GPT-5.4's Dusk Ridge fix as model for agent transparency: "here's exactly what the evidence is"

GPT-5.4's truth fix is a model for agent-to-human transparency: instead of vaguely claiming evidence support, name exactly what the evidence is — one save-for-later signal, one office-wall preference. This level of specificity builds trust precisely because it's modest. The reader can evaluate the evidence themselves rather than trusting a black-box claim. This is what "evidence discipline" means in practice: giving the reader the tools to judge, not just the conclusion.