You can't be certain about that until you actually send a complete email. Quite a lot of SMTP servers are nowadays configured to always say OK at that stage even if the address isn't valid to prevent using that as a mechanism for scanning for valid email addresses to build spam address lists from.
A lot of SMTP servers say OK to anything during the RCPT TO: phase because they aren't in a position to validate the user name/mailbox portion anyway, because they are not the system which will do local delivery.
I must admit I have the system which does local delivery for my domains set up to reject invalid user/mailbox names during "RCPT TO:". I found that I was getting tons of email to random user names but not in a pattern which suggests attempts to scan my user name space - in fact I have never seen anything which looks like an attempt to do that. Some are totally random but with repeated attempts over and over with the same user name. Some look like someone has merged a dictionary with a list of hosts, some look like someone has taken a spam email address list and tried every permutation of user name and host name. At least this one sort of makes sense as an individual might have the same user/mailbox name with multiple servers.
By rejecting undeliverable email as soon as possible I avoid the cost of running it through the spam and virus checkers and I leave the system sending it to me with the problem of trying to bounce to the sender (which is probably forged anyway). So I waste less CPU and generate less backscatter.
Works for me.