mdmp-protocol 0.1.0

MDMP core protocol for dataset contracts, grading, fingerprints, and AI lineage cards

MDMP

MDMP is an open protocol and tooling stack for dataset quality and AI training provenance.

Tagline: Know what your AI learned from.

30-second summary

MDMP gives every dataset a contract, a grade, and a fingerprint. It gives every model a lineage card that points to exactly which dataset fingerprints were used.

• Contract: schema, ranges, consent metadata.
• Validation: reproducible checks + deterministic grade.
• Fingerprint: immutable dataset identity (sha256:...).
• Lineage card: model-to-dataset traceability with stale detection.

Install

python -m venv .venv
source .venv/bin/activate
python -m pip install -U pip
pip install -e .

Quick Start

mdmp init --flavor health --output contracts/mdmp_contract.yaml
mdmp validate contracts/mdmp_contract.yaml data/demo_cgm.csv --output-json results/mdmp_report.json
mdmp report results/mdmp_report.json --output-html results/mdmp_dashboard.html

Core Commands

# grading + fingerprint
mdmp grade contracts/mdmp_contract.yaml data/demo_cgm.csv
mdmp fingerprint data/demo_cgm.csv

# stale lineage lifecycle
mdmp fingerprint-record data/demo_cgm.csv --output-json results/fingerprint.json --expires-days 365
mdmp fingerprint-check results/fingerprint.json data/demo_cgm.csv
mdmp lineage-card --model glucose_forecaster_v2 --dataset data/demo_cgm.csv --contract contracts/health_demo.yaml --output results/mdmp_model_card.yaml
mdmp lineage-card-refresh results/mdmp_model_card.yaml

# local registry scaffold
mdmp registry init --registry registry/mdmp_registry.json
mdmp registry push --registry registry/mdmp_registry.json --report results/mdmp_report.json --visibility public --model-id glucose_forecaster_v2
mdmp registry lookup sha256:YOUR_FINGERPRINT --registry registry/mdmp_registry.json
mdmp registry list --registry registry/mdmp_registry.json

# Hugging Face section export
mdmp hf-export --dataset-id python35/demo-cgm --report-json results/mdmp_report.json --output-md results/mdmp_hf_section.md

Documentation

• Spec: MDMP_SPEC.md
• Docs index: docs/index.md
• CLI reference: docs/reference/cli.md
• Launch checklist: docs/launch/launch-checklist.md
• Contributing: CONTRIBUTING.md

Release Automation

• GitHub release workflow: .github/workflows/release.yml
• PyPI publish workflow: .github/workflows/publish-pypi.yml
• Docs deploy workflow: .github/workflows/docs-site.yml

Design Boundaries

• MDMP stores metadata + fingerprints, not raw datasets.
• Local-first workflow; cloud registry is optional.
• Complementary to DVC / MLflow / W&B.

Status

• Spec version: v0.1-draft
• Research/provenance utility.
• Not a medical device and not clinical decision support.

Integrations

• IINTS integration guide: docs/IINTS_INTEGRATION.md