Guides

Each guide covers one job end to end. If you would rather start from a complete, runnable example, the use cases are recipes with measured results.

Seed a database#

• SQL DDL to schema: generate from CREATE TABLE statements, honouring keys, CHECK rules, UNIQUE and NOT NULL.
• Seed a live database: misata seed reads a Postgres or SQLite schema and fills it, parents first.
• Database seeding in Python: seed_database(schema, url) from code and from pytest.
• From a database: introspect, edit, regenerate.
• Django and dbt seed integration.

Test pipelines and code#

• pytest fixtures: data inside a test suite, with fingerprints.
• Validate and preview: profile a CSV, inspect what a story will produce.
• Realism report: the statistical tells of generated data.
• When Misata refuses: every refusal and warning, and how to make warnings fatal in CI.

Prove what the data holds#

• Attest a dataset: an in-toto attestation and a verify.py anyone can rerun with DuckDB.
• Certificate predicate v1: the format, field by field.

Share it#

• Publish to Hugging Face: verified before upload, with a dataset card generated from the attestation.

Hit a number a pipeline computes#

• Solve for a pipeline's output: move outcome curves until a dbt model, a SQL query or a Python function produces the targets.

Evaluate SQL models and agents#

• Build an evalpack: questions whose answers were declared before the data, checked on rotated databases.
• MCP server: give an agent tools to explain, generate, audit, attest, solve and seed.

Generate at scale#

• Streaming large datasets: bounded memory, straight to Parquet, JSON Lines, CSV or DuckDB.
• Export and Databricks and Spark.

Imitate a real file#

• Mimic: a twin fitted to a CSV. It is fitted to real rows and is not anonymous; use declared generation when the data must not derive from real records.

Shape the data#

• Outcome curves and rate curves: totals and rates per period that hold exactly.
• Distributions, correlations and conditional columns.
• Narrative patterns: growth stories in plain English.
• Processes and event logs, nested JSON and arrays, long-form text and geospatial realism.
• Lexicons, motifs and joint margins, anomaly injection, custom generators and use-case presets.

Comparisons#

• Faker vs SDV vs Misata
• Multi-table synthetic data
• Synthetic data for BI demos
• Python synthetic data generator