Predicting if, how and when ice crystals will form  in clouds (and our own cells!) is important to atmospheric science (and cryopreservation! ) and the manufacture of pharmaceuticals . Preventing crystals from nucleating is desirable instead in our brains (where aggregates of proteins can lead to neurological diseases ) and for the oil industry (where water-based crystals can block pipelines ). Computer modelling can, in principle, design strategies for optimal control over these processes: in fact, advanced atomistic simulation techniques are now able to calculate key properties such as crystal nucleation rates. Unfortunately, the comparison of these rates against experimental data is subject to uncertainties  which have not been rigorously quantified yet. Specifically, the choice of the particular mathematical object we use to identify atoms as part of a crystalline nucleus (the so-called “order parameter”)  has a major impact. We will use formal uncertainty quantification (which form part of the core HETSYS training) to rectify this, initially using simple models  – for which accurate results are available, and then moving onto scenarios of great practical importance - such as the formation of ice .