Research collaboration

Share an analysis as a small, immutable evidence package: raw-data identifier or access instructions, data dictionary and units, preprocessing script, discovery configuration, generated world bundle, validation data, simulation command, and a short statement of assumptions and exclusions. Reviewers should be able to reproduce the selected candidate without reconstructing intent from a notebook.

Separate exploratory candidates from promoted results. A promoted world should identify its author, source data, validation evidence, and compatibility version. Do not overwrite a bundle after review; create a new revision and keep the prior checksum so comparisons remain meaningful.

The local server core provides organization-scoped metadata, artifacts, and append-only events, but it is not a collaborative workflow system or a substitute for access review and authorship policy.