MetricCard

cards.MetricCard(self, id, version, name, description, metric_kind, measurand, task, requires_ground_truth, output, semantics, comparability, implementations, examples, provenance, aliases=(), citations=(), ground_truth=None, inputs=(), mappings=dict(), raw=dict())

A loaded and validated metric card.

The fields mirror the JSON Schema in schema/metric_card.schema.json. Nested structures (semantics, comparability, output, etc.) are kept as plain dicts, with typed accessors for the most-used fields.

Attributes

Name	Description
noise_floor	Smallest interpretable score difference, if the card declares one.
recommended_aggregation_across_datasets	Recommended way to aggregate this metric’s scores across datasets.
recommended_normalization	Recommended way to rescale this metric to [0, 1] before weighting.
score_of_random_baseline	Score a chance-level method reaches, if the card declares one.
target	Ideal value of a target_value metric, if the card declares one.