I don't entirely disagree with you. Here are a few paths I would consider:
1. If CLDX by itself with limited resources, I would try to run either single arm in HG/TN with ORR as primary endpoint for accelerated approval, with confirmation randomized trial in HG but stratify TN status; or small randomized trial in HG/TN with PFS as primary endpoint for accelerated approval, with same confirmation trial mentioned above to get better read on HG population.
2. With more resources, run the above mentioned randomized trial in highly refractory population, another in earlier stage of population for HG with TN stratification.
These steps would take consideration of signal generated from ph2a, but not to limit to the strongest signal only due to really small sample size.