From what I can tell from reading the exhibits in the court case, that's not what happened. This is a case of overfitting, rather than a conspiracy to create a process with "secret pass words".

They wanted to create a process that would not lead to disparate impacts, but would also give them competent controllers. So they took the biographical data that they had collected in the 80s and 90s and constructed a test such that the output of the test would give them the result that they wanted. This is clearly dumb, and they ended up with a model that was overfit to the old data and didn't generalize (or even make much logical sense). Likely there were a tiny number of black controllers when they gathered the biographical data, so those few data points, and their idiosyncrasies, had a large impact on the answer weights that they generated, making the model weird and pretty nonsensical.

It's not a conspiracy, it's just bad methodology in service of a dubious goal.

This is not just "bad methodology in service of dubious goal". It's blatantly illegal, and not just by weak "disparate impact" standard. According to disparate impact standard, procedures that result in disparate outcomes between protected classes are illegal, even if causing these outcomes was not the intent of the policy. Here, the disparate impact was not just unintended side effect, it was the deliberate goal. They deliberately constructed the test to discriminate based on race. This is even ignoring the alleged cases of insiders leaking answer key to some applicants.