We already had that discussion. The Monte Carlo model (from the white paper) assumes that all payers and payees employ the same algorithmic strategy for disambiguating double-spends. But there is no way to enforce that assumption without centralized servers.
That link contains an example with
completely random picking of tips.
Random picking is not modeling an attacker who is using a strategic algorithm.
Again instead of arguing about models, remove the centralization and let's observe.