agi-app-data-quality-gate

agi-app-data-quality-gate packages the data_quality_gate_project AGILAB app. It is a deterministic data contract, drift, leakage, and promotion-gate example for teams that need a concrete proof before a candidate dataset reaches model training or pilot promotion.

Purpose

Use this package to show how AGILAB can turn a data-readiness review into replayable evidence. The app generates a baseline dataset and a candidate dataset, validates the expected columns, profiles quality, measures drift, and writes a decision that can be reviewed before another system takes ownership.

What You Learn

The packaged project demonstrates the same contract-first workflow without requiring a source checkout. A first run shows the generated datasets, the quality profiles, the drift table, the gate decision, and the manifest that ties those artifacts together. It is intended to make a data promotion review easy to rerun and easy to inspect from AGILAB.

Installed Project

The distribution name is agi-app-data-quality-gate; the AGILAB project name is data_quality_gate_project. The package exposes both data_quality_gate and data_quality_gate_project through the agilab.apps entry point group, so AgiEnv(app="data_quality_gate_project") resolves the project without a monorepo checkout once this payload package is installed.

Install

pip install agi-app-data-quality-gate

This is the stable package install shape once this distribution is promoted to PyPI. For the current release artifact path, install the wheel directly:

pip install /path/to/agi_app_data_quality_gate-<version>-py3-none-any.whl

This app project is built as wheel and source-distribution artifacts in the GitHub Release archive, but it is not promoted to PyPI in the current release plan and is not pulled by the agi-apps umbrella. Install it directly only when validating the data quality gate package from a release artifact or a locally built wheel.

Run In AGILAB

Select data_quality_gate_project, open ORCHESTRATE, then run Deploy scheduler & workers and RUN. Open ANALYSIS or inspect the exported evidence directory to review the contract, drift metrics, gate decision, and artifact manifest.

Expected Inputs

The default run generates deterministic synthetic baseline and candidate datasets. It does not require private data, a model registry, a cloud account, an LLM, or an external network service.

Expected Outputs

The app writes baseline and candidate CSV files, JSON profiles, a data contract, drift metrics, a gate decision, a Markdown evidence report, a run manifest, and a data-quality summary with artifact hashes.

Change One Thing

Change only drift_strength, then rerun the app. Lower values should move the gate toward promote; higher values should move it toward manual-review or block. Keep seed=2026 when you want artifact deltas that remain easy to explain.

Troubleshooting

If the package resolves but custom data does not, rerun the default synthetic case first. Then verify that CSV and JSON paths are AGILAB-share-relative and that the candidate file contains every column required by the contract. A noisy or unexpected manual-review decision usually means the drift threshold was tighter than the candidate distribution, so inspect drift_metrics.csv before loosening the gate.

Scope

This is a compact data-quality gate example. It does not replace a full data observability platform, feature store, enterprise governance workflow, or production approval authority. Its purpose is to make one data-readiness review portable, deterministic, and evidence-backed.