Report-contingent scoring prompts increase upward misreporting in large language models

It excludes Python, Stata, and R programs; processed datasets; figures; derived analysis outputs; historical pilot batches; and unrelated mock or exploratory data. No human-subject data are included.

The package is intended to document the data underlying the manuscript and support provenance checks. Because several collections use provider APIs and separate collection batches, exact replication may depend on provider availability, endpoint state, and recorded collection metadata. Analysis programs are available from the authors upon request.