Market testing only counts if you will act on the answer.
Agreeing in advance that the answer might be no is the cheap part; honouring it when it is costs something.
A product taken all the way through App Review, a paid subscription tier and real market testing — then retired when the usage did not justify running the backend. The full loop, closed on evidence rather than exhaustion.
The hardest part of taking something through launch and into market testing is not the building. It is agreeing in advance that the answer might be no, and then honouring it when it is.
Running AI at consumer scale makes this sharper than it used to be, because every active user carries a recurring inference cost. A product that would merely have been unprofitable in a previous era is now unprofitable per person, per month, visibly. Reading that number honestly and stopping is a product skill, and the discipline transfers to everything built afterwards.