This article doesn't seem to account for the idea that removed and the announced have a different latency compared to a new announcement. For example if we withdraw a prefix and then announce it 10 minutes later propagation is WAY faster than if we announce a new prefix or withdraw for a few days and then announce again.

Edit, but cool article.

Why is it beneficial to quickly announce a 'test prefix'? Couldn't you position fast/slow propogation instead as trusting/untrusting providers?

It is beneficial for real changes to be propagated quickly, but it's irresponsible to use real traffic, so they used a test prefix -- a range of addresses without users.

Trusting / untrusting isn't likely to be the difference between propagating the route in 10 seconds vs 60 seconds. That's not enough time to do anything meaningful trust wise --- either the provider checks the routing dbs and routing PKI or they don't, but a survey of providers that accept or not poor routes is another story.