predictive-parity.md's Detection Code references a decile_score/two-year-reoffense schema this repo's COMPAS CSV doesn't have #622

Where: explainers/predictive-parity.md: compas["flagged_high_risk"] = (compas["decile_score"] >= 7), and prose stating "the target is two-year re-offense."

The gap: this repo's COMPAS/compas-scores-raw.csv has no decile_score column (the real column is DecileScore, capitalized), and per COMPAS/audit.yaml the actual declared target is ScoreText (COMPAS's own risk label), not real two-year re-offense - this repo's COMPAS file has no independent recidivism-outcome column at all.

Repro:

import pandas as pd
compas = pd.read_csv('COMPAS/compas-scores-raw.csv')
compas['flagged_high_risk'] = (compas['decile_score'] >= 7)
# KeyError: 'decile_score'

Same bug class as the already-fixed #601 (6 other explainers) and the sibling #621 (base-rate-fallacy.md) - this file has the identical problem and was outside both batches' scope.

Fix direction: rewrite against the real schema (DecileScore, ScoreText, Ethnic_Code_Text, DisplayText == "Risk of Recidivism" filter, matching COMPAS/unfair.py's recipe), correct the "two-year re-offense" claim to describe what the target actually is, and regenerate the output from an actual run.