Seq@K is an evaluation protocol that measures whether a model can solve a problem within K sequential attempts when it receives feedback after each previous failure. For an operator, the change is to add Seq@K to the model-selection and feedback-loop design process rather than ranking models only by Pass@1 or Pass@K, because those metrics cannot capture how a model uses feedback on prior attempts.
