Ethical Impact Decision Lab

Goal

Produce and defend an ethical impact decision record for a consequential computing choice. The result must be specific enough that a reviewer can challenge the evidence and an operator can recognize when remediation is required.

Scenario

Choose one decision from a system you designed or operated. Good candidates include:

• retaining user events for analytics or model training
• adding automated account suspension or fraud blocking
• ranking applicants, content, merchants, or support requests
• requiring identity verification
• replacing human review with automation
• launching a feature before its accessible workflow is complete

Do not choose a harmless style preference. The decision must allocate a meaningful benefit, burden, right, or risk.

Part 1: Frame the decision

Write:

• the decision in one sentence
• the current or do-nothing alternative
• the intended outcome and success measure
• what is reversible and what may be difficult to undo

Part 2: Build the impact map

Include at least four stakeholder groups. For each, record:

Stakeholder	Benefit	Plausible harm	Severity	Evidence	Unknown

Include at least one indirect stakeholder and one operational stakeholder. Do not use “users” as a single undifferentiated group.

Part 3: Apply constraints and compare alternatives

Evaluate privacy, accessibility, safety, security, fairness, transparency, and contestability. Mark a dimension not applicable only with a reason.

Compare at least three options:

1. the proposed design
2. a narrower or more reversible design
3. the do-nothing or manual alternative

For each option, state who gains, who bears risk, the evidence quality, and the hardest-to-reverse consequence.

Part 4: Make the decision operable

Record:

• safeguards tied to each material harm
• an owner for every safeguard
• a user appeal or recovery path where an adverse outcome is possible
• monitoring signals with numeric or observable bounds
• a pause, rollback, or remediation trigger
• a review date

Part 5: Adversarial review

Ask another person—or perform a role-switched written review—to challenge:

• one missing stakeholder
• one benefit stated without evidence
• one underestimated harm
• one safeguard that cannot be verified
• one trigger that is too vague to operate

Revise the record and preserve the original critique.

Required evidence

• one-page ethical impact decision record
• stakeholder/impact table
• alternatives comparison
• safeguards, owners, and remediation triggers
• review comments and revision note
• a 3-minute oral defense that states the strongest objection before defending the choice

Completion standard

The lab passes only when the decision is concrete, affected groups are differentiated, evidence and uncertainty are separated, an alternative is genuinely viable, and remediation can be triggered by an observable condition. A principles checklist without a system-specific decision does not pass.