nirs4all.pipeline.dagml.native_conformal_calibration module

Identity-bound calibration replay inputs for the native Methods lane.

This module only compiles host data into the existing DAG-ML PREDICT and conformal-truth contracts. It never computes residuals, quantiles, fingerprints, or interval bounds: those remain owned by DAG-ML.

exception nirs4all.pipeline.dagml.native_conformal_calibration.NativeConformalCalibrationError[source]

Bases: RuntimeError

A raw calibration cohort cannot be represented by the native lane.

class nirs4all.pipeline.dagml.native_conformal_calibration.NativeConformalCalibrationReplay(execution: DagMLReplayExecution, binding_id: str, calibration_relations: dict[str, Any], truth: dict[str, list[Any]])[source]

Bases: object

One native PREDICT replay plus identity-keyed calibration truth.

truth is intentionally a JSON-shaped value for dag_ml.TrainingResult.attach_conformal_calibration. It has no provenance fields: DAG-ML derives and validates those from the source outcome, replay and relation set at attachment time.

binding_id: str

calibration_relations: dict[str, Any]

execution: DagMLReplayExecution

truth: dict[str, list[Any]]

nirs4all.pipeline.dagml.native_conformal_calibration.compile_methods_conformal_calibration_replay(package: Any, X: Any, y: Any, *, sample_ids: Sequence[Any] | None, groups: Sequence[Any] | None = None, metadata: Mapping[str, Sequence[Any]] | Sequence[Mapping[str, Any]] | None = None, methods_library_path: str | None = None, dagml_module: str = 'dag_ml') → NativeConformalCalibrationReplay[source]

Compile a finite, explicitly identified calibration cohort.

The replay itself remains PREDICT, so the Methods provider never receives targets as execution inputs. Its envelope nevertheless carries the measured-target fingerprint required by DAG-ML, and the same stable sample ids occur in the point prediction request and in truth.