Package anchor
Weight Integrity
Question: Is this the intended model-weight artifact and has its integrity been preserved?
Model Weight Integrity & Trust/Governance · package anchor
Weight Integrity
A semantic anchor for the integrity, provenance, attestation, controlled movement, and verifiable use of machine-learning model weights.
Model weights can be high-value artifacts that are replicated, moved between environments, deployed, modified, substituted, exposed, or used for inference. This package supplies bounded public identities for discussing distinct security and governance questions around those artifacts.
Five-domain architecture
Five distinct trust questions around one model-weight artifact
These identities are an LJP semantic organization, not a standards-defined lifecycle or universal implementation architecture. Each remains an independent identity.
Origin and lineage identity
Weight Provenance
Question: Where did this model-weight artifact originate and what is its relevant lineage or custody history?
Evidence and attestation identity
Weight Attestation
Question: What evidence or claims support assertions about the artifact or its relevant state and context?
Controlled-movement identity
Weight Egress
Question: How is movement or export of the artifact across boundaries controlled?
Emerging verification and use identity
Machine-Verifiable Inference
Question: What machine-evaluable evidence can support claims about inference execution or use of an intended model artifact?
Relationship model
Related questions without interchangeable answers
ArtifactWeight Provenance frames origin and lineage.
Artifact stateWeight Integrity frames preservation and intended identity.
Claims and evidenceWeight Attestation frames supporting assertions.
MovementWeight Egress frames export and boundary crossing.
Operational useMachine-Verifiable Inference frames emerging evaluable evidence.
Evidence boundary
Authority, research, and LJP synthesis are kept separate
| Identity | Evidence class | Authoritative basis | Boundary |
|---|---|---|---|
| Weight Integrity | C | RAND model-weight security research; public security practice | The exact LJP identity is not a standard. |
| Weight Provenance | C | C2PA AI/ML guidance | Provenance context does not establish this package role as a standard. |
| Weight Attestation | C | RFC 9334 attestation architecture | The compound identity is an LJP application, not a ratified term. |
| Weight Egress | C | Public model-weight security practice | Does not imply an LJP enforcement service. |
| Machine-Verifiable Inference | B | emerging verifiable-ML research | Not an adopted universal standard. |
Buyer walkthrough
Evaluate a problem, not a promised deployment
Frontier or model-provider security
Which artifact is protected, can provenance and integrity be evaluated, and how is movement bounded?
Enterprise AI platform security
Which model artifact is deployed, what evidence accompanies it, and how are deployment boundaries governed?
AI governance and assurance
What supports a claim about an artifact, what is independently inspectable, and what remains operational assertion?
Security and infrastructure teams
Where are controls enforced, which systems remain replaceable, and which semantic identities persist across implementations?
Adjacent governance identities
Related Trust & Governance identities
These independent PKG7 identities are adjacent context, not components of the five-domain model-weight architecture.
Evaluation
Inspect the package directly
- Open all five identities and compare their distinct definitions.
- Inspect the cited public material and separate its scope from LJP synthesis.
- Review namespace.json, llms.txt, robots.txt, and sitemap.xml.
- Confirm that no page claims to operate model-weight infrastructure, certify an artifact, or impose a universal protocol.
Resources
Public sources and governed descriptors
Model-weight security
RAND research and public model-security practices.
Attestation architecture
RFC 9334 provides its own general evidence and attestation architecture.
Emerging verification
Verifiable-ML research frames an emerging technical direction.