Trace Sufficiency: Claim-Relative Evidence for Verifying Agentic LLM Executions — Data, Code, and Reproducibility Artifacts

Frozen data, code, protocol records, semantic and statistical closure artifacts, clean-root reproducibility evidence, and public execution-pipeline evidence supporting the article “Trace Sufficiency: Claim-Relative Evidence for Verifying Agentic LLM Executions.” The archive contains the 396,000-row controlled benchmark, controlled benchmark code and configuration, independent semantic-constancy closure, protocol-fidelity scenario-pair analyses, oracle/property/mutation/split evidence, bundle-dependency and ablation artifacts, and the completed 984-case AgentDojo public execution layer. AgentDojo evidence is provided as execution-pipeline transfer and is not presented as external semantic Trace Sufficiency validation.