I wonder how Aurora fares on this?
If I was some database vendor that sometimes plays fast and loose (not saying Microsoft is, just an example) and my product is good for 99.95% of use cases and the remainder is exceedingly hard to fix, I'd probably be more likely to pay for Jepsen not to do an analysis, because hiring them would result in people being more likely to leave an otherwise sufficient product due to those faults being brought to light.
From what I understand, multi-az has some setup with multiple semi synchronous replicas where only 1 replica needs to acknowledge the transaction.
Aurora doesn't use semi synchronous replication but uses clustered/shared storage with a different replication setup for cache invalidation
So unlikely v17 will make a difference.