one move: preserve an invariant
Train one scalar weight to match a target using the same gradient-descent loop. Return the loss path instead of hiding the learning process. In the Final Knowledge Intelligence System system, implement this level as a deterministic contract before composing it with the next service.
def train_neuron(weight, target, learning_rate, steps):
return [[step, weight, loss], ...]