How the concepts fit together
A project holds one version graph. A branch is a line of events inside it;main is protected. A commit names a checkpointed range of those events and carries a content root. The taxonomy (what an annotation may say) and the pipeline (how work moves) live inside commit content, so changing either is a commit and a disagreement about either is a merge conflict. Each Task keeps the taxonomy and pipeline versions it was created under. A Task is one unit of work over a Source at one stage; submitting its labels or finishing its review writes a commit. A review verdict binds the hash of the content it judged. A Dataset is a named selection over a branch, and a Dataset Version is that selection at one commit, frozen into a manifest with a digest per chunk.
Two ideas cut across all of that. The data boundary says what the control plane holds (events, references, hashes) and what stays in your storage (voxels). Provenance says who or what wrote each event, under which authorization Decision, and which earlier state it changed.
Some of these records have a dashboard screen (commits on the Versions tab, stages on the Pipeline tab) and some are read through the API only (Decisions and Ledger entries). The reference pages say which.
Next step
Read Data boundary to trace where data goes, or Version model to understand how history is kept, then pick the guide closest to the work you already do.| Plans | All plans |
|---|---|
| Last verified | 2026-09-28 |