Factorio Forums

I am fully aware that there is a 0.7% chance of getting a U-235 out of the centrifuge. I have followed the wiki guide about the new nuclear power, and so built a setup with a single reactor and a single centrifuge. This, however, did not cut it for me. I've let the setup do its thing and went away to remove some nearby biter nests, and honestly, I've forgot about the setup.
I've returned about an hour later, only to notice that I have not connected an inserter that would feed the reactor with fuel, so when I saw that the reactor was not running, I just thought that it's because there is no inserter to transfer the fuel from the assembler to the reactor, however there was only a single fuel unit in the assembler's inventory, and no other U-235 anywhere, suggesting that out of a constantly running centrifuge, I've received only a single U-235 per hour, which, given the 10 second crafting time, equates to 1 U-235 per 3600 uranium ore, which is a chance of 0.03%. That is ridiculously low, and I highly doubt that it's just a case of bad luck. Or maybe I just suck at math, which is the more plausible option...
Anyways, this seems odd to me, as the wiki guide states that a single centrifuge should be enough to provide a supply of U-235 to a single reactor, which simply is not the case in my gameplay.
I've expanded the power plant by now, it now has two reactors and nine centrifuges, and a steady supply of trains feeding it uranium ore, and still it often happens that one of the reactors is left without fuel.
EDIT: as I am writing this, I've noticed that while running in the background, one of my centrifuges produced at least three units of U-235 in a relatively short time, which, once again, seems to be defying the odds.
Is it somehow possible that the RNG is broken? Or was there some sort of update that decreased the odds of successful enrichment?

If you played games for long, you learn to hate RNG. Because RNG is just that, a computer code that gives a chance.

You probably wouldn't like dice games either, like Settlers of Catan. It's amazing how often with that game, you bet on certain numbers and they never show up. Though it is way against the odds.

Not to derail, but RNG isn't reliable. So you plan on having extra.

Big game hint, don't use your U-235's until you have Kovarex processing going.

At a guerss the wiki is saying that a single centrifuge running kovarex enrichment is able to support one reactor. I'm pretty sure you need a lot more than one centrifuge to separate enough 235 to sustain one nuclear reactor.

SPolygon wrote:... as the wiki guide states that a single centrifuge should be enough to provide a supply of U-235 to a single reactor, ...

I don't know the correct ratio for that. But your reading that too literally. Imagine someone told you

Someone wrote:A single lottery prize can sustain a familily for 100 years.

Would you now believe that you only need one lottery ticket to live happily? No you wouldn't. The wiki states that on statistical average the ratio is x:y. But to get up to statistically average production numbers you need more centrifuges, or more time. I remember that my last 50ish centrifuge setup had pretty much the expected numbers when i checked on it after 10-20 hours or so. You're ofc free to test this yourself, just build 100 centrigues and let each of them run 1000 cycles, and then we can discuss if the RNG is broken or not. Btw you can see the number of cycles any assembler type machine has completed over it's lifetime in the tooltip, so you don't need to estimate by hour.

Edit:
I took a look at the numbers and this is what i see:
1 centrifuge crafts 1 U-235 in 1904 seconds (1000/7 * 10/0.75)
1 U-235 lasts 2000 seconds per reactor. (10 cells per crafting)
So you need an average of 19/20 = 0.95 centrifuges per reactor. Without enrichment, or any sort of module.

Where there is RNG, there are threads to complain about RNG. Protip: Random chance is random.

I built a single centrifuge

Solo

One

I think I find your problem!

You will never have enough U235 in the early stages of nuclear power. Once you research Kovarex you'll have way too much. There is no middle ground.

An anecdote : Back when I was playing MMOs, I got as a loot on my 52th kill an item that was supposed to drop at 1/8 chance.
Random does not mean regular, and even unlikely results may occur ... rarely.
Well, seems you only have bad karma

1 U-235 per 3600 uranium ore, which is a chance of 0.03%

Your arithmetic's right, 1/3600 is 0.03%, but your math's wrong. 3600 uranium ore is 360 processing cycles, so over 360 cycles you averaged 0.3% U-235 output, where the guarantee is if you do it forever you'll converge on 0.7%.

So just how unlucky did you get?

The math on this is weird enough that for any particular problem it looks maybe not worth figuring, especially not for something you're doing for fun, but it turns out that this math shows up a lot. Like, everywhere. It's worth investigating how it behaves, and in the lingo of the people who did the investigation you've got sample size 360 from a distribution with just two values, 99.3% of them "no" and 0.7% of them "yes". What are the odds of getting just 1 "yes" in that sample?

Turns out, they're pretty high.

But you don't have to do the full math to work out exactly what to expect, you can get a feel for it by running trials right in the game console. Do
then repeat until you get a sense of what to expect.

quyxkh wrote:The math on this is weird enough that for any particular problem it looks maybe not worth figuring, especially not for something you're doing for fun, but it turns out that this math shows up a lot. Like, everywhere. It's worth investigating how it behaves, and in the lingo of the people who did the investigation you've got sample size 360 from a distribution with just two values, 99.3% of them "no" and 0.7% of them "yes". What are the odds of getting just 1 "yes" in that sample?

About 20%. It's even an 8% chance you get none with that number of trials. The mean is 2.52, but the standard deviation is a whopping 1.58.

One of the things people forget when playing video games: you're doing a lot of trials. There was another game I was playing where I triggered two super-rare (bad) events right in a row. Before raging too hard about the evil RNG, I calculated how much I would have to play to expect to see that happen once: it was only about 200 hours.

SPolygon wrote:there was only a single fuel unit in the assembler's inventory

If you mean uranium fuel cell and there was no inserters on output, then I assume it was 10 cells

SPolygon wrote:Is it somehow possible that the RNG is broken?

It actually not. One centrifuge produces one U-235 every 1905 seconds (31.75 minutes) on average. So after an hour of processing you would expect to get 1.89 U-235 on average. And you get pretty the same result.

The easiest way to deal with random is to have another centrifuge.
More complicated, but a better way is to adjust your uranium fuel cell consumption to your demand using circuit network and then place more reactors to get neighbor bonus

SPolygon wrote:one of my centrifuges produced at least three units of U-235 in a relatively short time, which, once again, seems to be defying the odds

Again, it's not. It's just we as humans expected random to be predictable:
http://www.dailymail.co.uk/home/moslive ... stand.html

As a Bernoulli trial with p = 0.007, U-235 production can be modeled with both binomial distributions and geometric distributions depending on what your goal is.

Probability of getting less than 2 pieces of U-235 from 3600 ore (360 trials) is a little under 30 percent.
http://www.wolframalpha.com/input/?i=bi ... 3D+360+<+2

The chance of getting 3 U-235 from 1000 ore (100 trials) is a little over 3 percent.
http://www.wolframalpha.com/input/?i=bi ... 3D+100+>+2

In the geometric distribution, the PDF is the chance that trial N is the first to produce U-235, and the CDF is the probability in N trials that any has produced U-235, but that's equal to the negative binomial distribution where n = 1.
http://www.wolframalpha.com/input/?i=ne ... 07+n+%3D+1

For 3 instead of 1:
http://www.wolframalpha.com/input/?i=ne ... 07+n+%3D+3

don't forget that centrifuge has crafting speed of 0.75, so there are 270 trials, and mean is 1.89, not 2.52

Another note about probability

If you have good enough chance of winning and lot of trials like playing slot machine, you would have results pretty close to your expectations, that's why slot machines is the most lucrative thing for casino - it just constantly brings money without noticeable risks. But If if you have chance of winning as low as 0.007, than 270 trials is not enough to develop statistic close enough to mean