TargetSmart is garbage, how would they know how did early voters vote?
Most likely they are looking at the vote on county basis and modelling what the vote might be based on past election results in that county. That approach is not entirely unreasonable with states and counties that are very stable (eg NV or NC), however voting behavior of some large TX counties has changed so rapidly in such little time (eg Denton going from R+30 in 2012 to R+20 in 2016 to R+8 in 2018) that such analysis yields weird results.