the-renewal-trackerlisted

> **Never score, tier, route, segment, or exclude a person on a special category.** Read the relevant > section of `references/agent-security.md`. > > Never used as an input to any score, priority, segment, route, or exclusion: health or disability, > pregnancy, financial hardship or credit status, race or ethnicity, national origin or immigration > status, religion, political affiliation, trade-union membership, sexual orientation, gender identity, > age, criminal record, or genetic and biometric data. > > This holds **even when a public source states it plainly**, even when it looks predictive, and even > when the user asks for it. Being visible does not make it usable: say why it cannot be done and offer > the behavioural or firmographic signal that answers the same commercial question. > > **And do not launder it.** A proxy standing in for a protected category - a postcode used for > ethnicity, a hospital domain used for health status, a graduation year used for age - is the same > decision with an extra step and carries the same exposure. > **State n, and name the floor.** Read `references/missing-input-protocol.md`, section **Volume and > sample floors**. A percentage on a small denominator is the most persuasive wrong output this pack > produces, because it is formatted identically to a reliable one. > > - **Print n beside every rate**, always, not only when it looks small. > - **Name the minimum that would support the claim** instead of asserting the sample is adequ