I do that with Cim berserker or Erii… but with rare spawns.

I understand why you’re doing that, but I’m having a hard time showing correlation between % chance to increase and number of points. This is of course compounded by the hidden 2nd roll, since it’s hard to factor that in unless you go 1 level at a time and record how many points they get, and ain’t nobody got time for that.

I was looking at all of the archers we did (specifically the RHuTS, who all seemed to come out with fairly low % on Accuracy) and went through half of them recording the % chance to level and the final Accuracy (without perks or armor) they had at 20. The results were all over the map:

Granted, a sample size of 19 is pretty small and since they only ranged from 40%-65% I guess one wouldn’t expect huge differences (25% chance with the 40% chance of a second roll should average out to 7 points), but you can see that my 45%/50% thralls ended up with values matching or exceeding the 65%.

My first Dalinsia is also quite like that.

+3 vit/+3 agi at 10

+3 vit at 15

+3 vit/+3 agi again at 20

OK…

Dang with a +9 vitality from perks alone I bet she has some crazy health.

Currently at the volcano. When I am back at the mainbase I can post a screenshot of her stats.

Not that high… and she didnt even got that much STR either… also a Cim with 90% rate in STR is kinda bad.

Compared to my Daicas:

Dalinsia with same chest+gloves (for str buff)

Do those attribute values include the effects of perks and/or armor they’re wearing?

No sir, they do not. I actually entered them on the sheet as “ = [Accuracy] - [+##] “. Nor do they include the extra 14% chance that all of them got from cooked fish.

Okay, so I actually went and dived back into the part of math I hated the most in high school: probability and statistics. Let me tell you something, this damn “hidden roll” is a major pain in the gluteal muscles.

But first, I wanna point out a slight discrepancy: you said your sample size is 19, but your graph has only 16 data points.

Right, moving on to the painful math.

It would be sooooo much nicer without the hidden roll. That would make each level a single “yes/no” experiment. That way the probability distribution for growing your attribute of choice over one or more levels would be a binomial distribution whose parameter *n* is the number of levels and the parameter *p* is the growth chance for the attribute.

The nice thing about the binomial distribution is that it has a simple cumulative distribution function *F(k; n, p)* which can tell you the probability of your thrall getting at least *+k* points to the attribute over the course of *n* levels for the growth chance of *p*.

That would be nice, but you don’t need to even think twice to realize things aren’t that simple: 5 of your thralls have more than 40 accuracy, and since RHuTS archers start with 20 accuracy, 40 would be the theoretical maximum over 20 levels, if it wasn’t for the hidden roll.

This is where stuff gets more complicated…

The hidden roll turns each level into an experiment that can have outcomes of +0, +1 or +2. The probability distribution for multiple independent experiments like that one is the multinomial distribution.

The multinomial distribution doesn’t have a nice cumulative distribution function like the binomial distribution, because the outcome of repeated experiments are specified as a vector that describes how many times each outcome occurred.

But our problem *can* be specified in terms of “at least *x*”, where *x* would be the total growth of the attribute after leveling, so there has to be a way to work with the multinomial distribution probability mass function to get the answer.

So, the first thing to do is to find the probability of getting a growth of “exactly *x*”. In order to do that, we have to find all the ways *x* can be expressed as a sum of +1s and +2s. Fortunately, that’s a variation of a well-known programming problem – making change given an unlimited supply of coins. With that, we can generate all the outcome vectors that would result in getting “exactly *x*” and the probability is simply the sum of all the probability mass function values for these vectors.

From there, calculating the probability of getting the growth of “at least *x*” is simply the sum of all “exactly *i*” where *i* goes from *x* to 40 (because the highest growth you can get is 20 times +2). If you have Python on your computer, you can calculate the probability using the script I’ll provide at the end of the post.

Of course, we don’t *know* the probability of the hidden roll, so that’s where we have to start speculating. Your number of samples is low in general, which would be problematic enough, but we need to look at each growth chance separately, which makes your dataset even worse: there’s only 1 data point for 55%, 60% and 65% each.

However, I did a bit of speculation anyway, just for *s*hits and giggles. Looking at the data points for the growth chance of 45%, I can see that one of them is +22, and half of them are at least +17. If I set the hidden roll to 80%, the probability of getting at least +17 is 46% and the probability of getting at least +22 is only 10%. If I set it to 60%, the probability of getting at least +17% is 29% and the probability of getting at least +22 falls to only 3%. I guess I could take +22 out as an outlier and play with the probabilities again, but I can’t be arsed

Anyway, all of this started because I wanted to point out that probabilities and statistics are really tricky, but I got actually intrigued and fell down the rabbit hole

If you wanna play with the numbers, here’s the link to the Python script: https://pastebin.com/8jaqbhDU

Not a discrepancy, just repeated data. I had values at 45% and 50% which appeared 2 or 3 times.

I went ahead and did the remainder of the RHuTS we have (34/40 Archer guards, the remainder are Hundred-Arrow) which isn’t much of an increase in sample size.

And the data, for your reference:

I do realize the issue caused by the second “secret” roll, but my point was not that the percentage is BS or anything along those lines but that due to the nature of the 2nd roll you can’t just throw out a thrall because he rolled a low chance in a stat. Trust me, I aced statistics in my MBA program.

I mean, yeah, you *can*

Obviously, you don’t want to do that with a Purge-only thrall it took you a lot of effort and luck to get in the first place. But a RHuTS or a Kisthis? It’s actually going to take you less time and effort to get one of them than to get it to level 20 (at least with PVE “0.25x” multiplier) and it’ll be more fun, too

I mean, yeah you *could*, but what I’m saying is that due to the nature of leveling, without perks even, it seems like many of my lower percentages are coming out the same or as low or high as those with better percentages.

As to the 0.25x XP in PVE (thank you for that, btw), you’re on your own there bud.

I’m at around 30 thralls this week that got the Break Bond. I am lucky. I am on a private PvE server with 4.0 to kill experience so I can get them to 15 fairly fast and see 2 perks as well as low primary. 25 of the 30 I dropped were test archers and world named fighters like Mei the Blade. I don’t like to chuck at 0 because I find even growth chance is locked on to a few stats for highest performance. Every so often one slips through though and you get more 2 hits than 1 or 0 on Attributes per level.

To be honest way to frustrated with Thralls to test anymore, hours of leveling to get crap over and over and over and all of you testing over and over again. Some of you spending hours doing charts and trying to figure out the hidden perks some thralls get and others do not and for what, no one dang word from Funcom on these broken thralls, when, or even if they are going to fix one things on them that is broken or we are stuck with the issues they have already. Open communication my butt. I will be leaving thralls in a box, if I was honest at moment thralls are just a time sink and a way to put up my blood pressure with no real end in sight as no one cares to give us any information of what will happen OR EVEN if a bit of a fix is coming.

We have had these issues for almost two months do any of you really thing they will do a thing now, no they are waiting for us to give up and just take the crappy thralls and deal with them. Far as I can see they ruined thralls and they do not give one piece of crap we are unhappy with them. Wish I was wrong but well , no more open communication now there there was before their , "we will be more open video. "

Well thats a lot of stinkun thinkun.

I love my thralls. Ive only got a couple I would really call “crap” and thats the giants and the yeti I broke to allow him to pursue his bliss. I take em all to level 10 if they finish the wheel, even at my tedious 1x xp rate on my server. Thrall leveling is the definition of endgame activity for me so I dont mind blowing hours on it. If I didnt, I’d just blow hours looking for a new base location… and breaking rocks. I dont care how tedious thrall leveling ever appears to anyone; breaking rocks is the worst. Its voluntary gulag duty…

So as of right now, food buffs are not working. I got a Dalinsia with 95% growth to Strength. 14% from gruel and 2 of 8 levels yielded no strength even though she has 109% growth rate.

Please submit that as a bug. It’s incredibly easy to write code that doesn’t properly check the bounds and then the results come out wrong.

Just saying, maybe once you get a whopping 10 top-tier thralls to level 20 and 4 or 5 of them turn out garbage, you might change your tune.

None of my thralls have gotten negative stats so far. I’ve only done 20 of them but every one has all positives. Sorry your luck is so bad.