Testing Kingmaker's RNG

[Nightbringer]

I've had a feeling that Kingmaker's RNG was badly skewed for a while now. Today I decided to play around for a while and actually track d20 rolls.

The set up was simple. Octavia in the Lonely Barrow pulling the Lonely Warrior through a Web spell. Track the reflex saves and see what the numbers look like, pretty simple. Octavia with a +11 REF save vs a DC 19 was expected to save ~65% of the time. The Lonely Warrior with a +5 REF save was expected to save 35% of the time. Octavia had basic buffs only: Mage Armor, a hearty meal, physical enhancement - DEX, and Haste (a futile attempt to stay ahead of the Lonely Warrior, he runs the same speed as a hasted character).

The results were far, far worse than I expected. Not only were the numbers heavily skewed, anytime Octavia succeeded on a REF save, she immediately rerolled. The second roll was, with a single exception, significantly lower.

The Lonely Warrior played by a different set of rules. While following the same path as Octavia, he had much higher average rolls and never rerolled a successful save, not a single time.

A small sample of my numbers are here. Parenthesis are times Octavias saves were rerolled with the reroll result. Octavia has more numbers than the Lonely Warrior due to her failing many saves and getting caught without the Lonely Warrior having to make any saves at all.

Octavia: 7, 3, (12-2), (16-5), (20-2), (18-7), (8-2), 1, (14-10), 19, (8-19), 1, 6. 1

Lonely Warrior: 17, 14, 18, 12, 19, 15, 9, 16, 10, 14, 18

I actually ran numbers like for several hours, but the results were pretty consistent the entire time. In three hours I managed to entrap the Lonely Warrior for longer than 1 round while having Octavia make it through the Web spell exactly twice.

Octavia average die roll used: 6.07

Lonely Warrior average die roll: 14.72

The numbers have nothing to do with skill, character build, or difficulty setting. It is strictly an observation of the RNG, which is NOT random. The average should have been ~10.5, testing showed the actual results to be skewed by ~4 points in the opponents favor on all REF saves.

I may take a look at other rolls another time, but I doubt the results will be much different. Much of the "difficulty" of the game comes not from the settings shown to the player, but from a hidden game mechanic where die results are adjusted against the player.

[InEffect]

Originally posted by Eonwe:

Originally posted by InEffect:

Um. No. At least not in my experience. Only spells worth casting are duration spells as those force a lot of checks. Some are bound to fail. And for yourself... You go out of your way to become immune to everything you can for exactly the reason of not being a subject to RNG.

Aww c'mon that's bullsh|t, you can't raise all your DC high enough that enemies only make their save on a 20. Enemies with all saves at 16-17+ are fairly common, even in the mid game, you're not going to have 37DC spells by then.

Just an example : in the beginning of the game it's a pretty good tactic to have your caster cast daze on the rogues in the palace as their will save is low, but they will still make their save on a 15 or so on hard, probably a 13 on unfair.
If (and that's a big IF) the rng is skewed, it changes a lot things.

Also you're not going to hit on a 2+ with your iterative attacks, at least not on tough enemies.

that's why I say casting single save or bust spells is largely useless. Duration effects like stinking cloud is where the money are at or something that you don't exactly care about. Spamming fireballs accepting that 70% cases will result in 1/2 damage is still a 5d6+ AoE.

What I mean is all good spells chalk out RNG completely. Spells that don't allow saves, Spells that You don't really care if enemy would save and duration spells that will work sooner or later regardless if 1st save was a success.

That or ramping your DC sky high which can be the case with necromancy sage sorc and illusion arcane sorc gnome.

I generally approach this game from the position of RNG minimization. The less you care about any particular roll the better. If you don't care what you roll at all means you are doing things right.
Pretty much the same logic applies to build evaluation. If you have to load - you lost the game. That means you are doing things wrong. The less you load - the better the build is. Logical conclusion is 70AC or total control builds are the only viable ones.

[fourfourtwo79]

This is a must watch regarding RNGs in RPGs (25 mins onwards). It even rubs paid QA badly wrong. https://youtu.be/MEewLWDpscA?t=1500


There's another rather popular game that rubs people the wrongn way. It's FOotball Manager. It's demonstrated over and over that a manager who keeps on creating superior chances (superior to score, er hit chance, if you will) whilst keeping the enemies low, have a far more consistent experience than ones who confuse having loads of shots would equal "managing", or playing "good football". Actually, some of the AI tactics happily let opponents have far more shots, mostly in tight spaces where it's less likely to score (deep block defending in football). The first patch of players also won't just change everything based on a single patch of bad luck (losing to three drirect free kicks, against all odds). The other will scream he's getting rigged every time he drops a point. Somebody brought up the term "RNG minimzation". It could be argued that managing, in-game or otherwise, is about lessening the impact of chance, recognizing bad/good luck and not overreacting to either. So games like that are a fascinating experiment on the human psyche. Unfortunately, a game like FM roll any hugely transparent numbers. And most of its playerbase is damn awful at assessing how big a chance roughly is (from subjective watching, as well as longer term playing experience). To be fair, they are tought how "easy" it was to score in football in general every week by sensationalist television commentary too (which the game even mimics).

Not sure about Pathfinder. But it would be rather odd (and unnecessary coding work) if Owlcat had rigged the RNG in favor of AI. The higher difficulty levels already tilt the odds in the AI's favor by bumping their stats -- one some levels, massively so. This massively, that some players of the pen&paper argue that the higher levels were akin to having a masochist GM. Why a) add another layer of AI advantage when the one in the game is already massive, whilst b) risking the trust of your playerbase in doing so? Doesn't make much sense. In either case, statistically too small sample sizes would be akin to somebody proposing marriage to the girl he's just met. Could go well, but... why rely on chance?

[Eonwe]

Originally posted by fourfourtwo79:

It's akin to somebody proposing marriage to the girl he's just met. Could go well, but... why rely on chance?

Are you saying the guy should propose marriage to tens of thousands girls he just met so he doesn't rely on chance ? 0o

[fourfourtwo79]

Originally posted by Eonwe:

Originally posted by fourfourtwo79:

It's akin to somebody proposing marriage to the girl he's just met. Could go well, but... why rely on chance?

Are you saying the guy should propose marriage to tens of thousands girls he just met so he doesn't rely on chance ? 0o

I'm saying that he should rather propose a date first to see if he at all gets along. :-) Then a second, a third, a fourth... and after a while, moving together to see if that works out. Anything that may help him make a better decision and conclusion. It's a bit of an awkard analogy, I agree. Still anybody can appear like a good fit in any moment -- similar to how on occasion football clubs buy players based on their performance in recent World Cups (three/four matches in the summer VS their entire career) -- transfers that oft spectacularyl fail.

As the human brain is that bad with grasping probability (-> gamblers fallacy, monthy hall problem etc.), there are games that rig the RNG in favor of the human player, actually. As argued, the least any developer wants is for their playerbase to feel "rigged" or "cheated" or they'd just stop playing. Given that the game on higher difficulties already "cheats" in that it makes opposition stats sky-rocket, it'd be a pretty stupid thing to add another, hidden and player's trust breaking mechanic on top of it. But then who knows, and if there was something to it -- it could also be a bug. https://www.gamasutra.com/view/news/116071/Analysis_Game_AI__Our_Cheatin_Hearts.php

[sdf_momar]

Originally posted by SoundofSilence:

Originally posted by Eonwe:

More to the point of OP, here is a thread regarding how "randomness" is handled in P:K

https://www.reddit.com/r/Pathfinder_Kingmaker/comments/9r88d6/rng_is_set_a_bit_low/

From that link someone logged about 370 rolls:

https://m.imgur.com/a/wYB9KqS

This one is more or less evenly distributed. From the context of that thread it seemed like they just grabbed the rng code or whatever the term would be and ran it 370 times. It's possibly there's either and error in game or another factor that would lead to the two different results. What difficulty was this OP? I'm wondering if they add a skew to the rng on harder modes or something. Also, do you have any mods running?

It's also possible one of the two results are simply wrong. In which case I'd be inclined to believe the no significant skew version.

Outside of the RNG distribution questions, has anyone confirmed the double save behavior? That seems more easily provable.

[Veganleader1]

I've watched the odds of succeeding at something such as a trickery check drop enormously right after a quicksave. An example would be that I need a 10 or higher to succeed, I roll 1-4 about 9 times in a row. It sure feels like there is a built-in roll-nerf for when you save before a check like this. I've seen this countless times. It seems to succeed only after a good 6+ rolls, with all of the previous rolls being either identical or within a very small range, and at the bottom of the spectrum.

[=[NK]= Col. Jack O'Neil]

Originally posted by Veganleader1:

I've watched the odds of succeeding at something such as a trickery check drop enormously right after a quicksave. An example would be that I need a 10 or higher to succeed, I roll 1-4 about 9 times in a row. It sure feels like there is a built-in roll-nerf for when you save before a check like this. I've seen this countless times. It seems to succeed only after a good 6+ rolls, with all of the previous rolls being either identical or within a very small range, and at the bottom of the spectrum.

Yeah I've had similar, especially when I only needed a 5 or 6.. But that's the joy of rng..

[superlovi]

@inEffect ( since you mentioned it)
I know that there are some build around of maxed out gnome illusionist sorcerers that revolvev around weird ( and as a lvl 9 spell it comes pretty late) The sage necro build you are referring to is something similar or have some different tricks? Wail of the banshee do the job of weird or there is some spell that work good enough even some level before?

[InEffect]

Originally posted by superlovi:

@inEffect ( since you mentioned it)
I know that there are some build around of maxed out gnome illusionist sorcerers that revolvev around weird ( and as a lvl 9 spell it comes pretty late) The sage necro build you are referring to is something similar or have some different tricks? Wail of the banshee do the job of weird or there is some spell that work good enough even some level before?

necro has a lot of damage spells throughout the game, but late it has a downside of needing 2 casts to kill stuff. And it struggles against undead.

Necro build revolves around items a lot. You basically need 16 base con to cover most negatives of cloak of sold souls and profane int staff from varnhold to boost DC's. Add +2 from sage and you have DC in the 40s(42 at level 20 for lvl9 spells, to be exact). As a bonus, Horrid wilting attacks Fort.

I guess you can go 7/14/16/19/7/16 Emberkin Aasimar and take a monk dip for AC if you know how to get to 20 fast(er) and take crane wing(as that actually still works with 2h weapons) if you want some decent AC, but I wouldn't recommend it for most people.

[superlovi]

Thks for the detailed reply

[Mithrals]

Originally posted by Eonwe:

Originally posted by InEffect:

Um. No. At least not in my experience. Only spells worth casting are duration spells as those force a lot of checks. Some are bound to fail. And for yourself... You go out of your way to become immune to everything you can for exactly the reason of not being a subject to RNG.

Aww c'mon that's bullsh|t, you can't raise all your DC high enough that enemies only make their save on a 20. Enemies with all saves at 16-17+ are fairly common, even in the mid game, you're not going to have 37DC spells by then.

Just an example : in the beginning of the game it's a pretty good tactic to have your caster cast daze on the rogues in the palace as their will save is low, but they will still make their save on a 15 or so on hard, probably a 13 on unfair.
If (and that's a big IF) the rng is skewed, it changes a lot things.

Also you're not going to hit on a 2+ with your iterative attacks, at least not on tough enemies.

https://imgur.com/idTDAji

I dunno, I got pretty close and that is on unfair.

Back on topic tho, I would need to see way more data to convince me that there is a problem with rng. Companies don't even write their own rng function these days, they all use libraries. If there is a problem with those they should be fixed and updated pretty quick since they are used millions of times every day.

[Seizure Storm]

n = 200 is a more than large enough sample size especially for a dice roll test. That the rolls for PCs are this depressed is mind blowing.

[Eonwe]

Originally posted by Mithrals:

I dunno, I got pretty close and that is on unfair.

I was talking about the early - mid game, i know you can get illusion spells to ridiculously high DCs by end game.

But to get to that point, you need those web / grease / cloud / whatever control spell you use, to work at least semi reliably in the earlier levels.

[zero]

looks like small sample size bias, how many times was this attempted? you will hardly even begin to see an accurate reprensation of data until you do 100 attempts, and even then the sample will be too small to be perfect, it will be skewed slightly still.

[hammerinn]

Originally posted by Grifta:

Originally posted by SoundofSilence:

I would be wary unless I see 1000 or more data points

This, although I would add a few more 0s. You took a dice with 20 possibilities, and you didn't even roll 20 times.

I've personally rolled 3 20s in a row at a table. If I applied that to the sample size that you're using, that would mean that I'm either a world-class magician, or my dice have the same internals at one of those BB-8 toys.

Read. He did it for far longer.