Guidelines
Applied Risk & Resilience Labs are expected to apply geospatial tools meaningfully, address a clear real-world need, and follow ethical and respectful practices with data. Because our work touches real communities and infrastructure, we also maintain a data governance policy that defines what we publish, what we withhold, and how we make those decisions.
Data Governance
Every dataset, model, and output is classified into one of four tiers before work begins. The tier determines where it is stored and who can access it.
Tier 1 · Open
Open-source GitHub. Methods, code, documentation, tutorials and results that do not point to a specific settlement, building or person. This is the default for anything another lab could reuse.
Tier 2 · Attributed
Open-source GitHub, with appropriate license and citation. Datasets we build from public sources such as Sentinel, Landsat, OpenStreetMap and HDX. We keep the upstream license and credit the source prominently, including HOTOSM, OSM and HDX contributors.
Tier 3 · Gated
Available on request. Results detailed enough to point at a specific place, such as building-level risk scores or corridor rankings. Hosted on Zenodo with access by request. Researchers, emergency managers and infrastructure owners are eligible.
Tier 4 · Restricted
Partner-only, never public. Results about vulnerable populations or critical infrastructure where release could put people or systems at risk are shared only with the sponsoring partner or agency under their governance.
Reliability and Use
Not for operational use
All outputs are learning artifacts produced by students. Every model and dataset must include uncertainty quantification and a clear statement that outputs are not intended for operational use without independent expert validation.
Repository Defaults
New repositories are private by default. Secret scanning is enabled on every repository from creation. Tier 3 and Tier 4 data are never committed to version history and large open datasets are kept with data-versioning tools rather than embedded directly.