heartland-synthetic: A Reproducible Generator and Benchmark Dataset for Rural Heart-Failure Research Workflows

Background: Research on rural heart-failure implementation often requires variables absent from general-purpose synthetic-patient generators, including distance to cardiology and structured social-support measures. Real clinical data may be unavailable during early protocol and interoperability development. Objective: To describe heartland-synthetic v0.2.2 and validate the integrity and reproducibility of its public 1,000-row benchmark cohort. Methods: The Python package generates tabular cohorts, HEARTLAND risk scores, longitudinal monitoring series, and REDCap/FHIR R4 exports from literature-informed distributions. Tests evaluated reproducibility, scoring, distributions, rural effects, outcomes, and exports. The public seed-42 benchmark was independently checked for shape, completeness, uniqueness, checksum, and descriptive distributions. Results: All 65 automated tests passed. The benchmark contained 1,000 unique rows and 31 columns, with no missing cells or duplicate patient identifiers. Conclusions: The package supports reproducible software testing, methods development, and training without real patient data. Its modeled distributions and outcomes are not external validation, population estimates, causal evidence, or clinical predictions.