Table S34 shows that over 20% of the control group either stayed home or didn't provide info on what they did. (Compared with ~4% of the intent-to-treat group)
So, sadly, they weren't able to directly compare 'public Montessori PK3' with 'public non-Montessori PK3'.
That lowers the statistical power of the experiment. It does not bias it. That's the beauty of intention-to-treat designs: they trade away bias for lower power -- a worthwhile trade every day of the week.
That doesn't change the fact that the title and article (which is all that most people will read) paints that as comparing public Montessori preschool vs public non-Montessori preschool.
But that's not what it's doing!
The baseline condition was NOT one where everyone had access to a public PK3 as a fallback if they didn't win the lottery.
So, sadly, they weren't able to directly compare 'public Montessori PK3' with 'public non-Montessori PK3'.