I managed a support forum in the late nineties/early aughts. It was a support forum for a particular mental health condition, so we expected a lot of not-always-perfect behavior and tolerated a lot of it. But with some people it was always hostile and over the top (e.g. somewhat credible death threats) . We always outright banned first (after fair warnings). It’s easiest. That worked for some, who emailed, apologized, and ended up being great community members.

Others not so much. When they came back repeatedly with a new account after multiple bans with no change in behavior they got shadowbanned. The couple of users that figured it out and came back again got the nuclear option. Imperfectly and without guarantee of no collateral damage, it linked IPs, accounts, etc of the user. They could register a new account but the shadow ban followed them unless they were very careful; in which case we just applied it manually.

The thing is, you have to make the system easier for you to control than it is for them to exploit. Otherwise they’ll just grind you down until they “win”.