This is very rarely seen in dev-stage biotech. Often, they underestimate (conservatively) what they saw in a non-randomized P2 trial. This is very common in combo situations. This rarely works out.
The key word is "assume" in all these situations. We were just looking for "different or not" as the primary measure. Once we had that answer, we attempted to answer the +/- question. All answers were 100% dependent on the assumptions we used. While we could statistically verify the outcome of our assumptions, this didn't prove our assumptions to be true.
Smart bio investors make this mistake more than any other, IMO. they forget all that pretty biostatistical math is 100% dependent on the assumptions they plugged in. It is very, very rare when doing biostatistical models of trial data can get you an answer that is better than your model. That's not to say modeling can't inform your risk assessment. It can and often does.
It just rarely, if ever, provides you an answer whether the trial is positive or negative.