Glossary
Use this page as a reference while navigating Studio and the API.
These definitions describe how concepts are used in BHANUJ. Follow the links for their full lifecycle and API contracts.
Candidate
An immutable experiment configuration that binds a prompt version, model version, dataset version, evaluator installation, Runtime Connection, and supported runtime parameters.
Evaluation Provider
A configured evaluator installation that scores model output against a stable provider contract. It is separate from the model runtime.
Runtime Connection
Tenant-scoped endpoint configuration and secret references used to invoke a managed model. Credentials are not returned to Studio or stored in candidate evidence.
Managed Asset
A prompt, model, or dataset version created or lifecycle-managed by BHANUJ.
Observed Asset
Immutable evidence reported by an external runtime or evaluation producer. BHANUJ records it but does not claim to author or deploy it.
Evidence
Facts used to support a decision, such as evaluation results, execution records, drift signals, or governed asset versions.
Lineage
The wider graph of relationships among assets, candidates, policies, jobs, and decisions. It is traceability context, not the decision input itself.
Policy
A versioned definition containing ordered structured rules that evaluate available evidence.
Governance Decision
A persisted, explainable outcome such as APPROVED, REJECTED, BLOCKED, or PROPOSED, with its policy and evidence context.
Replay
A new governed reproduction based on frozen source-execution evidence. It is not a retry of the original execution.
Continue with How AI Governance Platform Works for the relationship between these concepts.