Abstract: This Article presents evidence that some state prosecutors use their discretion to reduce racial disparities in criminal sentences. This finding challenges the prevailing view that prosecutors compound disparities. Given prosecutors' positions as mediators in a sequential system, this Article analyzes how prosecutors respond to disparities they inherit from the past--and interprets their impacts in light of the accumulated disparities that already exist when they first open their case files. Specifically, I estimate how the sentencing penalty for prior convictions differs by defendant race using North Carolina state court records from 2010 to 2019. I find that the increase in the likelihood of a prison sentence for an additional prior conviction was 25% higher for white than Black defendants with similar arrests and criminal records. While Black and white defendants without criminal records were incarcerated at similar rates, white defendants with records were incarcerated at significantly higher rates. And the longer the record, the greater the divergence. To understand this finding, I link an original survey of 203 prosecutors to their real-world cases. This survey-to-case linkage helps reveal how prosecutors' beliefs about past racial bias influence their decision-making. I find that the subset of prosecutors who attribute racial disparities in the criminal legal system to racial bias have lower prison rates for Black defendants with criminal records than facially similar white defendants, thereby offsetting past disparities. In concrete terms, racial disparities in North Carolina prison rates in 2019 would have increased by 20% had the state mandated equal treatment of defendants with similar case files. These findings should lead reformers to exercise caution when considering calls to limit or eliminate prosecutorial discretion. Blinding prosecutors to defendant race--a policy that jurisdictions are increasingly implementing--may inadvertently increase disparities by neutralizing the offsetting effects of some prosecutors. While race-blind charging ensures that prosecutors do not introduce new bias, it also ensures that any past bias is passed through to current (and future) decisions.