Algorithmic due process

The Right to Know and Appeal

A practical framework for notices, reason codes, preserved originals, human review, repair, and public accountability.

Overview

A thumbs-down button is not due process. When an automated system affects visibility, revenue, employment, education, credit, benefits, liberty, or access to essential services, the affected person needs more than a generic policy message.

A meaningful process identifies the action, rule, material inputs, degree of automation, consequences, and route to a reviewer who has authority to reverse the result and repair the record.

What is at risk

  • Generic notices that conceal the actual action
  • Appeals reviewed by the same automated system
  • Original content or data destroyed before review
  • A human reviewer who cannot change the outcome
  • Successful appeals that do not restore reach, revenue, or record state

Rights and safeguards

  • Durable notice and case ID
  • Machine-readable reason code
  • Preserved original and visible before/after state
  • Qualified human review for severe consequences
  • Written result and repair
  • Aggregate transparency reporting

What institutions and readers can do

  • Build appeal into the system before deployment
  • Measure reversals and repeated error classes
  • Give reviewers time, context, and real authority
  • Notify people when a corrected decision affected downstream systems

Related research

Starting sources

Limitations

This hub is a public-interest synthesis. Laws, technologies, deployments, and evidence can change. Consult the linked primary or authoritative source and the site’s evidence method before relying on a claim in a high-stakes setting.

Search the publication

Invisible moderationMental privacyCognitive Liberty Charter