StatGuardian

A Rust-native data quality engine with a declarative contract DSL for schema validation, drift detection, and anomaly detection for Pandas and Polars DataFrames.

Not to be confused with pipeline validators like PyDBTGuard or PyBeamGuard: those check pipeline definitions before they run; StatGuardian checks the quality of the data at runtime.

Which project do I need?

Part of the Mullassery open-source family. These tools have similar names — here is the short version of who does what:

Browse all projects →

What it is

StatGuardian validates data at runtime against versionable contracts written in a declarative .sg DSL. It catches schema violations, statistical drift, and anomalies before they reach downstream consumers. The core validation and statistics logic run in Rust while providing a Python API for integration into data pipelines.

Features / Highlights

• Declarative contracts: Define schema, quality rules, statistical drift thresholds, and anomaly checks in one .sg file.
• Rust-native execution: Core checks and statistics execute in compiled Rust for performance.
• Multi-backend support: Validate Pandas and Polars DataFrames with the same contract. Delta Lake and Apache Iceberg tables are supported as inputs.
• Drift detection: Detect distribution shifts between datasets using PSI and KS tests via the DSL.
• Anomaly detection: Identify outliers and duplicates with built-in rules.
• Blocking controls: Severity tiers let you mark critical checks as @blocking to gate pipeline execution.
• Detailed reports: Structured violation messages, scores, and per-check results.

Who it's for / Use cases

• Data engineering pipelines: Enforce contracts on batch or streaming inputs before writes.
• MLOps/feature pipelines: Catch schema or distribution drift between training and inference data.
• ETL/data integration: Validate external data against agreed schemas and quality thresholds.
• Data governance: Version and review contracts alongside code as the source of truth.

What it is not

• Not a full observability platform: Focused on contract-based validation; does not include dashboards/alerting infrastructure beyond returning structured reports.
• Not a generic testing framework: Designed for data contract enforcement at runtime and in pipelines.
• Benchmarks vary: Performance depends on workload (row count, columns, checks). Run benchmarks against your own data before making comparisons.

Project status

StatGuardian is in active development. Core features (schema validation, quality rules, drift, anomaly detection, multi-backend DataFrame support) are implemented. Streaming validation (StreamGuard) and additional integrations are documented under development paths; see docs/ROADMAP_HONEST.md for known limitations and planned work.

Installation

From PyPI

pip install statguardian

From source (Python + Rust)

pip install maturin
maturin develop

System requirements

• Python 3.8+
• Rust toolchain (for source builds)

Quick start

import polars as pl
import statguardian

contract = statguardian.DataContract.from_dsl("""
dataset orders {
    schema {
        order_id: string, not_null, unique
        amount:   float,  positive
        status:   string, not_null, enum=["pending","paid","cancelled"]
    }
    quality {
        completeness(order_id) > 0.999
    }
}
""")

df = pl.read_parquet("orders.parquet")
report = statguardian.execute(contract, df)
print(report.summary())
print(f"Passed: {report.passed}")

Usage

Validate a Polars DataFrame against a contract:

import statguardian

contract = statguardian.DataContract.from_dsl("dataset orders { schema { id: string, not_null } }")
report = statguardian.execute(contract, df)

Drift detection between reference and incoming data:

report = statguardian.execute(contract, incoming_df, reference=baseline_df)
for d in report.drift_results():
    if not d["passed"]:
        print(f"Drift: {d['column']} PSI={d.get('psi', 0):.4f}")

Load a contract from a .sg file:

contract = statguardian.DataContract.from_file("contracts/orders.sg")

Note: When validating a Pandas DataFrame, convert it to Polars first (e.g., pl.from_pandas(df)) before calling execute().

Documentation

• CLI documentation
• StreamGuard (streaming validation)
• Data contract DSL
• Benchmarking
• Honest roadmap & limitations

Contributing

Contributions are welcome. Please review CONTRIBUTING.md for guidelines on development setup, code style, and testing.

License

Licensed under the Apache License, Version 2.0. See LICENSE for details.