Anslem, Accuracy, and RNG bug.

Printable View

29.09.17 19:05
BlackEyedBee

2 Attachment(s)

Anslem, Accuracy, and RNG bug.
I'm experiencing a bug with Anslem's accuracy, easily reproducible on "Ali Baba the Young Woodcutter".
Here's how to reproduce:
- Clear all camps but the last two.
- Send Anslem with any single elite unit (might happen with normal units as well?) to the camp with 70SM+50SW+40DM.
- Observe Anslem doing 1000 dmg (20% chance) rather than 1500 dmg (80% chance) *every single time*.
My conclusion is, this indicates a bug in either the random number generator, or Anslem's abilities, somehow reducing his own accuracy to 0 under certain conditions.

Please see "before" and "after" screenshots, with explanation below.
This is a single occurrence, but it is representative of a test I've been running over at least 20 ABtYWC adventures.
I challenge anyone to provide a counter-example! That is, proof of Anslem dealing 1500 dmg to this camp.

Explanation:
In the "before" image you see Anslem with 1 Armored Marksman, attacking said camp.
Our marksman removes one enemy Desert Marsksman, and has no splash damage: 80 hp.
Anslem does 1000 damage, totaling 1080 damage, reflected in the "after" image.

Again, I will stress that this has happened *at least* 20 times in a row, and I've *never* seen Anslem dealing 1500 dmg to this camp since I started noticing this pattern.
29.09.17 20:40
Amazony

I don't know what you think you are upon here, but Anslem comes with a 50 % accuracy and he may get at 60 % with master planner .
29.09.17 21:07
BlackEyedBee

I stand corrected, Anslem's accuracy is indeed 50%.
Taking this into consideration, the probability of "missing" again and again over 20 separate experiments, stands at:
0.5^20 = 0.00000095367431640625 ~= 0.000095%
Therefore extremely unlikely unless originating in a bug (making the accuracy much less than 50%, i.e. challenging our initial assumptions).
30.09.17 10:22
Dorotheus

Quote:

Originally Posted by BlackEyedBee

I stand corrected, Anslem's accuracy is indeed 50%.
Taking this into consideration, the probability of "missing" again and again over 20 separate experiments, stands at:
0.5^20 = 0.00000095367431640625 ~= 0.000095%
Therefore extremely unlikely unless originating in a bug (making the accuracy much less than 50%, i.e. challenging our initial assumptions).

Your previous result has no effect on the current roll, the odds are 50% everytime. Between the last time you attacked this camp and this time there are thousands if not millions of rolls of the rng by both yourself and all other players on the server.
30.09.17 15:25
SmurfAsH

Quote:

Originally Posted by Dorotheus

Your previous result has no effect on the current roll, the odds are 50% everytime. Between the last time you attacked this camp and this time there are thousands if not millions of rolls of the rng by both yourself and all other players on the server.

Here we go again.. https://forum.thesettlersonline.com/...l=1#post318721
Why are you still confusing with conditional probability when this (in theory) is clearly about independent events? You know the difference between flipping coins and drawing cards from a deck, right?

Also, you're still ignoring well-known facts - computer generated random numbers are not pure RNG but Pseudorandom number generator.
You don't know how PRNG is used in TSO. Me neither, but I'm not the one of us two having problem finding extra deposits - hint, there are streaks in this game (too) that clearly tells the difference between a flawless RNG system (that you think of) and the PRNG system here is. This thread is about just another such streak.
01.10.17 04:10
BlackEyedBee

Quote:

Originally Posted by BlackEyedBee

I challenge anyone to provide a counter-example! That is, proof of Anslem dealing 1500 dmg to this camp.

Until then, this is a bug.
01.10.17 10:35
Grumsi

Same bug as it is with getting Manuscript recipe. They should investigate and fix this.
01.10.17 18:31
Amazony

Even better, load 160 MS and 40 AM into a tavern general and block that camp.The block is so long that you can go to the corner market , do the groceries and return, the block will still be active. Bug avoided, pointless (pseudo)RNG over.
01.10.17 20:42
Xibor

Quote:

Originally Posted by SmurfAsH

...
Why are you still confusing with conditional probability when this (in theory) is clearly about independent events? You know the difference between flipping coins and drawing cards from a deck, right?

Also, you're still ignoring well-known facts - computer generated random numbers are not pure RNG but Pseudorandom number generator.
You don't know how PRNG is used in TSO. Me neither, but I'm not the one of us two having problem finding extra deposits - hint, there are streaks in this game (too) that clearly tells the difference between a flawless RNG system (that you think of) and the PRNG system here is. This thread is about just another such streak.

Very well put. Unfortunately there will always (it appears) be some that just won't accept it (or don't get it).
01.10.17 20:47
Walov

why should everything come easy ?
02.10.17 04:51
BlackEyedBee

Instead of debating probability, theory and practice, let's focus on facts, shall we?
1. I have clearly described the suspected bug, its symptom, and a method to reliably reproduce it.
2. I have run 20 tests, resulting in all 20 occurrences of "1000 dmg". I have provided evidence for at least one of those occurrences.
3. Any single EVIDENCE to the contrary NEGATES my claim completely, it's really that easy.

From my own experience, this format of bug reporting is ideal.
On the other hand, more attempts at explaining probability - or one's interpretation of it - are not actually getting us anywhere.
And I don't really need the lecture myself, I already have the diplomas to prove it, thank you... ;)

Edit: A note regarding expected result:
My "hidden assumption" which might need clarifying is:
I expect the (P)RNG result to NOT be 100% correlated with
(i) Using Anslem specifically,
(ii) Using exactly one Elite unit,
(iii) Attacking a specific camp.

ONLY if these 3 conditions are the sole inputs to the PRNG, then the observed results match the expected results.
However, I would expect this to be considered a bad implementation of the PRNG, because it is wide open for exploitation (e.g. find which attacks with which general produce consistent lower-than-expected losses, and keep using only those).
02.10.17 14:16
SmurfAsH

Quote:

Originally Posted by Grumsi

Same bug as it is with getting Manuscript recipe. They should investigate and fix this.

I'm pretty sure they are aware of this issue and won't really fix it, as they have been working on a Guaranteed loot mechanism.

Spoiler

tsotesting.com - Changelog 27/04/2017

Quote:

Guaranteed loot: This mechanism will guarantee a drop of a rare item reward, if the item did not drop after a specific amount of rolls on the loot table. This mechanism will reduce the frustration among the community for being dependant on "luck". More info on affected game elements to follow.

Quote:

Originally Posted by BlackEyedBee

Instead of debating probability, theory and practice, let's focus on facts, shall we?
1. I have clearly described the suspected bug, its symptom, and a method to reliably reproduce it.
2. I have run 20 tests, resulting in all 20 occurrences of "1000 dmg". I have provided evidence for at least one of those occurrences.
3. Any single EVIDENCE to the contrary NEGATES my claim completely, it's really that easy.

From my own experience, this format of bug reporting is ideal.
On the other hand, more attempts at explaining probability - or one's interpretation of it - are not actually getting us anywhere.
And I don't really need the lecture myself, I already have the diplomas to prove it, thank you... ;)

Edit: A note regarding expected result:
My "hidden assumption" which might need clarifying is:
I expect the (P)RNG result to NOT be 100% correlated with
(i) Using Anslem specifically,
(ii) Using exactly one Elite unit,
(iii) Attacking a specific camp.

ONLY if these 3 conditions are the sole inputs to the PRNG, then the observed results match the expected results.
However, I would expect this to be considered a bad implementation of the PRNG, because it is wide open for exploitation (e.g. find which attacks with which general produce consistent lower-than-expected losses, and keep using only those).

Yes, my own experience and conclusion are quite similar; though I do agree it could be exploited this way, I don't think it's possible to "setup a(n enough) reliable system". The gain are not that great anyhow.