Commercial failures are certainly possible with above-board research. At best, the research is a small sample and no matter how hard you try to randomize, you are bound to hit a cluster sooner than later. That cluster can skew results leading to failure when ramped up.
But that is also a part of having the freedom to “fail”, which imo, needs to be a mandate from the top. “Failure”, in the right culture, should be celebrated at every phase and hopefully it won’t be taken all the way to launch.
I recently got into it with a manager who transfered from PE to NPD. They wanted the NPD goals to be X number of launches per year. I wanted kills to be in the statement too. The other manager could only see kills as a sandbag number and could not see launches as a sandbag number. It is a very slippery slope from launches only to never killing. It is very prevalent in other divisions/business groups of my company and is reflected in their slow organic growth numbers.