nirs4all.api.native_witness module
Live, process-local provenance for the strict Methods execution lane.
The public engine="native" path already has an intentionally separate
Dag-ML entry point: it supplies native Methods inputs and never installs a
Python operator callback. This module does not turn that local fact into a
portable attestation. It retains the exact live dag_ml.TrainingResult
facade only long enough to make the local execution boundary observable and to
release its process-local resources deterministically.
The durable evidence remains the signed outcome and portable package owned by
Dag-ML. The internal live witness relies only on Dag-ML’s public Python
lifecycle surface, is deliberately neither serializable nor copyable, and is
invalid after detach()/close().
- class nirs4all.api.native_witness.NativeMethodsExecutionClaim(schema_version: int, execution_entrypoint: str, execution_mode: str, outcome_fingerprint: str, methods_library_mode: str, portable_artifacts_required: bool)[source]
Bases:
objectAudit-only description of one locally witnessed strict execution.
This is intentionally a small immutable projection. It is not a signed receipt and cannot be used to establish provenance after the live witness has been detached or a result has crossed a process boundary.
- class nirs4all.api.native_witness.NativeMethodsTerminalExecutionClaim(schema_version: int, execution_entrypoint: str, execution_mode: str, outcome_fingerprint: str, terminal_run_id: str, receipt_fingerprint: str, methods_library_mode: str, portable_artifacts_required: bool)[source]
Bases:
objectAudit-only description of one attached strict terminal execution.
The native receipt remains the authoritative evidence. This projection only identifies the live local boundary that still owns the attached
TrainingResultused for lifecycle management.