Forge Lake
Raw reality: ingest, dedupe and version the data that feeds every label.
Lake
Label takes uploads one dataset at a time. Lake is the hose behind that: cameras and object stores in, a versioned corpus out.
- Bulk ingest from cameras, S3 / Azure Blob and R2, with content-hash dedupe.
- Dataset versioning so a Train run pins the exact slice it saw.
- Lineage from a raw frame through Label into a Registry artifact.
stub ยท ships after Label