Pipeline testing

Contract-to-Test synthetic data

Turn schema contracts and a supported SQL workload into deterministic Pandas datasets containing bounded predicate, join, NULL, and boundary witnesses.

The compiler parses SQL but does not execute it or read production rows. Coverage applies only to the documented Phase 1 subset.

Read the full guide API reference

Compile a workload

from great_generator.testing import compile_test_dataset

result = compile_test_dataset(
    schema=tables,
    queries=[sql],
    coverage=["predicates", "joins", "boundaries"],
    seed=42,
)

assert result.validate()["passed"]
print(result.coverage_report())

Measured artifacts

The result includes clean DataFrames, separate labeled negative cases, parsed query contracts, scenario row bindings, bounded expected outcomes, explicit coverage denominators, and a reproducibility manifest.

Current execution boundary

Phase 1 validates Pandas and acyclic simple relationships. Spark-native workload witnesses, generic composite/cyclic generation, temporal consistency, and arbitrary full-query oracles remain gated later phases.

Evidence, not claims

The repository includes one controlled missing-filter mutation fixture. It is a reproducible local result, not a universal benchmark or originality claim.

Implementation and research details

Feature-gap analysisCode-evidenced baseline and phased gaps. Prior artRelated tools, published approaches, overlaps, and testable differences. Evaluation protocolControls for fair, reproducible mutation experiments. Query-aware generationUse required values, selectivity, partitions, and basic join coverage without SQL compilation.