CICI:IPAAI:CERTIFY:Contractual Integrity for Scientific AI-Ready Data Pipelines

Scientific artificial intelligence systems increasingly depend on datasets collected, cleaned, and transformed across multiple institutions. The integrity, provenance, and authenticity of these datasets determine whether scientific results are valid and reproducible. Current approaches such as logs, checksums, and anomaly detection provide useful evidence, but they do not guarantee that data transformations preserve the scientific properties required by downstream analyses. The CERTIFY project creates a framework for verifiable integrity in artificial-intelligence-ready scientific data pipelines. The project enables scientific data objects to carry machine-checkable proofs that specified integrity and provenance requirements are satisfied. It also uses temporal contracts to describe required properties of data curation, cleaning, and transformation steps, and compiles those contracts into lightweight operating-system-level monitors that prevent buggy or compromised code from violating declared policies. Large language model-based agents help configure and manage these components, reducing the expertise required to apply formal verification methods in scientific workflows. The project is evaluated through two science drivers: agricultural sensor networks used to study livestock health and crop yield, and electronic health records that require strong provenance and privacy safeguards. The project produces open-source software, documented example datasets, course modules, and training materials that help prepare researchers and students to secure AI-driven scientific workflows. This structure supports reproducible discovery while making integrity checks practical for collaborative, cross-institutional scientific teams. The results strengthen trust in data-intensive science by helping researchers verify where data came from, how it changed, and whether it remains suitable for scientific use. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.