Human and LLM Strategy Submissions for Colonel Blotto Tournament Experiments

215 anonymized human strategy submissions form the core of a study comparing human and large language model performance in a Colonel Blotto game variant. The dataset includes raw and final valid strategy sets across three tournament environments, with allocations to nine states and submission timestamps. Author Dmitry Dagaev published this replication data via Harvard Dataverse in May 2026.

Use Cases

• Compare human and LLM strategic behavior based on the described tournament environments.
• Analyze the distribution of resource allocations across nine states based on the final sheets.
• Study the validity of strategy submissions based on the separation of raw and final data.
• Examine temporal patterns in strategy submission based on the included timestamps.

Strengths

• Contains 215 total human-submitted strategies across three tournament environments.
• Provides separate raw and final valid strategy sets, with final sets containing 205 to 207 valid human strategies.
• Includes a controlled comparison with 22 to 207 LLM strategies in Tournaments 2 and 3.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for individual sheets is unknown, which may limit suitability assessment.

Provenance

Source

Harvard Dataverse

Collection Method

Collected from human and LLM submissions for a controlled tournament experiment.

Freshness

Last updated 2026-05-11 14:21:39; freshness should be verified.