Technology
Reproducibility system
The technical foundation for deterministic dataset generation \u2014 same Blueprint, same seed, same dataset. Always.
Start FreeHow it works
Determinism through seeded PRNG and Blueprint hashing.
LiteSeed achieves reproducibility through two mechanisms: a deterministic PRNG seeded with a user-defined integer, and a Blueprint hash that captures the complete schema specification at generation time.
Mulberry32 PRNG
A deterministic pseudo-random number generator seeded with a user-defined integer. Same seed = same random sequence.
Blueprint hashing
A SHA-256 hash of the complete Blueprint specification is computed and stored at generation time.
Version locking
Every generation run locks the Blueprint version and seed — preventing accidental schema changes from affecting reproducibility.
Implementation details
Mulberry32 PRNG
The Mulberry32 algorithm provides a high-quality, deterministic PRNG in pure JavaScript.
- \u219232-bit state, seeded with user-defined integer
- \u2192~500M samples/second in V8
- \u2192Passes statistical randomness tests (BigCrush)
- \u2192Same seed = identical sequence across all environments
Blueprint hash
A SHA-256 hash of the serialised Blueprint specification is computed at generation time.
- \u2192Hash computed from canonical JSON serialisation
- \u2192Stored in dataset version as blueprintHash
- \u2192Hash mismatch detection on reproduction attempts
- \u2192Hash included in compliance export metadata
Seed storage and retrieval
The generation seed is stored in the dataset version metadata and can be retrieved for reproduction at any time.
- \u2192Seed stored as integer in dataset version
- \u2192One-click reproduction via Reproduce Run button
- \u2192Seed + Blueprint hash = complete reproduction specification