Inside the August architecture bundle

On August 6, 2026 the Lab froze a small archive of seven files: a README, a source registry, three schema and taxonomy files, an architecture note, and a validation plan. Together they define how an observation is recorded. They do not include a running ingestion service, a completed field experiment, or benchmark results. Calling this an architecture bundle preserves what was built without inflating its status.

Begin with a citable observation

The central record type, called a StormAtom in the archive, represents one observation or one derived value. Its required fields are an identifier, a schema version, an evidence class, a variable, a valid time, a spatial support (the footprint the value applies to), the value, its unit, its source, its provenance, a quality flag, and a rights statement.

The evidence classes separate observations, derived values, model analyses, forecasts, human reports, inferences, hypotheses, and unresolved states. That distinction matters because a model output must not silently become a direct measurement when it enters a map or a summary.

Keep the layers connected

The archive sketches bundles of observations, event records, longer-term summaries, and explicit missingness records. Its architecture note requires source lineage, versioned corrections, retained source-quality flags, and separate treatment of derived values.

The source registry lists candidate environmental sources. Listing a source is not confirmation that access was configured, that data were ingested, or that redistribution was authorized. Each of those is its own field.

Next

The next step for the schema is to exercise it on a small permitted dataset, which is what the ice-storm file provides, and then test whether the resulting package is understandable and reproducible outside the original workspace.