QAEDA-TST-008 · feedback control

Measure efficacy and make rollback real

How will we know the intervention reduced the stated harm, what adverse effects will be measured, and what mechanism actually reverses or narrows the system if it fails?

Pass condition

What a defensible proposal should be able to show.

The proposal predeclares success and failure measures, evaluates privacy/liberty/error costs, publishes aggregate evidence where possible, and identifies who can narrow, suspend, or terminate the intervention.

Warning signs

Signals that the proposal is drifting.

  • More flags, more removals, or more surveillance are treated as proof of effectiveness.
  • No baseline or counterfactual exists for comparing outcomes.
  • Adverse effects such as false positives, chilling, exclusion, security risk, or bias are not tracked.
  • There is no practical rollback path once infrastructure and institutional dependencies form.

Evidence to demand

What should exist before confidence rises.

  • Pre/post or comparative outcome measures tied to the original harm.
  • Adverse-effect metrics and subgroup error analysis where relevant.
  • Independent evaluation or reproducible audit evidence.
  • A named authority, technical mechanism, and timetable for rollback or redesign.
Failure mode

A system can become permanent because it produces measurable enforcement activity even when its marginal contribution to safety remains unknown.

Decision discipline

No single passing test licenses the intervention.

Apply the full sequence. A defined harm does not excuse a weak nexus; a narrow mechanism does not excuse secret secondary use; an appeal process does not prove efficacy.