Random% Difficulty Dial

[pmcneill]

Thoughts about an idea I've been tossing around, this being one I'm hoping to work on a bit this spring and summer when not teaching.

Background:
The random% and MR flagsets are kind of one-size-fits-all -- you might get an awful seed or it might be a cakewalk. It makes the mode hard to practice and scarier than maybe it has to be.

What:
Start by giving every randomizer option a number of points, indicating its relative increase or decrease in difficulty. Rolling 2 lives might add 100 points, but 5 lives is -50 points. Starting candle? -10. No hints? +25. From here, we can come up with an overall difficulty score for a seed roll (visible in testing mode only, of course) and get a sense of whether it's reflecting reality.

The second step is to add a "difficulty" option, either a 1-11 dial or just an absolute number. After a seed's options are rolled, the randomizer looks at the rolled score compared to the difficulty requested. If it's too hard, it randomly picks a hard option to remove or an easing option to add, until it gets near the requested score. If it's too easy, the opposite happens. The changes it makes would be within the rest of the options selected, though. If a seed is too easy, it might turn hints off to up the difficulty -- but only if hints being on/off is set to random. If hints are forced on, it'll find something else to change, or decide it can't reach the requested level and just move on to generation.

Why:
This feels like a natural next step for a random% or max-rando style tournament, where the difficulty can increase as the tournament goes on. The score doesn't affect what can be chosen, so the flags are consistently the same throughout, but later round would see the randomizer favor harder choice combinations.

[Ellendar]

This is a consideration I'd had for a while. (This is probably a superset of #38). There are a few challenges associated with this.

1. Some properties are only difficult in combination. 2 lives start isn't that bad if you're starting with 8 hearts, but it's absurdly rough with only 1 heart. The amount more difficult 2 lives 1 heart is than either of those settings individually is literally much greater than the sum of its parts. While the difficulty calculation doesn't need to be linear, it does add complexity to building it.
2. There is a challenge finding the right place to put something like this. Right now, not all properties are randomized up front. The later in the process you put the decision about whether the difficulty for the computed settings exceeds the limit, the more properties you can include in it, but the greater the performance hit from repeatedly recalculating those values. This may involve moving some things depending how broad we want to make it.
3. In order for the setting to work with the flags, it needs to have a number of discrete values. Probably this is less of a dial and more of just a list, but in either case there is a cost to flag length bloat if we have to many options. Probably 8 is sufficient, but you get 2^n possibilities per n bits in the flag space we use at 6 bits per character.
4. We're going to need a way to test this. Ideally we implement this up front, and then have people rate their subjective experiences with the flagset and compare to help tune it.

Probably this gets connected to adding spoiler logs, which I need to make an issue for at some point.