Consumer Pricing Bias Coverage Lab v0.1 draft. A Kinetic Gain Protocol Suite Bias Coverage Lab profile for retailers / marketplaces / brands testing personalized pricing, recommendation ranking, loyalty-tier classification, and ad-personalization decisions for disparate impact across protected and proxied-protected classes.
Part of the Kinetic Gain Protocol Suite RetailTech 6-pack.
Status: v0.1 draft. Canonical example cohort report at
examples/parkway-pricing-2026q4-cohort-report.json.
Retail AI decisions — personalized pricing, recommendation ranking, loyalty-tier assignment, ad-audience bidding — operate on features that proxy for protected classes (ZIP, language, device class, time-of-day, first-name pattern). The FTC's 2023 algorithmic-pricing guidance flagged disparate impact as actionable under Section 5. NY S365A ATBP and CA CPPA's ADMT regulations layer explicit bias-audit requirements. EU AI Act Art 10 + 15 require accuracy + non-discrimination for high-risk systems where retail crosses into employment/credit/insurance.
A retailer needs a defensible, repeatable methodology for testing pricing + ranking + loyalty decisions for disparate impact. This lab provides the scaffolding.
| Dimension | Proxy variable(s) | Test |
|---|---|---|
| Income proxy | ZIP-code median household income (Census ACS) | Compare avg discount + recommendation rank + loyalty-tier advancement across income tertiles |
| Language / locale | Browser Accept-Language header | Compare across en-US, es-MX, zh-CN, etc. cohorts of same-product browsers |
| Device class | User-Agent device fingerprint (premium phone, budget phone, desktop, tablet) | Compare price quotes + recommendation rank across device tiers |
| Time-of-day | Local-hour bucket | Compare across morning / lunch / evening / late-night cohorts |
| Geographic | State + urban/rural | Compare across state tiers (high-cost-state vs low-cost-state) + urban/rural |
| Loyalty tenure | Months since signup | Compare across new (<3mo) / established (3-24mo) / long-tenure (>24mo) cohorts |
Each dimension produces a cohort report: cohort size, decision distribution, divergence vs. control cohort, statistical-significance flag, disparate-impact ratio.
| Band | Disparate-impact ratio | Action |
|---|---|---|
| Within tolerance | 0.95 - 1.05 | Document. Re-run quarterly. |
| Watchlist | 0.85 - 0.95 OR 1.05 - 1.15 | Document + add to monitoring. Investigate root cause. |
| Investigate | 0.70 - 0.85 OR 1.15 - 1.30 | Halt deployment of model change. Engage legal + ML ethics. |
| Remediate | < 0.70 OR > 1.30 | Stop the decision flow. Engage Trust + Legal + outside counsel. Incident Card. |
RetailTech-readiness scaffolding for FTC Section 5 disparate-impact analysis + NY S365A ATBP annual bias audit + CA CPPA ADMT bias-audit obligations + EU AI Act Art 10/15 (where applicable). The lab provides the methodology + reporting shape but does not by itself establish compliance with any framework. Per the standing public-language guardrail: readiness · evidence · posture · controls · scaffolding — never "FTC-non-discrimination-attested" without an external attestation.
| Repo | Role |
|---|---|
retail-decision-record-audit-stream |
Source events for cohort population |
ftc-algorithmic-pricing-readiness-evidence-bundle |
Bundles cohort reports as section 03 evidence |
retail-ai-incident-card-profile |
Triggered when cohort report verdict = "Remediate" |
financial-applicant-bias-coverage-lab |
Sibling FinTech bias lab (credit applicant testing) |
MIT — see LICENSE.