The missing piece is that hiding bad news should be harder. For example, if you're a researcher and all you claim from your research are the good news, people (and the ideal Bayesian agent) should immediately suspect "maybe they're hiding the bad news".