A large-scale study led by the Centre for Genomic Regulation and the Wellcome Sanger Institute overturns the notion that protein core mutations universally destabilize structure. Testing thousands of variants of the SH3 domain revealed considerable tolerance to core alterations, with only specific residues critical for stability. Machine learning models trained on this data broaden understanding of protein stability and enable accelerated, predictable protein design, offering significant potential for biotechnology and therapeutic development.