Making genes into gene products is subject to predictable errors, each with a phenotypic effect that depends on a normally cryptic sequence. The distribution of fitness effects of these cryptic sequences, like that of new mutations, is bimodal. For example, a cryptic sequence might be strongly deleterious if it causes protein misfolding, or it might have only a minor effect if it preserves the protein fold and tweaks function. Few sequences have effect sizes that fall in between. Strongly deleterious sequences can be subject to some selection even while they are cryptic, and expressed only at low levels that depend on a molecular error. Robustness to the potentially deleterious effects of cryptic sequences can be achieved globally by avoiding making errors (e.g., via proofreading machinery) or locally by ensuring that each cryptic sequence has a relatively benign effect. The local solution requires powerful selection acting on every cryptic site, and so evolves only in large populations. Small populations with less effective selection evolve global proofreading solutions. However, we also find that for a large range of realistic intermediate population sizes, the evolutionary dynamics are bistable and either solution may result. The local solution, which does not occur in very small populations, facilitates the co-option of cryptic sequences and therefore substantially increases evolvability. This can occur even in genetically uniform populations, illustrating that neighbourhood richness or "quality" in genotype space can be more important to evolvability than the quantity of genotypes in a population spread across that space.