The second premise is very dependent on the algorithm and the hardware itself. "Faster" can be very different when parallelization comes into play. Baring that, ASIC's also a thing. Also, Moore's law has not been very relevant since around 2020? The new Moore's law is power efficiency and not transistor density. Say for example, we've peaked at 5GHz on a single core performance with x86 ISA. Then the next step is how to reduce the power consumption for the same 5GHz.
Now, the readme mentioned about using Argon2id. A cpu-memory-hard function. So you are also limited by the theoretical maximum memory that you could allocate to crack the puzzle faster.