VAST-Bench v0.6: Corpus, Model Outputs, and Reproducibility Artifact

VAST-Bench v0.6 is the reproducibility artifact for the controlled exploratory study “VAST-Bench: A Controlled Exploratory Study of Evidence-Conditioned Vulnerability-Claim Labeling.” The archive contains twelve controlled vulnerability scenarios and executable local oracles; the counterfactual evidence-state constructor; frozen prompts, policies, scoring, and aggregation code; the exact model manifest; corrected raw model outputs and parsed results; and the codebook, data card, analysis plans, and SHA-256 manifest. The original pilot evaluates free-text primitive-to-impact overclaiming under direct and structured claim-labeling conditions. The exploratory counterfactual arm varies local-only, reachable-only, full-chain, and negative-control evidence states with and without state-specific evidence. The artifact is intended to reproduce the reported results; it does not represent operational bug-bounty performance or general vulnerability grounding. Artifact archive SHA-256: 6dd30acd0f780b3a752cd16577cd615126bce8b648a7ec21d6ad9cd0cfe62626