Alloy design is the deliberate choice of composition and processing route to obtain a specified set of properties, rather than the empirical adjustment of an existing grade. Thermodynamic calculation has long been the backbone of the work: assessed descriptions of the Gibbs energy of each phase let the equilibrium constitution, the transformation temperatures and the driving force for precipitation be computed before an alloy is melted, which is what the CALPHAD approach provides ^[1]. Such calculations have been used to set the chromium, niobium and carbon contents of heat-resistant cast steels where the target is creep strength near 1000 degrees Celsius ^[2]. Electronic-structure calculation adds quantities outside thermodynamics, among them stacking-fault energy and elastic constants, and is now routinely applied to screen refractory multi-principal-element alloys ^[3]. Entropy-based descriptors are used to rationalise which concentrated compositions form a single disordered solid solution instead of intermetallic compounds, although how far they predict is contested ^[4]. Machine-learning models trained on calculated and measured data are increasingly inserted between these steps; how much they shorten the search depends on how representative the training set is, and the transfer of a model fitted on one alloy family to another remains debated ^[5].