Harvest gambling question

Ignoring the cost of yellow lifeforce, has there been any empirical testing of the E.V. of the harvest div card gamble. I see a lot of people claiming that it is zero, but I haven't seen any empirical testing...

Fairly sure it is zero. I haven't tested my self, but someone wrote a good paper on strategies for getting a full stack. I'll see if I can find it.


Uh, this is the paper,

https://www.reddit.com/r/pathofexile/comments/p9lvhu/following_steelmages_confusion_on_stream_today_i/

It seems to be partially blocked behind a scribed account at the moment. I've read the whole thing, so it is/was available somewhere and it's very well written.

I believe the conclusion or gist of the paper was the EV on cards is 0, no matter what you do, but if you are going for a full stack you should gamble the amount of cards that will result in a full stack and not more.

I don't know about Scribed, but it is a very good paper. A lot of other people have done other analysis on this. I think I have another post about a year back suggesting the paper to someone in the gameplay section, it might have some more information if you can find it. I would bring it up for you, except the search function on the POE forum is not great and think it might take a while to find.


(Never mind, found that post, https://www.pathofexile.com/forum/view-thread/3512331 , basically same information I passed on to you and goes to the same file location, except the is a link I've got to someone else's work as well. Paper says it is written by NerdyJoe, maybe that person in available in community if you want to read the whole thing and don't wish to go through the site.)

Thanks for the detailed response, but I'm looking for any empirical evidence, like poor fishwife / sirgog etc. do. The reason is, I was sure someone had stated it was 0.02 back the league after harvest came out, but I can't find any evidence of any large test trials one way or the other.

The markov chain stuff, is actually stats nonsense, if the e.v. is zero there is no strategy to improve your long run odds, just improve your chance if you plan to do a limit trial with a defined stopping point. You can actually use negative binomial pdf for the stats for that instead of markov I think.

"

The reason is, I was sure someone had stated it was 0.02 back the league after harvest came out, but I can't find any evidence of any large test trials one way or the other.

Maybe, I haven't heard that though. I'm not a statistician but you would likely need millions of harvest mats to get anything approaching definitive.

I would say the most reliable data would be if the community accepts the outcomes to be of uniform distribute with 4.5 years of using the mechanic. It's possible that it is not completely even, but to get a more certain understanding than that it would probably need compilation of data from a large community, or beyond that, data mining the probability from the code or confirmation from the developers.

I've seen people doing this recently with trying to crack the values for shipping data. They've got a lot of useful information, but it's not definitive. Similar with combination probabilities.

I've not yet seen someone provide definitive probabilities for double corrupting skill gems, and that has been around for a long time.

Perhaps if there is a statistician around they can comment on how big a sample would be needed to be relative sure about the even distribution.

https://www.calculator.net/sample-size-calculator.html?type=1&cl=95&ci=0.5&pp=51&ps=&x=Calculate

~38k samples to find out with 95% confidence that the E.V is 1-2%

So not a solo task, but should be doable by a team.

And no, I wouldn't trust the community saying it is zero :) Stats is hard to judge for things you don't do often enough.

So about 60 million mats, which I think is about 10 mirrors in settlers currently without accounting for availability and impacting the market.

If you had 100 people compiling data, 600k mats each is not that hard I guess, but you also need the data to be recorded accurately which is another story.

I don't see why GGG would code in a slight bias weighting of 2% personally. This would be a lot of work to complete to guess something that they know just by looking at the code.

"

And no, I wouldn't trust the community saying it is zero :) Stats is hard to judge for things you don't do often enough.

I think someone would of noticed it by now and be using it to the point it was noticeable and known through out community. Maybe I'm wrong, maybe people know and have kept a lid on it.

If you find the post of the people claiming it's skewed, I'm interested to read it.

People keep claiming EV is zero with no evidence. The counter evidence is exactly as you said:
1) Yellow juice has value purely from the gamba alone
2) Lots of people are doing a "EV zero" activity
3) Card like house of mirrors have significantly higher value than their trade in value because they can be used in the gamba...

I'm a bit out of my depth on this one. I haven't used it much myself. For my purposes I assume it to be even, because I'm not doing the kind of bulk high end transactions you are talking about where it would matter. I'm largely playing SSF where I would be going for one stack.

It may well be as you say. I haven't seen compelling evidence either way. I tend to think it is even and not biased, but I wouldn't bet you on it without knowing the coding.