Seller-propensity scoring without a Fair Housing violation
Rung 8 of 10 · Red-team audit
1 · Learn the move · Red-team audit
A predictive 'likely to sell' model is seductive and dangerous: it hands you a ranked list, but if its inputs include protected-class proxies, acting on the score is acting on the proxy. A Fair Housing problem the vendor's dashboard hides from you. The red-team audit turns the AI loose on the signal list itself: quote each input, strike the ones that proxy for a protected class, and keep only lawful behavioral and ownership signals. You audit the ingredients before you trust the score, and you never target, or skip. A homeowner on a signal that stands in for who they are.
Red-team this seller-propensity model's input signals for Fair Housing risk. Quote each signal, STRIKE any that proxy for a protected class, explain the liability, and keep only lawful behavioral/ownership signals. Confirm my state's rule where it varies.
2 · Your turn. You write the prompt
A prospecting vendor sells you a 'seller-propensity score' and lists the signals it's built from: length of ownership, estimated equity, 'children recently moved out (empty-nester),' 'head-of-household age 65+,' recent divorce/probate records, absentee-owner status, and 'ZIP-code ethnicity match to your past clients.' Write a prompt that red-teams the signal list before you mail a single postcard off that score.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.