Documentation

Create a project, upload a seed file, review the extracted Blueprint and generate your first dataset in under 5 minutes.

Complete reference for the Blueprint JSON schema — field types, distribution parameters, constraint types and generation policies.

When to use Sandbox mode (test data for software development, up to 10k rows) and when to use Training mode (ML/AI training data at scale, up to 1M rows).

How Blueprints are structured, versioned and linked to dataset versions. Includes Blueprint hash verification and parent-child lineage.

Reference for all 8 supported distributions: normal, lognormal, gamma, uniform, poisson, categorical, rare_event, mixture.

Hard and soft constraints, the 6 constraint types (formula, range, regex, date_order, not_null, enum_only) and resampling behaviour.

How the Mulberry32 PRNG, randomSeed and blueprintHash guarantee bit-for-bit reproducibility across runs.

How the chunk-based streaming engine (CHUNK_SIZE=10,000) generates 5M+ row datasets with ≤256 MB peak RAM.

Dataset versioning, the lineage graph (Seed → Blueprint → Versions), experiment tracking and the Dataset OS primitives.