What is the optimization tag for? What should its description be?

Question

Currently our optimization description reads:

for challenges involving optimizing something.

Which is not very specific. Due to our rules on objective winning criteria every question (except maybe pop-cons) is about optimizing something. The tag wiki:

This indicates that the challenge involves optimization, which could be maximization, minimization, or something else. It might be performing some sequence of events in the most effective way, or it could be improving efficiency of retrieval or processing of input in a certain way.

is no more specific about what the tag should be used for.

The tag might be about optimizing the output in which case it would be a synonym of optimized-output.

The questions with these tags don't seem to have a common thread to me, which seems to be an indicator that this tag is not very useful. But perhaps with a more descriptive description questions that fit the tag would end up being tagged properly more frequently and questions that don't less so.

So what should the tag be used for? What should its description be?

Answer 1

Note: This is my best understanding of what an optimization puzzle is. It reflects my opinions that were formed based on my limited perspective.

I think optimization puzzles tend to focus on three areas:

1. There is exactly one observable parameter ("score") which measures the performance of attempting a puzzle in a particular approach. As examples, this could be a mixed function between code size, objectives reached (e.g. count from 1 to N where score depends on increasing N), constraints met (e.g. penalty on repeating characters), or any arbitrary but well defined set such as playing a game and measuring your score, etc.

2. The performance (measured score) involves the loosely computer sciencey definition of "optimization", that is a "better" approach works more efficiently or use fewer resources. (Resources is generally interpreted as code size on this site, but there may be alternatives.)

3. The problem does not have an obvious solution, i.e. solutions may compete on out-performing each other rather than solving the problem with the optimal methods. More specifically, in most optimization problems there exists a playing field that arises from having the output be a continuous function (i.e. for two similar approaches, they may be measured to be distinct from each other based on details) or the optimal algorithm is very hard to find / compute. (e.g. unit allocation)

Because optimizing code size already has its own field (code golf), and optimizing output size is generally considered golf over a puzzle (e.g. proof golf), optimization puzzles tend to target a different field, such as exploring a problem as deep / consistently as possible, or using less fuel to move a robot to a destination, sorting new structures which is graded by efficiency, or devising an algorithm to play a game and have your code try to reach as high of a score as possible.