model-drift.md cites a fabricated single-run gap/p-value for unfair.py (6.39%/p=0.348 instead of the real 7.16%/p=0.2564)

Where: explainers/model-drift.md - "The original unfair.py audit reports a single 6.39% gap from one random 80/20 split, not statistically significant at n=1,000 (p=0.348)."

The gap: the real, deterministic (random_state=42) output of German Credit Lending/unfair.py is a 7.16% gap with p=0.2564 - not 6.39%/p=0.348.

Repro:

$ python3 "German Credit Lending/unfair.py"
Fairness Gap: 7.16%
Permutation test p-value: 0.2564 (not statistically significant at α=0.05)

Ran 3 times - byte-identical output each time (the script is fully seeded). The correct 7.16% figure is also what's cited elsewhere in this same repo for this exact script (explainers/reject-inference.md, explainers/selection-bias.md, llms-full.txt) - model-drift.md is the only file citing 6.39%/p=0.348 for it.

Why it matters: the rest of the explainer's own table (5 sequential rolling-window rows: 4.3/11.7/13.6/15.1/10.2%) independently reproduces correctly and is internally consistent - only this one anchor sentence is wrong, giving a reader a false single-run number to compare the rolling-window swings against.

Suggested fix: replace "6.39%... p=0.348" with "7.16%... p=0.2564", matching the script's actual (and elsewhere-correctly-cited) output.