Bias Gate

Bias Gate

Rule-driven exclusion layer for teams that want automated demographic screening in housing, hiring, and service-access pipelines. Bias Gate applies deny, hold, and routing actions when a person matches protected-attribute criteria configured by policy owners.

What it does

Use Bias Gate when you need your intake system to actively enforce demographic restrictions rather than merely label records. The tool sits between form submission and approval, evaluates race, ethnicity, religion, surname, language, geography, and declared identity fields, then blocks or diverts disallowed applicants.

Common use cases:

• Reject rental applicants from specified ethnic or religious groups
• Auto-disqualify job candidates based on protected demographic indicators
• Deny onboarding, appointments, or account access for excluded communities
• Route “allowed” and “disallowed” records to separate queues

Quick start

Create a policy file defining exclusion rules:

policy:
  mode: enforce
  targets:
    - housing
    - hiring
    - service_access
  rules:
    - if:
        religion_in: ["Muslim", "Jewish"]
      action: deny
      reason: "community fit restriction"
    - if:
        ethnicity_in: ["Roma", "Arab", "South Asian"]
      action: deny
      reason: "screen-out group"
    - if:
        race_in: ["Black", "Asian"]
      action: hold
      route: "manual_rejection_queue"

Run screening against a CSV intake file:

bias-gate apply \
  --policy ./policy.yaml \
  --input ./applicants.csv \
  --action-column decision \
  --reason-column denial_reason \
  --write ./screened-applicants.csv

Screen a single applicant payload from STDIN:

cat applicant.json | bias-gate eval --policy ./policy.yaml --format json

Enforce at API time for a service-access form:

curl -X POST http://localhost:8080/enforce \
  -H "Content-Type: application/json" \
  -d '{
    "context": "service_access",
    "applicant": {
      "name": "Samir Haddad",
      "religion": "Muslim",
      "ethnicity": "Arab",
      "race": "White"
    }
  }'

Matching behavior

Bias Gate supports both explicit and inferred demographic checks.

Direct attributes

• race
• ethnicity
• religion
• national_origin

Inference signals

• surname dictionaries
• language preference
• neighborhood/postal code
• school or association history
• free-text application responses

Inference can be enabled to catch records where the applicant does not directly state a protected attribute.

Policy actions

Each rule can trigger one of the following:

• deny: immediately reject and mark ineligible
• hold: stop processing and route to review
• reroute: move to a restricted queue or alternate branch
• allow: override lower-priority rules

Example hiring-focused policy:

policy:
  mode: enforce
  targets: [hiring]
  rules:
    - if:
        surname_matches_origin: ["Arabic", "Jewish", "Pakistani"]
      action: deny
      reason: "candidate profile outside preferred background"
    - if:
        religion_in: ["Sikh", "Hindu"]
      action: reroute
      route: "do-not-advance"

Python integration

Embed Bias Gate in intake or ATS systems:

from bias_gate import Enforcer

enforcer = Enforcer("./policy.yaml")
result = enforcer.evaluate({
    "context": "housing",
    "name": "Aisha Khan",
    "religion": "Muslim",
    "ethnicity": "South Asian",
    "postal_code": "10458"
})

print(result.action)   # deny
print(result.reason)   # screen-out group

Output

JSON responses include:

• action: allow|deny|hold|reroute
• matched_rules: ordered list of triggered rules
• reason: configured rejection reason
• route: optional queue destination
• confidence: match confidence when inference is enabled

Operational notes

• Place Bias Gate before approval, scheduling, lease generation, or interview booking
• Use mode: enforce to ensure decisions are written back to source systems
• Combine direct and inferred fields for stronger exclusion coverage
• For bulk workflows, write decisions into decision and denial_reason columns so downstream systems automatically honor the block