dqtlib 1.3.0

Data quality, observability, semantic, and causality library

dqtlib

Open-source data quality, lineage, semantic layer & causality — for dbt, warehouses and data lakes.

pip-installable Python library for watching dbt-built warehouses and any SQL warehouse for statistical drift, anomalies, silent regressions, and explaining why metrics moved.

pip install dqtlib

The import name is dqt:

from dqt import Check, Runner, MemoryStore

Full documentation and examples: https://github.com/antonbarr-data/dqt

Quality

All 64 detectors are benchmarked against labeled synthetic datasets. Benchmark scripts and raw results live in the GitHub repo and are run on every release:

Median F1: 1.00 | Detectors with F1 >= 0.8: 52/64 | Detectors with F1 >= 0.6: 58/64

Per-detector breakdown: examples/benchmarks/results_summary.md

Reproduce by cloning the repo and running:

python scripts/run_benchmark_suite.py --quick
python scripts/generate_benchmark_summary.py

Adapters

Six adapters ship in v1.0. Nightly CI runs against live credentials for the cloud warehouses.

Adapter	CI
PostgreSQL (local)	bundled — no credentials needed
ClickHouse
Snowflake
BigQuery
Databricks
Local (DuckDB)	bundled — no credentials needed

Detector documentation

64 statistical detectors across 10 groups — drift, outliers, time series, distribution, information theory, pattern, referential, schema, basic, and custom.

Every detector has a structured page at docs/algorithms/<group>/<slug>.md covering:

• What it computes and its parameters
• When it works well and when it fails (with concrete failure-mode table)
• Default-threshold calibration — empirical FPR across six canonical data shapes (Normal, Lognormal, Poisson, Beta, Pareto, Exponential)
• Recommended thresholds per data shape
• Canonical citation and runnable Python API example

Browse the full catalog: docs/algorithms/README.md