while there is variance in how my models do against a test set, its highly unlikely my rank 10 model is going to dethrone a #1 rank model, nor a rank 1k model g

while there is variance in how my models do against a test set, its highly unlikely my rank 10 model is going to dethrone a #1 rank model, nor a rank 1k model going to beat mine. Its possible, but only because I over fit the public leader board, and good ML practices help prevent that (such as if your cross validation improves a model but it does worse on the public leader board, be inclined to trust your cross validation).