Documented losses persist at institutions where standard preservation policy is in place

The cohorts that produce the 73-93% non-delivery baseline (C-0002) are not policy-naive populations. The Vines 2014 ecology cohort (E-0001), the Wicherts 2006 APA psychology cohort (E-0003), the Gabelica 2022 biomedical cohort (E-0002), and the Acciai 2023 PNAS / Nature-portfolio cohort (E-0004) were all sampled from journals operating under data-sharing requirements; the Wicherts and Gabelica cohorts were further filtered to authors who had explicitly committed in writing to share. The losses documented across these populations occurred at institutions with data management policies in force, at universities that funded repositories, and on datasets that had been deposited in recognized platforms.

The implication is the rhetorical hinge between §1.2 and §1.3. The empirical record cannot be read as "loss occurs where mitigation is absent" — the standard tools of research data management were in place across the cohorts that produced the documented losses, and the data disappeared anyway. Whatever the loss-producing mechanism is, it is not one that the operational policy stack as currently constituted prevents.

This claim is the empirical motivation for the architectural thesis (C-0001). It does not by itself establish that the cause is architectural; it establishes that the cause is not addressed by the operational reforms already deployed. The architectural claim follows in §1.3 and is developed across the rest of the paper.