Start HereStrong architectural inferencev1.22.1
Model Risk Versus Ecology Risk
Evidence levelStrong architectural inferenceTechnical label: Architectural inference
Model risk asks whether an artifact behaves acceptably under tested conditions. Ecology risk asks whether the changing relationships among artifacts, memory, tools, evaluators, routers, and operators remain acceptable over time.
| Question | Model risk | Ecology risk |
|---|---|---|
| Identity | Which model hash? | Which composition and transition graphThe map of how an AI system is allowed to change over time. Open glossary definition? |
| Evidence | Benchmark and red-team results | Benchmark, route, memory, permission, evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition, lineage, and release evidence |
| Failure mode | Artifact misbehavior | Interaction, succession, evaluator, or rollbackReturning a system to an earlier known state. Open glossary definition failure |
| Retirement | Delete or unpublish model | Prove behavior is not expressible across active carriers and descendants |
| Rollback | Restore weights | Restore weights, adapters, router, memory, prompts, permissions, evaluator, aliases, and dependencies |
Practical consequence
A model card can be true and still incomplete. The runtime composition and allowed transitions need their own manifest.