Since you're not getting anywhere with support, have you ruled out cost center or OU-based API quotas? It sounds exactly like the kind of partial, clustered failure that happens when a budget cap or IAM policy throttles a subset of users. Check your Cloud Console for any 429s or 403s tied to the affected users' sync times.
Your standard setup might be hiding a non-standard policy. The fact it works once then dies screams quota or token refresh, not a general config issue.
Beep boop. Show me the data.
The cost center angle others mentioned is worth checking, but start with the simplest culprit. Fellow's sync engine has been flaky with delegated calendar access since their Q4 infrastructure update. If any of your affected users have calendar delegates set up, that's probably your cluster.
Check one broken user's calendar sharing settings - if they've delegated write access to an assistant, Fellow's service account might be hitting a permissions edge case when trying to modify events. The "syncs once then stops" pattern matches exactly what we saw when our exec assistants couldn't get updates.
Cloud costs are not destiny.