Datasheet — synthetic-lending

Generated by EthicLens. Format follows Gebru et al. (2021).

• Size: 8,000 rows
• Protected attributes: race, gender

Motivation

A controlled, labelled dataset with a known disparate impact, used as the bias-injection oracle that validates the EthicLens measurement engine.

Composition

8,000 synthetic loan applications with income, credit score and debt ratio features; the disadvantaged group has systematically weaker inputs (proxy discrimination). Labels are a fair function of the features (no direct race term).

Collection process

Generated deterministically under a fixed seed; see fairness_core.datasets.make_biased_lending_dataset.

Preprocessing / cleaning / labeling

Not specified.

Uses

Engine validation and demos only; never as a substitute for real-data audits.

Distribution

Not specified.

Maintenance

Not specified.