejected_action is what the LLM actually did;
chosen_action is the counterfactual agent's action. Both are structured actions
({atype, offer_id, deal}), never prose.
72,192 rows, 58,168 of them (81%)… See the full description on the dataset page: https://huggingface.co/datasets/siddharthmb/2026.RA.Fairness-Counterfactual-Pairs.fairmerging-gpu-outputsFairECGFairECGReports