Artifact Graphs and the Preservation of Computational Work

Abstract

A single artifact tells only part of the story. Most computational results are derived from prior results, which were themselves derived from earlier computation — chains of dependency that, taken together, reveal the full structure of the work performed.

In agent-based computational ecosystems, these relationships naturally produce artifact graphs: directed graphs in which nodes represent computational artifacts and edges represent derivation relationships between artifacts.

These graphs encode the structure of computational work performed by autonomous systems.

This document introduces the artifact graph model, establishes the Artifact Graph Axiom, and derives the Artifact Graph Theorem, demonstrating that the loss of artifacts results in the collapse of portions of the computational graph.

This leads to the principle of Computational Work Conservation, which motivates the need for persistent artifact availability in distributed agent ecosystems.

1. Artifact Graphs

As established in the previous note, computational artifacts possess identity and derivation relationships. Artifacts frequently depend on other artifacts that were produced earlier in a computational workflow.

Examples include:

datasets produced by data collection processes
analyses derived from those datasets
reports summarizing analytical results
summaries derived from reports

These relationships naturally form a directed dependency structure.

When artifacts and their derivation relationships are represented explicitly, they form an artifact graph.

In an artifact graph:

nodes represent computational artifacts
edges represent derivation relationships between artifacts

Artifact graphs therefore capture the structural relationships between computational results.

2. Artifact Graphs as Representations of Computation

Artifact graphs encode the structure of computational workflows.

Each node represents the output of a completed computation, and each edge represents a derivation relationship between artifacts.

This structure implicitly describes:

the order of computational steps
the dependencies between stages of computation
the lineage of derived results

Artifact graphs therefore provide a structural representation of computational work performed within a system.

3. Artifact Graph Model

To reason about artifact graphs more precisely, we define a simple model.

An artifact graph is a directed graph G = (V, E) where:

V represents the set of computational artifacts
E represents derivation relationships between artifacts

An edge (a → b) indicates that artifact b was derived from artifact a.

Under this model:

artifacts correspond to graph nodes
derivation relationships correspond to directed edges
computational workflows produce new nodes and edges

The resulting graph captures the structural relationships between computational results.

Artifact graphs therefore represent the accumulated structure of computational work performed within a system.

4. Artifact Graph Axiom

The structure of artifact relationships leads to a foundational observation.

Artifact Graph Axiom

Any non‑trivial computational system produces artifacts whose derivation relationships form a directed graph of computational work.

As computational systems increase in complexity, these graphs grow correspondingly larger and more interconnected.

Agent-based systems frequently produce deep artifact graphs as agents perform chains of computational tasks across distributed environments.

5. Artifact Graph Theorem

A direct consequence of the Artifact Graph Axiom is the following.

Artifact Graph Theorem

If artifacts within a computational artifact graph are not persistently available, the graph becomes partially or wholly irreconstructible.

If an artifact disappears, any artifacts derived from that artifact lose the ability to verify or reconstruct their lineage.

For example:

a dataset artifact may be lost
an analysis artifact derived from that dataset may still exist

Although the analysis artifact remains available, the derivation relationship connecting it to the dataset can no longer be validated.

As artifacts disappear, increasing portions of the artifact graph become structurally incomplete.

In sufficiently complex systems, artifact loss results in the collapse of large portions of the computational graph.

6. Computational Work in Artifact Graphs

Each artifact node represents the result of computational effort.

Producing an artifact requires:

computational resources
execution time
data processing
algorithmic computation

The artifact graph therefore represents the accumulation of computational work across a system.

Each node represents completed work, and the structure of the graph reflects how that work was derived.

Preserving artifact graphs preserves the results of this work.

Destroying artifacts destroys the work required to produce them.

7. Computational Work Conservation

The structure of artifact graphs reveals a deeper property of computational systems.

Computational Work Conservation

Systems that preserve the artifacts produced by computation conserve computational work. Systems that lose artifacts destroy computational work.

In distributed computational ecosystems, artifacts represent the results of completed computation. These artifacts form dependency graphs that capture the structure of computational workflows.

When artifacts disappear, the corresponding portions of the artifact graph become irrecoverable.

As a result, the work required to produce those artifacts must be recomputed.

In sufficiently complex systems, repeated artifact loss results in large-scale destruction of accumulated computational work.

Preserving artifact availability therefore becomes necessary for conserving the results of computation across agents, workflows, and time.

8. Implications for Agent Systems

Agent ecosystems frequently involve long chains of derived computation.

Artifacts produced by one agent may become inputs to multiple downstream agents performing additional analysis, synthesis, or transformation.

Without persistent artifact availability:

artifact graphs become fragmented
computational lineage becomes unverifiable
downstream workflows lose reproducibility
previously completed work must be repeated

Maintaining artifact availability is therefore essential for preserving the integrity of computational workflows.

Conclusion

Artifact graphs provide a structural representation of computation within agent ecosystems.

Each node represents the result of computational effort, and each edge captures a derivation relationship between artifacts.

The disappearance of artifacts destroys portions of this graph and results in the loss of accumulated computational work.

Preserving artifact availability is therefore necessary for maintaining the integrity of computational workflows and conserving the results of computation across distributed systems.

Artifact graphs reveal that computational work is not merely performed but structured and accumulated through artifacts. When artifacts persist, the work they represent persists as well.

If artifact graphs represent the structure of computational work, preserving computational systems requires preserving the artifacts that compose those graphs. The next paper examines why traditional storage systems fail to maintain this structure and why new infrastructure is required.

Discussion and Feedback

The ideas presented in this document are part of an ongoing exploration of architectural requirements for agent-based computational systems.

Comments, critiques, and alternative perspectives are encouraged.

Feedback may be submitted through issues or discussions within this repository.

Future notes in this series examine why traditional storage systems fail to preserve artifact graphs, introduce the Artifact Availability Layer required to maintain them, describe deterministic artifact identity, and explore the principle of Computational Work Conservation.

Citation

If referencing this work, please cite:

Kopcho, Rich. Artifact Graphs and the Preservation of Computational Work.
Agent Artifact Availability (AAA) Series. Technical Note, March 2026.