Error, consequence, repair
False Positives
False positives are the moment a weak signal, stale record, bad match, or overconfident inference becomes a real burden on a person who now has to prove the system wrong.
Publication information
Trace The Error Chain
Start with the trigger, source record, matching rule, model output, reviewer action, downstream sharing, and adverse consequence. Repair depends on knowing where the bad claim entered the system.
Name The Human Cost
False positives can cost time, money, travel, housing, employment, schooling, benefits, immigration status, protest access, reputation, and mental health. Error rates matter because people absorb the burden.
Require Notice Where Possible
People cannot correct secret files they never learn about. High-impact systems should provide notice where lawful, reasons, source access, human review, appeal, and a written outcome.
Audit For Patterned Mistakes
Ask whether errors cluster by language, disability, race, neighborhood, device sharing, poverty, protected activity, or data-source quality. A civil-liberties audit looks for who pays for mistakes.
Delete Or Quarantine Bad Records
Correction should include deletion, quarantine, annotation, downstream notification, appeal closure, and retention review. Leaving stale suspicion in place converts a mistake into a permanent shadow file.