Resources
Technical specifications
Formal specifications for the LiteSeed platform — Blueprint schema, performance targets, reproducibility guarantees and the generation engine.
Platform specifications
Performance targets
Generation throughput≥ 50,000 rows/sec
Peak RAM (5M+ rows)≤ 256 MB
Chunk size10,000 rows
Single run capacity5M+ rows
Hard constraint resampling50 retries max
Sandbox mode (dev/test data)up to 10k rows
Training mode (ML/AI data)up to 1M rows
Reproducibility guarantees
PRNG algorithmMulberry32
Seed type32-bit integer
RNG instances per run1 (shared)
Blueprint hash algorithmSHA-256
Generation identity tuple(blueprint_id, seed, rows, mode)
Reproduce endpointPOST /api/trpc/runs.reproduce
Cross-machine determinismGuaranteed
Schema specifications
Database schema
projects
blueprints
blueprint_versions
seed_uploads
runs
dataset_versions
run_exports
experiments
experiment_metrics
dataset_insights
dataset_recommendations
users
sessions
project_members
audit_log
Export formats
CSVUTF-8, comma-separated, header row
JSONLOne JSON object per line
ParquetApache Parquet via @dsnp/parquetjs
SQLiteSingle-table SQLite database
ZIPAll formats bundled
SQL dumpINSERT statements for any SQL database
