GitHub for Science
Git gave code a shared lifecycle: version, diff, branch, merge, review, and cite. Commonscience's 5-year architectural direction is to give scientific artifacts — molecules, datasets, trained models, protocols, assay results, notebook entries — that same lifecycle.
Version — every entity is designed to carry a hash-chained history. Diff — comparisons are meant to be structural, not textual: a molecule diff is not a text diff of a file. Branch — alternative hypotheses or parallel parameterizations designed as first-class divergence, not copies. Merge — reconciling divergent branches with provenance meant to stay intact, not overwritten. Review — protocols designed to be reviewed against methodology, models against training data and reproducibility, not just against style. Cite — the intent is a stable, versioned, addressable reference for any artifact.
This lifecycle only works if every artifact is modeled consistently underneath — which is what the entity model and its provenance event chain are designed for. Version/diff/branch/merge/review/cite is the visible behavior; the entity system and provenance chain are the substrate that's meant to make it possible.
This is a 5-year North Star, not a claim that full git-style branching and merging across every artifact type is available now. GUIDE is the first client being designed toward it — its notebook is the first surface being built around this framing; the rest of the lifecycle (structural diff, branch/merge across all entity types, formal review workflows) is architecture the platform is designed to grow into, not a finished capability list.