Crate agent_evaluate_contract

Expand description

HTTP API contract for the Evaluate service.

Evaluate is a control plane over four immutable catalog resources — Case, Environment Profile, Evaluator and Suite — plus the durable Task that a Runtime pulls from. Case is the node that owns the dependency edges, so it carries its own id, version and digest like every other catalog resource; a Suite is only an ordered selection of Case versions.

§Layering

agent-evaluate-contract   paths + DTOs        (this crate)
agent-evaluate-runtime    control plane + verdicts
agent-evaluate-server     router + auth + bin

Re-exports§

pub use gate::*;
pub use metric::*;
pub use objective::*;
pub use review::*;

Modules§

gate: Report comparison and the declarative release gate.
metric: Metric taxonomy, report slices and metric health.
objective: Objective metrics derived deterministically from one authoritative Trace partition.
review: Human review queue and the error-analysis sampling protocol.

Structs§

AdminCasePage
AdminCaseSummary
AdminReviewPage
AdminRunPage
AdminRunSummary
AdminSuitePage
AdminSuiteSummary
AgentEvaluationInput: Stable input passed to a repository-owned Agent evaluation function. It is deliberately independent from the function implementation language.
AgentEvaluationOutput: Stable output produced by one repository-owned Agent evaluation function.
ApiError: Error body shared by every non-2xx response.
AssertionVerdict
CancelRunRequest
CaseRunPage
CaseRunSummary
ClaimEvaluationCaseRequest: Runtime-facing pull protocol. evaluate_task_id is the durable EvalOperation::id returned when the SDK/CLI configures a run. Infra never needs a callback URL for the Runtime.
ClaimEvaluationCaseResponse
CreateEvalRunRequest
EnvironmentProfileVersion
EvalAccessScope
EvalBudget
EvalCaseResult
EvalCaseVersion
EvalOperation
EvalReport
EvalSuiteVersion: Ordered selection of immutable Case versions. A Suite adds no content of its own, so changing one Case never rewrites the others.
EvalTurn
EvaluateAdminOverview
EvaluationCaseLease
EvaluationSubject: Agent is an optimistic precondition; Runtime ID remains the execution selector and Runtime is the authoritative owner of their binding.
EvaluatorVerdict
EvaluatorVersion
EvidenceRef: Immutable evidence object owned by the scoped Evaluate namespace.
ExperimentVariant
GetOperationRequest
GetReportRequest
JudgeCalibration: Human-label agreement accumulated for one evaluator id.
JudgeEvaluationEvidence: Attested outcome of a model-graded evaluator. Produced by the isolated executor; the control plane never calls a judge model itself.
ListAdminCasesRequest
ListAdminReviewsRequest
ListAdminRunsRequest
ListAdminSuitesRequest
ListCaseRunsRequest
PluginEvaluationEvidence
RegisterEnvironmentProfileRequest
RegisterEvalCaseRequest: Immutable Case content. A Case is the node that owns the evaluation dependency edges, so it is registered and versioned in its own right and a Suite only references it.
RegisterEvaluatorVersionRequest
RegisterSuiteRequest
RenewEvaluationCaseRequest
RenewEvaluationCaseResponse
SandboxExecutionEvidence: Bounded evidence returned by the isolated executor. Verdict calculation remains owned by Evaluate and is deterministic from this typed structure.
SandboxExecutionRequest
SubmitEvaluationEvidenceRequest
SubmitEvaluationEvidenceResponse
VerifyCommand
VersionedResourceRef: Immutable reference to a catalog resource. Evaluate snapshots the resolved version into a CaseAttempt; mutable aliases are deliberately unsupported.

Enums§

CaseFailureKind
CaseRunState
EvaluatorImplementation
OperationState
TypedAssertion: Versioned typed assertion schema used by immutable Case versions.

Crate agent_evaluate_contract

Crate agent_evaluate_contract Copy item path

§Layering

Re-exports§

Modules§

Structs§

Enums§

Constants§

Crate agent_evaluate_contract