Model development
Tool 01 · E.1 — Synthetic Augmentation Decision Tool
Check a gender approval-rate gap, diagnose its source, and determine whether synthetic augmentation is appropriate.
Participant library
PAL practitioner tools
Review the four tools before completing the questionnaire. Open each interactive tool and follow the instructions at each step.
Before you begin
Removing sex from a credit model does not remove gender disparity. Address, occupation, income stability, transaction history and an applicant's history with the lender can carry the same patterns. In many institutions, fairness work is split across data science, compliance, model risk and leadership, and the gaps appear in the handoffs between them. Each PAL tool covers one of those handoffs.
Your ratings test whether the tools are relevant, feasible, proportionate and defensible inside a real institution. The questionnaire does not test your knowledge.
How the four tools fit together
PAL works as a sequence, not four separate controls. When monitoring finds a problem, the finding starts the next step instead of ending in a report. Commitments are recorded before training, the data is checked and corrected during development, the model is monitored and its decisions explained after deployment, and every step leaves evidence.
PAL has two intended use cases for a lender. Building a fair model (Layers 1–2) uses Tool 04 and Tool 01 before the model is approved. Running a fair model (Layers 3–4) uses Tool 02, Tool 03 and Tool 04 once it is live. For the borrower, the result is one of two outcomes: an unfair decline caught before it reaches her, or a fair decline that comes with reasons she can act on.
PAL has two intended use cases for a lender — building a fair model (Tools 04 and 01) and running one (Tools 02, 03 and 04).
Back to Tool 04 · Decisions and audit trailTool 04 appears twice because it frames the whole lifecycle. Its template is signed before training, its decision points name who acts when a threshold is crossed, and its checklist collects evidence from every other tool.
After deployment
Tool 02 · E.2 — Fairness Monitoring Protocol
Check a live credit model's fairness, see what each result means, and who must act.
Each credit denial
Tool 03 · E.4 — Adverse-Action Explanation Reporting Template
Check that decline reasons are accurate, understandable, actionable and not stand-ins for sex.
Before training, then throughout
Tool 04 · E.3 — Pre-Registration Template & Audit-Trail Checklist
Fix fairness rules before training, name decision owners, and retain evidence throughout the lifecycle.
The final items, Q29–Q33, ask about the four tools together: completeness, proportionality, willingness to pilot, which tool you would adopt first, and what is missing.
Share your assessment
Keep the tools open while reviewing the questionnaire.
Review the questionnaire →