Evaluator lineages in world-model benchmarks

Versioned replication package for “Evaluator lineages expose pseudo-replication in world-model benchmarks”. The release contains the 112-row evaluator-lineage registry, a machine-verifiable evidence ledger, frozen derived analysis outputs, configurations, tests, figure and table source data, environment specifications and hash-verification utilities. It supports reproduction of the reported evaluator-lineage, reaggregation, sensitivity and leaderboard analyses. Third-party benchmark implementations, source videos and model weights are not redistributed; their versions, licences and public locators are documented in the package.