It can be a very subtle issue, though. How would you recommend testing backups for rare corruption? Especially when live data is changing.

With this I had used a backup and nobody even noticed the problem for three weeks, thankfully I had some more direct backups too...

This is oft-repeated advice, but, in this context, it ends up being little more than a platitude.

Since the complaint here is of corruption that is silent, the level of testing required to catch it would be extraordinary.