Every soft fork or consensus change involves a (very small) non-zero risk of a network split. That risk is considerably lower for a soft fork than say a hard fork (where all nodes need to upgrade). That’s why soft forks aren’t attempted every month or year. All you can do is minimize that risk.
Aaron lays out some scenarios that are theoretically possible. Any incompatibility between “Bitcoin Core” and “Bitcoin Taproot” during the Speedy Trial deployment is in my view highly unlikely. If Speedy Trial fails to activate and we reach November 2022 (please note 2022 not 2021) without miners activating then we are in a similar scenario to the UASF in 2017 where it depends on what the economic majority is running. I can’t predict what the economic majority would be running in November 2022 but I highly suspect the delaying of Taproot activation would be at the top of everyone’s minds.
You do have to weigh up these risks of a network split with miners deliberately blocking Taproot activation potentially forever. If we were to say no more UASFs ever again because we don’t want to take any network split risk that would be handing miners a permanent veto to block the activations of soft forks that have community consensus. So you have to weigh up the risk of the latter which would be just as concerning (if not more concerning) to people.
So in summary these are subtle trade-offs. A number of developers have worked hard to minimize the risk of a network split. But it doesn’t get to zero unless you literally never try a soft fork again. And that would mean that Bitcoin would never seriously improve again.