how is this digital treatment similar to BetterHelp? how would it possibly be bad in factors other than efficacy, like BetterHelp was due to data nightmares and advertising a different mechanism? this isn't even online

99 is a more than enough sample size if your RCT's Bayes factor is 114 and 15.8 for better efficacy than -control and -regular treatment respectively, which corresponds to "extreme" and "strong evidence" (Lee and Wagenmakers 2013, p. 105; adjusted from Jeffreys, 1961). The Lancet also peer-reviewed the claim "The Bayesian adaptive trial design enabled efficient evaluation with early stopping when convincing evidence was reached (n=99).[2]"

indeed further testing is needed to establish subgroup effects and improve generalizability but this is already quite promising