MoonDataDrift

Project identifier: moondatadrift

Repository:

• GitHub: https://github.com/pla678889/MoonDataDrift
• GitLink: https://gitlink.org.cn/bla6677/moondatadrift

MoonDataDrift is a MoonBit library and CLI for comparing a baseline table with a current table and surfacing likely data drift.

It focuses on practical checks that are useful in ML monitoring:

• missing-rate changes
• numeric quantiles
• PSI
• KS-like distribution distance
• categorical frequency shift

Why this project

This repository is meant for the 2026 MoonBit open-source competition. The scope is intentionally compact enough to be finished well, but open-ended enough to grow into:

• file-based CSV ingestion
• richer charts and HTML reports
• drift thresholds and alerting
• CI integration for model pipelines

Current scope

The current version works on in-memory tables and ships with a runnable demo report.

Sample data lives in data/, and the sample report outline is documented in examples/demo-report.md.

What it does

• compares two tables with the same schema
• classifies each column as numeric or categorical
• reports missing-rate movement
• computes numeric p10 / p50 / p90 summaries
• computes PSI for numeric bins and categorical distributions
• computes a KS-like cumulative distance
• emits a Markdown report from the CLI

What it does not do yet

• it does not read files from disk yet
• it does not render HTML dashboards yet
• it does not implement advanced outlier or time-window detection yet

Run

moon run cmd/main

Test

moon test

License

Apache-2.0

Notes for competition review

• MoonBit is the primary implementation language.
• The repo includes a clear README, a license, a runnable demo, and tests.
• The project direction is aligned with monitoring and MLOps, while remaining small enough to maintain cleanly.
• The source notes and competition checklist are in docs/.