Scope 16 / Phase 1: Compute tail asymmetry signal in NumericProfiler

[DEVunderdog]

Parent

#105

What to build

Compute tail_asymmetry_ratio and classify TailAsymmetryTag in NumericProfiler from the already-available PercentileSnapshot. No additional data pass over the column is required.

Computation (_numeric_profiler.py)

For each numeric column, after percentiles are computed:

tail_asymmetry_ratio = (p99 − p95) / (p5 − p1)

• When p5 == p1 (flat left tail — denominator is zero), store tail_asymmetry_ratio = None and leave tail_asymmetry_tag = None. Do not raise.
• Otherwise, classify using NumericProfileConfig thresholds:
  • ratio > tail_asymmetry_right_threshold → TailAsymmetryTag.RightHeavy
  • ratio < tail_asymmetry_left_threshold → TailAsymmetryTag.LeftHeavy
  • otherwise → TailAsymmetryTag.Symmetric
• Store both values on NumericStats.tail_asymmetry_ratio and NumericStats.tail_asymmetry_tag

Acceptance criteria

• A column with a disproportionately heavy right extreme tail (p99 − p95 >> p5 − p1) stores TailAsymmetryTag.RightHeavy
• A column with a disproportionately heavy left extreme tail stores TailAsymmetryTag.LeftHeavy
• A column with balanced extreme tails stores TailAsymmetryTag.Symmetric
• A column where p5 == p1 stores tail_asymmetry_ratio = None and tail_asymmetry_tag = None without raising
• tail_asymmetry_ratio is computed from PercentileSnapshot values already present — no second pass over the raw column data
• tail_asymmetry_right_threshold and tail_asymmetry_left_threshold are read from NumericProfileConfig (configurable, not hard-coded)
• All existing NumericProfiler tests pass (no regressions)

Blocked by

• Blocked by Scope 16 / Phase 1: Data model — BimodalStats, TailAsymmetryTag, new NumericFlag members, NumericStats fields, NumericProfileConfig thresholds #229 (Scope 16 Phase 1 data model — TailAsymmetryTag, NumericStats.tail_asymmetry_tag/ratio must exist)