You have to distinguish between validating the transaction (by attaching a new tip to it) and confirmed transactions.
There are still multiple confirmation models in IOTA research, and as far as I know it is not yet clear what exact consensus model (with which parameters) will be eventually used in IOTA 2.0. (IOTA 1.5 uses a simple coordinator-based approach where every transacion validated by a milestone transaction from the coordinator is automatically considered confirmed).
Why cannot you confirm a transaction as soon as there is a new tip? As every participant in the network can create new tips (at some rate limit), it would be too easy to create two conflicting transactions and then validate both of them by different tips, causing a double spend.
Common models include "all tips except one", or some majority rule.
But in the picture above, I assume it was quickly drawn for illustration, and does not follow any exact rules, as the bottom node in the last confirmed layer is validated by three tips and the second-to-top node in the first unconfirmed layer is validated by four tips (so either the former should be unconfirmed or the latter should be confirmed).