Artifact Inspection
A completed Forge execution produces more than a computational result.
It may also produce execution artifacts.
Artifacts preserve additional evidence generated during execution and provide a richer view of the computation than the primary result alone.
Unlike the primary result, which answers the computational question, artifacts explain how that answer can be inspected.
Artifact inspection is therefore an essential part of independent verification.
Verification Objective
After completing this document you should understand:
- what execution artifacts are
- why artifacts exist
- how artifacts differ from computational results
- how artifacts strengthen independent verification
- what conclusions artifacts support
- what conclusions artifacts do not support
Prerequisites
Before continuing, complete:
- Verification Overview
- Canonical Smoke Suite
- First Verified Execution
- Inspect Result
- Replay & Determinism
You should already have a completed execution with generated artifacts.
What Is an Artifact?
An artifact is an execution product generated by the runtime in addition to the primary computational result.
Artifacts preserve information that would otherwise be lost after execution completes.
They extend the observable evidence available to evaluators without changing the execution outcome.
Artifacts as Evidence
Forge treats artifacts as part of the execution evidence surface.
Canonical Execution
│
▼
Primary Result
│
├─────────────┐
▼ ▼
Execution Summary Execution Artifacts
│ │
└──────┬──────┘
▼
Independent VerificationThe primary result answers the computational question.
Artifacts provide additional evidence supporting that answer.
Artifact Categories
Artifacts vary by capability and execution family.
Common artifact categories include:
Summary Artifacts
Provide a concise overview of the execution outcome.
Typical examples include:
- summary statistics
- confidence summaries
- optimization summaries
- aggregate metrics
These artifacts are often the first surface inspected during verification.
Distribution Artifacts
Describe the distribution of execution outcomes.
Representative examples include:
- histograms
- probability distributions
- percentile summaries
- frequency tables
These artifacts help evaluators understand variability rather than only point estimates.
Structural Artifacts
Describe relationships discovered during execution.
Examples include:
- graph topology
- propagation paths
- dependency chains
- critical nodes
- influence maps
Structural artifacts are particularly important for graph-oriented workloads.
Ranking Artifacts
Capture ordered execution outputs.
Examples include:
- ranked candidates
- optimization frontiers
- best candidate
- worst candidate
- candidate score tables
Ranking artifacts preserve decision context.
Temporal Artifacts
Describe execution behavior over time.
Representative examples include:
- timelines
- progression curves
- simulation evolution
- execution snapshots
These artifacts support longitudinal inspection.
Diagnostic Artifacts
Provide additional information useful during verification.
Examples include:
- confidence maps
- disagreement maps
- mutation traces
- execution previews
- sampling diagnostics
Diagnostic artifacts strengthen the evaluator's understanding without changing the computational result.
Artifact Interpretation
Artifacts should always be interpreted together with:
- the execution contract
- the execution context
- the primary result
- the replay metadata
An artifact viewed in isolation rarely provides sufficient verification evidence.
What Artifacts Prove
Artifacts strengthen confidence by demonstrating that additional execution evidence was preserved.
Depending on capability, artifacts may demonstrate:
- execution completeness
- statistical behavior
- ranking rationale
- structural relationships
- simulation characteristics
- optimization trade-offs
Artifacts therefore extend observable execution evidence.
What Artifacts Do Not Prove
Artifacts should not be interpreted as independent proof that a computation is correct.
They do not replace:
- execution identity
- canonical validation
- replay verification
- execution contracts
Artifacts explain execution.
They do not authorize execution.
Artifact Availability
Not every capability produces the same artifacts.
Artifact availability depends on:
- execution family
- primitive
- profile
- requested outputs
- execution policy
The canonical capability contract defines which artifacts may be produced.
Verification Strategy
Artifact inspection should follow successful execution.
Recommended order:
Execution Identity
│
▼
Primary Result
│
▼
Summary Artifacts
│
▼
Supporting Artifacts
│
▼
Replay MetadataInspecting artifacts before confirming execution identity is not recommended.
Common Misinterpretations
"Artifacts are the result."
Artifacts complement the result.
They are not the computational result itself.
"More artifacts mean better execution."
Artifact quantity is not a measure of execution quality.
Different capabilities naturally produce different evidence surfaces.
"Artifacts prove correctness."
Artifacts strengthen verification.
They do not independently prove computational correctness.
"Every capability produces identical artifacts."
Artifact production is capability-dependent.
The canonical capability contract defines expected outputs.
Verification Outcome
Artifact inspection demonstrates that Forge preserves observable execution evidence beyond the primary computational output.
This additional evidence enables evaluators to:
- inspect execution behavior
- understand computational context
- interpret domain-specific outputs
- compare replayed executions
- communicate verification findings more effectively
Artifacts therefore extend verification rather than replace it.
Next Steps
Continue with:
Capability-specific verification demonstrates how these evidence patterns are applied to individual execution families.
Related Documentation
- Verification Overview
- Canonical Smoke Suite
- First Verified Execution
- Inspect Result
- Replay & Determinism
- Negative Validation
- Execution Model
- Trust Overview
Final Principle
The primary result answers the computational question.
Execution artifacts preserve the supporting evidence required to inspect, interpret, and communicate that answer.
Artifacts therefore extend the observable evidence surface of a Forge execution rather than serving as additional computational outputs.