Scope 16 / Phase 1: Data model — BimodalStats, TailAsymmetryTag, new NumericFlag members, NumericStats fields, NumericProfileConfig thresholds

[DEVunderdog]

Parent

#105

What to build

Add all new schema elements introduced by Scope 16 to the Phase 1 data model. No computation is implemented here — this slice is the pure type / field / config foundation that every other Scope 16 slice builds on.

New types (_numeric_config.py)

• BimodalStats dataclass with fields: dip_statistic: float, dip_p_value: float, center1: float, center2: float
• TailAsymmetryTag StrEnum with members: Symmetric = "symmetric", RightHeavy = "right_heavy", LeftHeavy = "left_heavy"
• NumericFlag.Bimodal = "bimodal" and NumericFlag.HighOutlierDensity = "high_outlier_density" added to the existing StrEnum

NumericStats new fields

• bimodal_stats: Optional[BimodalStats] = None
• tail_asymmetry_tag: Optional[TailAsymmetryTag] = None
• tail_asymmetry_ratio: Optional[float] = None
• outlier_density: Optional[float] = None

to_dict() and from_dict() extended to round-trip all four new fields. BimodalStats serialises/deserialises as a nested dict (None when absent).

NumericProfileConfig five new threshold fields

• bimodal_dip_p_value_threshold: float = 0.05
• tail_asymmetry_right_threshold: float = 2.0
• tail_asymmetry_left_threshold: float = 0.5
• outlier_sigma_threshold: float = 3.0
• high_outlier_density_threshold: float = 0.05

to_dict() and from_dict() extended for all five fields.

Acceptance criteria

• BimodalStats dataclass exists with the four specified fields and is importable from dataforge_ml.profiling._numeric_config
• TailAsymmetryTag StrEnum exists with Symmetric, RightHeavy, LeftHeavy
• NumericFlag.Bimodal and NumericFlag.HighOutlierDensity are present in the NumericFlag StrEnum
• NumericStats carries the four new fields, all defaulting to None
• NumericStats.to_dict() serialises bimodal_stats as a nested dict when present and None otherwise; from_dict() reconstructs a BimodalStats from the nested dict
• NumericStats.to_dict() / from_dict() round-trips tail_asymmetry_tag, tail_asymmetry_ratio, and outlier_density correctly
• NumericProfileConfig exposes all five new threshold fields with the specified defaults
• NumericProfileConfig.to_dict() / from_dict() round-trips all five new fields; missing keys fall back to defaults
• Existing NumericStats and NumericProfileConfig round-trip tests pass without modification (no regressions)
• All in-scope symbols carry numpy-style docstrings per ADR-0034

Blocked by

None — can start immediately.