Variance-based truncation remains an effective way to derive successful search models from higher quality contact restraint-assisted ab initio decoys. The percentage of sequence in the search model is mapped against the root-mean-square deviation (r.m.s.d.) over all Cα atoms of the first representative of each search-model ensemble derived from the largest cluster against the native structure. Successful structure solutions of individual search models are highlighted in blue and unsuccessful solutions in red. Progressively darker shades of either colour correspond to increasing numbers of overlapping points. Progressive truncation is shown for (a) Rosetta decoys and (b) PconsC2+bbcontacts decoys (or PconsC2-only decoys for all-α targets).