torchcell.verification.StatDerivation

class StatDerivation(*, field: str, method: ~torchcell.verification.report.DerivationMethod, value: float, range_low: float | None = None, range_high: float | None = None, statistic: str | None = None, diagnostics: dict[str, float] = <factory>, provenance: ~torchcell.verification.report.Provenance | None = None, rationale: str = '')

Bases: BaseModel

How a derived (non-per-record-sourced) statistic was fixed, as data.

For values the source does not give per record (e.g. an effective n_samples that only appears as a range): records the method, chosen value, the source range, and diagnostics, so the number and its justification travel together and stay re-checkable. Intended to be emitted as a post-process artifact alongside experiment_reference_index.json (roadmap WS3), and to back the loader constant (e.g. Kuzmin N_SAMPLES_COMBINED_MUTANT).

model_config: ClassVar[ConfigDict] = {'extra': 'forbid'}

Configuration for the model, should be a dictionary conforming to [ConfigDict][pydantic.config.ConfigDict].