Machine Learning BOM
Gives visibility into security, privacy, safety, and ethics of ML systems by listing models and the data that produced them.
Spec focus
components[type=machine-learning-model], modelCard, data
Graph labels
BomDocument, Component, ModelCard, Dataset
Lifecycle
build
Mapping. modelCard fields become a :ModelCard node. Datasets are first-class so lineage queries can walk TRAINED_ON.
Extension. The anomaly model used on the HarborPulse gateway is linked to the SBOM (onnxruntime) and the HBOM (edge device it ships on).
Documents
PulseAnomaly v3 ML-BOM
urn:uuid:7a2e1c90-aa11-4b33-9d01-mlbom004
Subject PulseAnomaly · v4 · 2026-03-01T11:15:00Z
Open in graphNeighborhood
Drag to pan · scroll to zoom · 6 nodes · 6 edges
Select a node to inspect CycloneDX properties.
Sample Cypher
All BOM documents
Every CycloneDX document in the Atlas graph, with its type.
ML model lineage
From dataset through model card to the component that ships the model.
Components shared across BOM types
Inventory that appears in more than one BOM variety — the knowledge-graph payoff.
Knowledge concept overlay
How CycloneDX instances map onto the Atlas ontology.
Graph size by label
How many Atlas nodes exist for each CycloneDX class.
Neighborhood of HarborPulse API
One- and two-hop context around the application component.
Models, datasets, training lineage, and intended use.