If a user crashed with an oplogTruncateAfterPoint set, deleted their oplog, went into initial sync, didn't apply any operations, and then cleared their initial sync flag, and then crashed, they could start up with a stale oplogTruncateAfterPoint and delete part of their oplog. If that oplog was used to commit a majority write (which it can be since initial sync nodes send progress updates), this could lead to majority writes being rolled back.