Rykuta wrote: Sun Nov 17, 2024 8:18 pm
Drundia wrote: Fri Nov 15, 2024 11:55 pm
Rykuta wrote: Wed Nov 13, 2024 9:07 pmSo why did I choose the quantity of the parts rather than any sort of complexity weighting? To me it just seemed like the simplest thing to implement and the simplest thing for users to understand at a glance. It also seemed the most in-line with the current implementation as far as how many parts of each quality it would output. (But also my probability math is definitely rusty, so I could be wrong on that point).
Let's pick red science as an example. You need 1 gear assembler for 10 red science assemblers, but it would contribute the massive 50% to the output quality. Picking highest ROI ingredient for quality will be cheap. On the other hand bumping up quality in recycler is expensive. That's why ignoring quality of ingredients may be a good solution. After all you still need to do fine work to get fine results, fine ingredients are not enough.
I'll try to explain what I mean when I say that it seems most in-line with the current implementation.
In the current implementation, in order to get a 100% guaranteed uncommon with a 10% chance to roll into a rare (for our science example); you have to successfully roll two 10% chances (the plate and the gear). Alternatively, you just have to roll a 10% on the end science product. Note that rolling two 10% in a row is actually harder than just rolling the end product. But this is totally worth it right? Because now with the guaranteed uncommon you have a 10% chance to roll a rare. Well, the odds of sequentially rolling two 10% rolls successfully is 1% right? Which as it turns out is the same as the odds of rolling that blue without any of the other modules being uncommon.
Ultimately the goal is to just make it so that putting quality modules on intermediates is worth doing and more accessible than it currently is.
Aha, I see what you talk about. However, you are focusing a lot on chances of single output in a mass production game. If you roll an uncommon gear, but not copper plate, it's not like uncommon gear is lost. In the long run you'll get an average of 10% successful rolls on both and it doesn't matter that they don't happen at the same time.
Let's keep red science example and stick some 2.5% Quality modules everywhere.
For a baseline 100 units:
Code: Select all
common uncomm rare
Miner (+7.5%) 92.50 6.94 0.56
Furnace (+5%) 87.88 10.98 1.14
Assembler (+7.5%) 79.09 17.79 3.12
So, copper ends on Furnace and gear end on Assembler. With your suggestion we'll average 83.48 common science, 14.39 uncommon, 2.13 rare. A total of 118.65 science points for 300 raw resources or 0.3955 science points per 1 resource.
Let's also use recycler to bump quality of copper plates to the level of gears. Did some short calculations for a rough estimate. In my setup I need to mine extra 56.86 copper (for a total of 156.86) and send 75.81 copper plates into recycler, which loses 56.86 units and 18.95 units are remaining to be processed in recycler with 10% quality upgrade chance.
Now the original table needs to be scaled for new amount of raw copper:
Code: Select all
common uncomm rare
Miner (+7.5%) 145.10 10.88 0.88
Furnace (+5%) 137.84 17.23 1.79
Recycler loss -56.86
Recycler gain +17.06 +1.71 +0.19
Final 79.09 18.94 1.98
I did fast matching on number of commons, we would still need to recycle some more to convert several uncommons to rares, but let's pretend that out copper plates are as good as gears. So we'll get 79.09 common science, 17.79 uncommon, 3.12 rare. A total of 124.03 science points, but for 356.86 raw resources, which is 0.3476 science points per 1 resource. Your suggestion brings overall better result and eliminates usage of recycler on intermediates because it's simply cheaper to not.
The biggest problem is also illustrated in the calculations above: balancing more than 2 qualities. Currently besides storing items until better times the only option is a Recycler. However it's deeper in the tech tree and requires some complex factory design (5 subfactories for each quality and logistics to handle upgrades). I'd call for 3 options with their pros and cons.
1. The basic option. Basically the name of the topic. Higher quality items can go to lower quality recipe. Extra quality is wasted.
Pros: very simple factory (can use quality when necessary and ignore it completely where not needed)
Cons: higher quality is wasted.
2. Some king of upgrading machine, available earlier than recycler. Similar to recycler: loses 75% or so of the input and the output can be upgraded using modules.
Pros: relatively simple factory (mostly without loops, you only need to loop output of upgrader into itself if item wasn't upgrader)
Cons: fewer modules than Recycle+Recraft chain, can't benefit from innate prod bonuses of Electromagnetic plant or similar.
3. Recycle+Recraft chain.
Pros: best chanced to upgrade items thanks to two sets of modules past and possible innate prod bonuses.
Cons: very complex factory (need pretty much a factory for each quality and each product in the chain, especially in the long run, also need to set up logistics)
dragon_gawain wrote: Tue Nov 19, 2024 1:26 am
Hmm, mayhaps?
I haven’t done the math on recycling vs downbinning in terms of effective resource loss, but it’s certainly possible that downbinning is more costly than recycling…
I might try to crunch those numbers a bit to see what it comes out to.
And yes, of course with a large enough buffer the odds of being screwed by the odds becomes very small.
Edit: after reading through that testing thread, allow me to amend: a large enough buffer would lead to microscopic odds (i.e. balanced distribution of crafts are balanced) assuming that the odds are true, and that the random doesn’t have a bias. As they mention in the other thread though, those odds are being done to the limit of as time approaches infinity.
I’ll be curious to watch further results of that thread.
What is more costly depends on the final goal. For example in the above example with red science in this post, downbinning is more cost efficient than recycling. The number is 113.27 science points or 0.3776 science points per 1 resource.
However, if your ultimate goal is producing higher quality final product you either recycle the underdog (75.81 copper plates) or recycle the final product (75.81 red sciences). Obviously recycling final product is more costly.
Anyway, quality isn't free. You lose some machine speed and the opportunity to use other modules.
Probability theory says that for multinomial distribution (which is what happens with quality rolls) variance goes up as the number of trials goes up. Sometimes the outputs balance themselves over time, sometimes they instead grow apart, randomly.
(kinda)
