This is not how I understood the phrasing. Layoffs can include a mix of randomization, removal of non-critical areas or groups, and quantitative low performers (reviews). The initial output of the layoffs may have skewed in one direction incorrectly representing marginalized communities or maybe not. Twilio says they would attempt remove a bias to baseline.