Journal — Open source
Building with AI, keeping the record
A long-running project needs an archive that survives a change of tool, model, or workspace. Here is how this journal was assembled and what it does not contain.
Digital Sidewalk development has involved three AI assistants over two years, each with a different slice of the history. That makes the archive broader than any one conversation, and it makes it easy for an idea, a generated file, and a tested result to blur when the work is later summarized.
How this edition was made
A first draft of this journal was assembled by one assistant from the documents it could inspect: the conference manuscript, the August bundle, the readiness plan, the global protocol, and the public site. It correctly labeled proposals as proposals and declined to promote an unaudited dataset claim into a finding.
It also had a partial history. It did not know about the April 2025 naming decision, the January 2026 cavity model and its HVAC lineage, the ice-storm dataset's provenance, the mathematics, the August parent frame, the observation-design instrument, the calibration protocol, the open-source package, or its own August specification. A second pass, made with the assistant that had that history, added those entries and corrected two claims: the field-test console is live, not unreachable, and the dataset exists with stated limits rather than being an unverified number.
Separate proposal, artifact, and result
A proposal describes what should be done. An artifact shows what was produced. A result requires a method, inputs, and an output that supports a particular claim. Agreement between assistants does not replace those requirements, and every entry here carries a status label for that reason.
Article dates describe the period of the underlying work. They are not invented original publication dates. Every entry in this first edition was published on the same day, September 15, 2026, and says so.
What is not here
Private source documents, proprietary detector internals, and full assistant transcripts are not included. The held entry on the Lab's relationship to a sister unit's forthcoming product is written and will be published when that product is public. Missing conversations remain missing in the record until their contents can be recovered.
Making the next update easier
Each future addition records whether it contains private material, third-party content, or a result that still needs reproduction. The aim is continuity: a reader should be able to see what changed, what survived inspection, and what remains to be established.