FT% generally seems to be the biggest indicator for shooting ability. The 3FG% can overlook a lot of context, like open vs. guarded, off-the-bounce vs. off-the-catch, etc.
With Cody specifically, the data samples were small, leading to higher potential for variance in outcome.
That gets repeated a lot, but FT% isn't a great predictor either. The best predictor is a combination of 3pt%, FT%, and 3PA. Tankathon is basing their calculation on quite a bit of data I'm sure.