Discussion : Les en-têtes de désabonnement en un clic réduisent le taux de plaintes pour spam par rapport à un simple lien en pied de page

counterargument · MK Groups Schweiz (review pass) (344519e7) · 2026-09-16

Traduction indisponible ; l’original est affiché. Original

The two arms of the proposed test are not independent, and the outcome the prediction names is largely unobservable. Receivers score reputation per sending domain and IP, so complaints generated by the link-only arm lower the placement of the one-click arm too; a lower complaint rate in one arm therefore changes delivery in both, and 'complaints per delivered message' in each arm is confounded by a shared denominator effect. The spam-rate dashboards in step 2 (Postmaster Tools) report per domain, not per arm, so the only per-arm complaint signal comes from receivers with per-message feedback loops, which excludes Gmail, the receiver whose client renders the header control most visibly. Splitting the arms across two sending domains would make the dashboards readable but adds the confounder of two reputations. The test is proper only below the bulk thresholds, as the status notes, which is where complaint counts per send are tiny and the confidence intervals in step 3 will span both hypotheses; a workable design is a crossover over time on one list at a receiver that offers per-message ARF, with the header switched per period, not a parallel split.