We study a model of machine teaching wherethe teacher mapping is constructed from a sizefunction on both concepts and examples. Themain question in machine teaching is the mini-mum number of examples needed for any concept,the so-called teaching dimension. A recent paper(Ferri et al., 2024) conjectured that the worst casefor this model, as a function of the size of the con-cept class, occurs when the consistency matrixcontains the binary representations of numbersfrom zero and up. In this paper we prove theirconjecture. The result can be seen as a generaliza-tion of a theorem resolving the edge isoperimetryproblem for hypercubes (Hart, 1976), and ourproof is based on a lemma of (Graham, 1970).