Okay, though experiment about the record company.
Suppose the company is organized into two departments. Department A evaluates the potential market for each project (artist and album) pitched to them by artists' agents. They take all factors into account including the artist's race, insofar as it affects the market estimates. They then produce a breakdown for each project: most likely range of cost, and most likely range of revenue, presented as 90% confidence intervals, over a whole range of marketing scenarios. Let's say they're pretty good at this, and historically, 90% of their estimates have in fact turned out to fall within the projected 90% confidence interval, regardless of the music style.
This information is then passed along to Department B, but only the financial projections. The names, backgrounds, music styles, appearances, and races of the artists, and even what the music sounds like, are withheld from the dossier. Department B makes the decisions on which projects to fund and which marketing scenarios to choose, taking into account the various offered risks and returns, and the company's current financial position.
Let's then say that the result is similar to the actual historical period in question; that is, black artists are less likely to have their projects funded because overall, though there are specific exceptions, the (accurate) market analysis of Department A predicted higher risk or lower revenues for black artists' projects, so the money decisions made in Department B less often went that way.
Which department contains the racists: A, B, both, neither?
Respectfully,
Myriad