Satan wrote:
It's a balancing thing. It's very difficult to define exactly what's right, but simple to tell when something is wrong.

I think that this either doesn't address what I wrote, or proves it.

Define "difficult", "define exactly", "right", "wrong".

But separately from that even, when you have a discrete number of configurations, even if parts of the problem are probabilistic, it's a trivial thing to solve for the optimal configuration using mathematics (linear programming).

*Integer* programming. Straightforward LP is usually completely useless when dealing with discrete configurations.

And this doesn't really apply here - we are analyzing [days(x), minutes(x)/mettle(x)]. Not linear problem at all.

(we can always find range of x that minimizes days(x), and then use that as another condition, but can't really get rid of the reciprocal).

Of course, this is relatively easy to solve by just enumerating all possibilities.

When you have continuous configurations, it's much more difficult to solve

Unless

1) you can simplify using differentials (or variations). Then you might even get analytic solutions, sometimes.

or

2) it's not linear

especially if parts of it are abstract enough to not be represented by mathematics well.

"We can't use math to solve this if we can't use math to solve this" is always true.

Important point incoming.

At that point you have to rely on true understanding of the mechanics and experience. When it's discrete, things can happen where a certain configuration is more likely to win out even when you don't have mastery of the mechanics or tons of experience. It's not always like that, but in this case it feels like it is. Yes, you can switch over to that configuration once you realize it, it just doesn't feel interesting. In the IotM category, once you solve the problem the first time, that same configuration is usually going to be the winning one (unless we get something like an IotM computer, or something else that really saves some time).

How do you know that this is wrong?