preprocessing.target_corruption.TargetCorruptionReport

preprocessing.target_corruption.TargetCorruptionReport(
    fired,
    n_flagged_cells,
    n_flagged_hours,
    spans,
    action,
    first_flagged_hour,
    pre_policy_last_target,
)

Immutable summary of a target-corruption detection and policy run.

Parameters

Name	Type	Description	Default
fired	bool	True when the detector identified at least one flagged hour-slot; False means no corruption was found (or the detector was not configured).	required
n_flagged_cells	int	Number of native-cadence (e.g. 15-min) slots that were flagged across all target columns.	required
n_flagged_hours	int	Number of distinct calendar hours that contain at least one flagged slot.	required
spans	List[Tuple[str, str]]	Contiguous flagged hour-runs as (start_iso, end_iso) string pairs. Empty list when the detector did not fire.	required
action	str	The policy action taken: one of "abort", "heal", "truncate", or "noop".	required
first_flagged_hour	Optional[pd.Timestamp]	The earliest flagged calendar hour, as a tz-aware pd.Timestamp floored to the hour, or None when the detector did not fire.	required
pre_policy_last_target	Optional[pd.Timestamp]	The last observed-target timestamp BEFORE any policy mutation. None when the detector did not fire. Used by the "truncate" path in prepare_data to compute how many hours were retracted so that predict_size can be bumped idempotently.	required

Examples

from spotforecast2_safe.preprocessing.target_corruption import (
    TargetCorruptionReport,
)

report = TargetCorruptionReport(
    fired=False,
    n_flagged_cells=0,
    n_flagged_hours=0,
    spans=[],
    action="noop",
    first_flagged_hour=None,
    pre_policy_last_target=None,
)
print(report.fired, report.action)
assert not report.fired
assert report.action == "noop"

False noop