EMEP Component Dependency Graph: 17 Internal Abstractions

The EMEP platform is composed of 17 internal abstractions, each with a defined interface, responsibility, and dependency set. This page documents which components call which, which components produce data consumed by others, and which are leaf nodes with no downstream dependents within the platform.

The 17 Components

#	Component	Responsibility
1	ModelRegistry	Catalogs all models, their metadata, lifecycle states, and provenance
2	ModelLoader	Loads model weights and configurations into memory
3	ModelCompatibilityAnalyzer	Determines compatibility states between model pairs
4	TensorEngine	Executes low-level tensor operations (add, scale, interpolate, mask)
5	MergeEngine	Orchestrates merge execution using a selected strategy
6	MergeStrategy	Abstract interface for merge algorithms (Linear, SLERP, TIES, DARE, etc.)
7	CandidateGenerator	Packages merge outputs into candidate models with status and lineage
8	EvolutionEngine	Drives mutation, crossover, and selection over candidate populations
9	EvaluationEngine	Orchestrates benchmark execution and metric aggregation
10	BenchmarkEngine	Executes individual benchmark tasks and collects raw scores
11	FitnessEngine	Computes scalar or vector fitness from evaluation results
12	ExperimentTracker	Logs all experiment state transitions, parameters, and results
13	ArtifactStore	Stores, retrieves, and verifies all artifacts by hash
14	DatasetRegistry	Catalogs datasets, versions, licenses, and contamination status
15	QuantizationEngine	Reduces model precision for deployment
16	InferenceBackend	Serves model inference requests
17	DeploymentManager	Manages deployment, health checks, rollback, and updates

Dependency Rules

Acyclic: The component graph must remain a directed acyclic graph (DAG) at all times. No component may directly or indirectly call itself.
Interface-based: All dependencies are through defined interfaces, not concrete implementations. This enables substitution and testing.
Eventual consistency: Components that write to shared stores (ModelRegistry, ArtifactStore) do not require synchronous confirmation from readers.
Failure isolation: A failure in one component must not cascade to unrelated components. The ExperimentTracker logs all failures independently.

Dependency Graph

Dependency Matrix

The following table shows direct dependencies. A checkmark indicates that the row component directly calls or reads from the column component.

	MR	ML	MCA	TE	MS	ME	CG	EvE	BE	FE	ET	AS	DR	IB
MR
ML	check
MCA	check	check
TE
MS				check
ME	check	check	check		check
CG	check					check					check	check
EE	check						check			check	check
EvE	check								check		check		check
BE													check
FE	check							check
ET	check											check
AS
DR
QE												check
IB														check
DM	check											check		check

Leaf Nodes

The following components have no downstream dependents within the EMEP platform. They are terminal nodes in the dependency graph.

TensorEngine: Leaf computation node. All merge strategies depend on it, but it has no internal EMEP dependents.
ArtifactStore: Leaf storage node. Many components write to it; none read from it for further platform computation.
DatasetRegistry: Leaf catalog node. BenchmarkEngine reads from it; no component depends on DatasetRegistry outputs for further processing.

The interfaces between these 17 components are defined in API Specification and Configuration Specification. Any change to an interface requires an ADR and updates to all dependent components.

System Dependency Map for the end-to-end pipeline view
Data Flow Map for weight, vector, and artifact movement
API Specification for component interface definitions
Architecture Diagrams for visual architecture overviews

Project Context

Research Foundation

System Maps

EMEP Component Dependency Graph: 17 Internal Abstractions

The 17 Components

Dependency Rules

Dependency Graph

Dependency Matrix

Leaf Nodes

​The 17 Components

​Dependency Rules

​Dependency Graph

​Dependency Matrix

​Leaf Nodes

​Related Pages

The 17 Components

Dependency Rules

Dependency Graph

Dependency Matrix

Leaf Nodes

Related Pages