Replication Data for: Not Yet: Humans Outperform LLMs in a Colonel Blotto Tournament

This dataset contains anonymized strategy submissions used in the study “Not Yet: Humans Outperform LLMs in a Colonel Blotto Tournament.” The study examines strategic behavior of humans and large language models (LLMs) in the Electoral Race game, a discrete variant of the Colonel Blotto game. The dataset includes submissions for three tournament environments. Tournament 1 contains human strategies competing only against other human strategies. Tournament 2 adds strategies submitted by several popular LLMs to the human strategy pool. Tournament 3 matches the number of LLM strategies to the number of valid human strategies. The workbook includes both raw human submissions, including invalid entries, and the final valid strategy sets used in the tournament analysis. The file contains the following sheets: “Tournament 1 humans,” “Tournament 2 humans,” and “Tournament 3 humans” include all 215 human-submitted strategies for the corresponding tournaments, including invalid submissions. “Tournament 1 final” includes 205 valid human strategies. “Tournament 2 final” includes 207 valid human strategies and 22 LLM strategies. “Tournament 3 final” includes 207 valid human strategies and 207 LLM strategies. The final sheets report allocations to States A–I, the total number of trips, and submission timestamps.本数据集收录了研究《Not Yet: Humans Outperform LLMs in a Colonel Blotto Tournament》中使用的匿名化策略提交成果。该研究针对人类与大语言模型（Large Language Model，LLM）在选举竞赛博弈（Electoral Race Game）——一种离散形式的上校布莱托博弈——中的策略行为展开了系统性考察。 本数据集涵盖三类锦标赛环境下的策略提交内容：锦标赛1仅允许人类提交的策略之间相互对战；锦标赛2在人类策略池中加入了多款主流大语言模型生成的策略；锦标赛3则使大语言模型策略的数量与有效人类策略的数量完全匹配。 本数据集附带的Excel工作簿同时收录了原始人类提交内容（含无效条目）以及用于锦标赛分析的最终有效策略集。 该文件包含以下工作表："Tournament 1 humans"、"Tournament 2 humans"及"Tournament 3 humans"分别收录对应锦标赛的全部215条人类提交策略，包含无效提交内容；"Tournament 1 final"包含205条有效人类策略；"Tournament 2 final"包含207条有效人类策略与22条大语言模型策略；"Tournament 3 final"包含207条有效人类策略与207条大语言模型策略。 上述最终工作表还记录了针对州A至I的资源分配方案、总分配次数以及提交时间戳。