Pytest Fixtures with Misata

misata.testing provides pytest fixture factories and built-in fixtures so your tests get realistic synthetic data without database setup, migrations, or fixture files.

Install

pip install misata pytest

Or install the dedicated plugin and skip the conftest wiring entirely:

pip install pytest-misata

The plugin registers the built-in fixtures with pytest automatically, so saas_tables, ecommerce_tables and friends work in any test file with zero configuration. It is listed in pytest's plugin index as pytest-misata.

Quick start

Define fixtures in conftest.py using misata_fixture:

# conftest.py
from misata.testing import misata_fixture

saas_tables   = misata_fixture("A SaaS company with 500 users", rows=500)
fintech_tables = misata_fixture("A fintech with 200 customers and 2% fraud rate", rows=200)
hr_tables     = misata_fixture("An HR company with 300 employees", rows=300)

Use them in any test file, no imports needed:

# test_billing.py
def test_invoice_fk_integrity(saas_tables):
    invoices = saas_tables["invoices"]
    subs     = saas_tables["subscriptions"]
    assert invoices["subscription_id"].isin(subs["subscription_id"]).all()

def test_churn_rate(saas_tables):
    users = saas_tables["users"]
    assert 0.10 <= users["is_active"].mean() <= 0.90

# test_fraud.py
def test_fraud_flag_rate(fintech_tables):
    txns = fintech_tables["transactions"]
    # ~2% fraud rate from the story description
    assert 0.01 <= txns["is_fraud"].mean() <= 0.05

misata_fixture(story, rows, seed, ...)

Parameter	Default	Description
story	required	Plain-English dataset description
rows	1000	Row count for the primary table
seed	42	Random seed: same seed = identical data every run
smart_correlations	False	Auto-add Pearson correlations between related numeric columns
min_quality_score	None	Retry generation until FidelityChecker score meets threshold

misata_schema_fixture(story, rows)

Returns a SchemaConfig (no data generated), useful for testing schema parsing and domain detection:

# conftest.py
from misata.testing import misata_schema_fixture

saas_schema = misata_schema_fixture("A SaaS company with users and subscriptions")

# test_schema.py
def test_domain_detection(saas_schema):
    assert saas_schema.domain == "saas"

def test_table_names(saas_schema):
    names = [t.name for t in saas_schema.tables]
    assert "users" in names
    assert "subscriptions" in names

Built-in fixtures

Three fixtures are available to import directly without conftest.py setup:

misata_generate

Injects misata.generate, use when you want to generate different datasets within a single test:

from misata.testing import misata_generate   # imported for type hints only

def test_multi_domain(misata_generate):
    saas    = misata_generate("A SaaS company", rows=100, seed=1)
    fintech = misata_generate("A fintech company", rows=100, seed=2)
    assert "users" in saas
    assert "customers" in fintech

misata_parse

Injects misata.parse for schema inspection tests:

def test_saas_domain(misata_parse):
    schema = misata_parse("A SaaS company with 5k users")
    assert schema.domain == "saas"

misata_preview

Injects misata.preview for DetectionReport tests:

def test_detection_confidence(misata_preview):
    report = misata_preview("A fintech with fraud detection and 2% fraud rate")
    assert report.domain == "fintech"
    assert report.domain_confidence in ("high", "low")

Reproducibility

Every misata_fixture call with the same seed produces identical data. Tests are fully deterministic:

saas_a = misata_fixture("A SaaS company", rows=100, seed=42)
saas_b = misata_fixture("A SaaS company", rows=100, seed=42)

def test_determinism(saas_a, saas_b):
    import pandas as pd
    pd.testing.assert_frame_equal(saas_a["users"], saas_b["users"])

Scope

By default fixtures have scope="function", fresh data for each test. If you want shared data across a test module (faster), define the fixture manually:

# conftest.py
import pytest
import misata

@pytest.fixture(scope="module")
def shared_ecommerce():
    return misata.generate("An ecommerce store", rows=5000, seed=42)