Critical systems such as those developed in the aerospace, railway or automotive industries need official documents to certify their safety via convincing arguments. However, informal tools used in certification documents seldom cover the uncertainty that pervades safety cases. Several works use quantitative approaches based on belief functions to model and propagate confidence/uncertainty in the argument structures (particularly those using goal structuring notation). However the numerical uncertainty information is often a naive encoding of qualitative expert inputs. In this paper, we outline a qualitative substitute to Dempster-Shafer theory and suggest new qualitative confidence propagation models. We also propose a more faithful encoding of expert inputs.