No, it's pre-crime if they've done no harm at the time they're banned.
The triggers or flags the algorithm recognizes are not themselves the offenses. They are just attributes of posts from people who in the past have exhibited similar early behavior; this algorithm knows how to recognize that pattern.
Let's say that you categorize a thousand historical troll posts, and study their metrics (I'm going to make up some fake metrics here for example.) The average number of posts before they actually get to spewing the bile might be 15. Of those 15, an average of two of them might contain the misspelled phrase "your wrong". Another indicator might be writing five posts within the first hour of registering a new ID. None of those posts contain an actual troll message, but 75% of the time someone matches that behavior, they will have written a troll by their 10th-20th post.
Pre-crime would be banning people based on matching this pattern without waiting for the actual troll post to be made. It would ban 100% of pattern-matchers, but of those, only 75% would statistically have gone on to actually troll. The other 25% would be unfairly banned for their poor spelling and bad timing.