There are two major families in cryptanalytic attacks on symmetric ciphers: statistical attacks and algebraic attacks. In this position paper we argue that algebraic cryptanalysis has not yet been developed properly due to the weakness of the theory which has substantial difficulty to prove most basic results on the number of linearly independent equations in algebraic attacks. Consequently most authors present a restricted range of attacks which are shown experimentally to work with their computer but refrain from claiming results which would work on a larger computer but have not yet been tested. For example in recent 2015 work of Raddum we discover that (experimentally) ElimLin attack breaks up to 16 rounds of Simon block cipher however it is hard to know what happens for 17 rounds. In this paper we argue that one CAN predict and model the behavior of such attacks and evaluate complexity of the attacks which we cannot yet execute. To the best of our knowledge this has never been do ne before.