DataTerminology.comNEW
Also known as: Model Lineage
Documenting model origin, training data, versions, and transformations.
DESCRIPTION
Model provenance records where a model came from and how it was produced: data snapshots, features, code, hyperparameters, environment, and training runs. It enables reproducibility, debugging, and compliance, and supports responsible release and rollback. Provenance is often captured via experiment tracking and model registries integrated into CI/CD for ML.
ALTERNATIVE DEFINITIONS
W3C PROV
Provenance is information about entities, activities, and people involved in producing a piece of data or thing, which can be used to form assessments about its quality, reliability or trustworthiness.
Provides the foundational concept extended to model artifacts and training processes.
MLflow
MLflow Tracking is an API and UI for logging parameters, code versions, metrics, and artifacts when running machine learning code and for later visualizing and comparing the results.
Operational tooling to capture the model’s provenance across runs.
OpenLineage
OpenLineage is an open standard for lineage metadata collection and analysis for data pipelines.
Focuses on standardized lineage linking data and model artifacts.
RELATED TERMS
Traceability Data Lineage Reproduction Model Registry MLflow Metadata Management Versioning Experiment Tracking Model Cards Documentation
COMPETING TERMS
Black-box Modeling
RELATIONS
- Broader: Provenance
KNOWLEDGE AREAS
MLOps Metadata Management AI Governance Data Management
TARGET AUDIENCE
MLOps Engineer AI Engineer Data Scientist Data Steward Data Architect
EXTERNAL SOURCES
- Wikipedia - Provenance
- W3C - Core provenance standard
- MLflow - Experiment tracking for provenance
- OpenLineage - Lineage standard for pipelines
- Microsoft - MLflow integration on Azure
TOP QUESTIONS
Why does Model Provenance matter?
It enables reproducibility, debugging, compliance evidence, and safe rollback, all crucial for responsible ML operations.
Category: Relevancy
What should Model Provenance capture?
Data snapshots, feature code, model code and version, parameters, environment, training metrics, and artifacts with immutable identifiers.
Category: Requirements
How is Model Provenance implemented?
Use experiment tracking, model registries, lineage capture, and CI/CD integration to automatically log and link all artifacts.
Category: How-To
What is a common mistake about Model Provenance?
Storing only final model binaries without run context makes reproduction and audits nearly impossible.
Category: Mistakes
How does Model Provenance relate to Traceability?
Provenance documents model creation; traceability connects the model to data sources and downstream decisions.
Category: Relation
How do you evaluate Model Provenance quality?
Attempt full reproduction using recorded artifacts, verify hash matches, and check lineage completeness across inputs and outputs.
Category: Testing
What tools support Model Provenance?
MLflow, model registries, metadata stores, and lineage standards like OpenLineage enable end-to-end capture.
Category: Tooling
How long should Model Provenance be retained?
Retention should align with regulatory, contractual, and risk-based policies; high-risk domains often require extended retention.
Category: Management
What is the risk of weak Model Provenance?
Inability to reproduce or defend decisions, longer incident recovery, and increased compliance exposure.
Category: Risks