solutions_backend.classification.manager

Durable, fail-closed classification orchestration manager.

Classes

AgentsExecutionClient

Bases: Protocol

Authenticated Agents workflow execution surface.

execute

execute(request: ClassificationWorkflowRequest) -> ClassificationWorkflowResult

Execute one side-effect-free classification workflow.

ClassificationCapacityError

Bases: RuntimeError

This Backend process has reached its explicit execution capacity.

ClassificationConflictError

Bases: RuntimeError

One stable intake identity was reused for different normalized content.

ClassificationExecution

Internal operation receipt distinguishing execution, replay, and active lease.

ClassificationManager

Own durable intake, Agents execution, and deterministic result admission.

classify

classify(intake: ClassificationIntake) -> ClassificationExecution

Persist, claim, execute, revalidate, and atomically admit one classification.

close_admission_and_drain

close_admission_and_drain(timeout_seconds: float) -> None

Stop new execution claims and wait boundedly for owned work to finish.

get

get(run_id: UUID) -> ClassificationRunState

Return one safe durable run state.

ClassificationNotFoundError

Bases: LookupError

The requested durable classification identity does not exist.

ClassificationTerminalError

Bases: RuntimeError

The durable run is terminally failed and cannot be retried implicitly.

ClassificationUnavailableError

Bases: RuntimeError

Classification could not safely reach an admitted terminal result.

IdentityRepository

Bases: Protocol

Post-Agent identity revalidation surface.

revalidate_identity

revalidate_identity(source, expectation)

Compare canonical identity to the pre-Agent expectation.

ObservationRepository

Bases: Protocol

Immutable observation persistence surface.

insert_or_validate

insert_or_validate(observation: SourceObservation) -> SourceObservation

Insert or validate one canonical observation.

RunRepository

Bases: Protocol

Durable run/attempt state-transition surface.

append_attempt

append_attempt(start: ClassificationAttemptStart, lease_owner: UUID) -> None

Append one running attempt.

claim

claim(**kwargs) -> ClassificationRunClaim

Claim or inspect one durable execution.

complete

complete(result: ClassificationWorkflowResult, finish: ClassificationAttemptFinish, lease_owner: UUID) -> ClassificationRunState

Atomically admit a successful result.

fail_retryable

fail_retryable(finish: ClassificationAttemptFinish, lease_owner: UUID) -> ClassificationRunState

Finish one attempt in reclaimable state.

fail_terminal

fail_terminal(finish: ClassificationAttemptFinish, lease_owner: UUID) -> ClassificationRunState

Finish one attempt in terminal failed state.

get

get(classification_id: UUID) -> ClassificationRunState

Return one durable run.

renew

renew(**kwargs) -> ClassificationRunState

Renew one currently owned and unexpired durable lease.