Idea Sieve — Learning Loop

Learning from failure

A miss is not a dead end — it is a signal. Every blinded prediction that fails against held-out ground truth is classified into WHY it failed, and that classification is fed back into the sieve's priorities. Rules that keep failing blinded prediction get downweighted; rare over-performers get preserved. This is where the search strategy starts to evolve on its own.