Astronomy existed before telescopes, microbiology before microscopes, physics before particle accelerators. These are all just tools that increased measure-ability leading to a more accurate scientific approach. Things like sonar are an initial run at quantifying things like readability and rules compliance and they do a fantastic job.
You can predict with some degree of accuracy what a programs optimal efficiency would be, or did you miss that in your computer science classes? You are right that different languages/environments change the results, and while it adds a large layer of complexity it is still quantifiable. There is and will always be a "best possible solution." I work in an enterprise and not in research, so nothing I do will ever reach that point, but that doesn't prevent it from existing.
Lets take a quick look at the definition of science: systematic knowledge of the physical or material world gained through observation and experimentation. Everything I have said has revolved around this definition, observation and experimentation to gain understanding. Its as if you are intentionally being obtuse.
On the other side art: the quality, production, expression, or realm, according to aesthetic principles, of what is beautiful, appealing, or of more than ordinary significance. You don't do whats beautiful or appealing when coding. You do what is right. You don't right a recursive loop when you can do it in a normal loop, just because it looks prettier.
As for memory vs speed vs readability, you can weight requirements making them more or less important and there by altering the target optimal performance. Its simple math.
At this point you are either just a troll or are an incredibly incompetent coder. Do you honestly sit in dev meetings and when asked why you did something respond with "because its art man"?